<h1>Introduction</h1>Groq is on a mission to build the computer for the next generation of high-performance machine learning. Founded by former Google engineer, Johnathan Ross the company has doubled in size since its inception and is on track to double its headcount again by the end of 2021. In late September, Ross presented as part of the online AI summit 2020. His keynote gave key insights on the industry, the market, and the Groq architecture capable of producing the world’s first PetaOp chip. We’ve pulled out some of the highlights from the presentation.<h1>An answer to the end of Moore’s Law</h1>In 1971, Intel co-founder Gordon Moore predicted that the number of transistors we can pack into a microchip would double every 18-24 months. A statement that has proved itself true for a time, but as Moore himself noted “the nature of exponentials is that you push them out and eventually disaster happens”. Computing now finds itself on a precipice where<a href="https://www.wevolver.com/article/what.a.little.more.computing.power.can.do" rel="noopener noreferrer" target="_blank">&nbsp;cost, power, and efficiency&nbsp;</a>compete for priority to deliver the computing ambitions of the next generation.Rock’s Law or Moore’s second law states that the cost of a semiconductor chip fabrication plant doubles every four years. In line with this prediction, next-generation factories are estimated to cost more than $20 billion.Realizing the enormity of what these two laws mean for the future of computing placed Jonathon Ross, <a href="https://groq.com/" rel="noopener noreferrer" target="_blank">CEO, and Founder of Groq</a>, on his current path.Ross recounts when working for Google back in 2012, Jeff Dean gave a presentation to the leadership team, containing just two slides. The first slide was good news; machine learning finally works. The second slide was bad news; It would require another 20 to 40 data centers to be able to afford it - and that was just for speech recognition. At the end of the presentation, Ross started the TPU (Tensor Processing Unit) at Google.Fast forward several years and Ross, now the head of Groq, unveils the world’s first and only PetaOp chip. The chip’s performance is equivalent to one quadrillion operations per second, or 1e15 ops/s. Groq's architecture is also capable of up to 250 trillion floating-point operations per second (FLOPS).<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1603820876791-Groq+banner.jpg" style="width: 300px;" class="fr-fic fr-dib"><h2>Architecture over speed</h2>“The industry often focuses on faster speeds,” begins Ross. “That’s what so many chip makers are focused on today. Everyone in this field is trying to make a faster chip. That approach of throwing more transistors onto a die means cost increases with incremental performance improvements. Following a trend that is running out of steam isn't a win for customers who are planning for the future. What we've done is come up with something revolutionary. I call this deterministic or predictable computing. A radically simplified approach which is scalable for customers. This is what enabled us to create the first and only PetaOp processor.”<h2>The shortest path to performance</h2>“How did we get there?” Ross reflects at the start of his keynote address. &nbsp;“We got there by doing things differently. A good example of this, we built on a 14-nanometer process when others were building on a 7-nanometer process. We simply didn't need the bleeding edge lithography to produce the fastest chip. Instead of adding more transistors, our performance came from our architecture. Everyone here believes Moore's law will run out. We talk about it and no one is acting any differently than they did before.”Ross asserts that the real question is not the end of Moore’s law but what happens if ML accelerator hardware doesn't get any better. The answer, Ross believes, is in designing and delivering a new architecture not based on the familiar systems of FPGA, CPU, or GPU. “It’s something completely different,” says Ross, “we call it the TSP or the tensor streaming processor.”<h2>Compute as power</h2>“Compute is a utility like power,” states Ross. “The country that controls compute will be in an advantaged position.” The future of computing is likely to need a careful mix of private enterprise, government, and academic input, research, and funding to ensure both its <a href="https://www.wevolver.com/article/a-global-collaboration-to-move-artificial-intelligence-principles-to-practice" rel="noopener noreferrer" target="_blank">continued development and shared access</a>. It isn’t only owning or controlling the technology that will give a nation an advantage but rather the overall deliberate direction and ambition around legislation, use, access, and dominance.“The future isn't guaranteed,” says Ross, “but you do get the future that you stand up for.” For Ross being deliberate about your approach is key to success. But he also understands this needs to happen across the industry to be effective for as many as possible.&nbsp;“At Groq we believe in the shortest path to performance,” continues Ross. “We have to stand up for the future we want by creating new trends with our approach. Trends like:<ul><li>Inference being different than training and being bigger than training.</li><li>Batch Size 1</li><li>Determinism and predictability, and</li><li>Performance and software first, which to us means the complexity and intelligence of a task is managed in the software stack.</li></ul>“Defining these trends originated with what we’re accomplishing at Groq but more importantly, these are going to enable customers' AI applications in production, that today are simply too costly.”<h1>Hopes and Challenges for the Future of AI</h1>Key investors and stakeholders in the AI field joined Ross in the presentation to share their hopes and challenges for the future. &nbsp;Each speaker pointed to where their priorities are for the development of the technology and gave insights into where their organization is directing resources. Speakers included Nicolas Savage, the Managing Director of TDK ventures the, corporate VC arm of TDK, CEO of Canadian quantum computing company, &nbsp;1QBit, &nbsp;Andrew Fursman, &nbsp;Teddy Gleeser, &nbsp;a partner at D1 Capita and CTO of the ASIC business unit at Marvell, Igor Arsovski. Their full remarks can be viewed in the video excerpts below:<h2>Predictability and reliability</h2>“I’m personally excited about future advancements like bioengineering, finance, government applications, and especially autonomous vehicles because it will save lives,” shared Ross. &nbsp;“My cousin was a passenger in a car that got into an accident. The driver of the other car had been drinking and he slammed into the side of my cousin's car. This didn't have to happen and if we had autonomous cars it wouldn't have happened. I want our chip in those cars because we need the level of predictability that I know our chip can bring but there's something else that AI brings you. When cars can make deliveries to you when auto-editing makes you a better photographer, when auto-sorting your inbox or when speech recognition keeps an EMT's hands-free so they can focus on their patient in an emergency.”“One major challenge in today's AI's systems is when you look at AI accelerators which typically push both performance but at the same time try to drop the voltage as low as possible so they can achieve really high energy efficiency. This tops per watt metric is kind of critical. Furthermore, these accelerators typically get integrated into data centers and automotive applications that have critical uptime and reliability requirements. I know reliability is not as sexy as the tops per watt metric and because of that it's taken a backseat to many AI discussions and many AI conferences but reliability is really critical to deliver useful systems.”<h2>Time is a Limited Resource</h2>“Silicon Valley has gotten a lot of criticism for focusing on convenience but I’m not going to apologize for that whether it's a safety issue or convenience time is a limited resource and we only get so much of it. What are the obstacles keeping us from that new reality? You’re probably thinking about speed but that's only part of the equation. Performance is a big challenge but reliability, predictability, and certainty in the real world is key. Chip makers like us need to deliver that performance not just in the lab but also in the real world,” details Ross.“Training cost isn't an issue. Train the model once and use it as much as you want. With inference, each response has to be real-time and it costs money and energy that's a scaling issue. Now instead of scaling with the number of ML researchers, you're scaling with the number of users and the number of queries. Inference at scale is harder than training.”“When I was at google roughly 80% of the ML models that were trained ended up having to be abandoned because they were too expensive to run in production. AI success is hinged on speed for inference as well as training but for the workloads of the future total cost of ownership is dependent on scalability. Just like when the team and I created the TPU at Google and it saved billions of dollars, we as an industry have just begun tackling these challenges.”<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1603874293788-groq+in+body.jpg" style="width: 788px;" class="fr-fic fr-dib"><h1>The future isn't guaranteed</h1>One of the guiding mantras at Groq is that the ‘future isn’t guaranteed’. This is essentially a guiding principle that the future is both unknown yet directable. The intention within the development of technologies such as AI and computing must be deliberate and understood within a wider socio-economic context. Ross takes the responsibility that he and his company, and others like it have the power to change the future, very seriously. It's a way of thinking that is embedded in the company's methods and products.“At Google, the team and I created the TPU for existing workloads that we could make better as well as new and emerging workloads,” says Ross. &nbsp;“To plan for the future means understanding those workloads. Those were yesterday's challenges and we have new challenges today like power efficiency. To Groq those obstacles are the path and that's where we're going.”“Today, we're shipping to our customers both as individual PCIe cards and systems with eight cards each and there's even more on the roadmap to come. I know 2020 has been a crazy year. Together we're pushing through and making great strides. Groq is going to grow our <a href="https://groq.com/careers/" rel="noopener noreferrer" target="_blank">headcount by 4X by the end of next year.</a> We’re not going to lose who we are because it's scrappy maneuverability that lets us innovate faster and more affordably than everyone else. That’s how we built the first PetaOp chip and that's how we're going to build what's next.”<h1>Conclusion</h1>The power of computing is at a tipping point where major players are making decisions about the direction they will pursue that will have huge impacts on the future of technological innovation and potential. Groq is one of these companies that understands it has the potential to completely change the direction of computing and massively influence the potential of AI. With that potential comes responsibility - something Ross and his team have demonstrated they are ready to shoulder. Interested in learning more about Groq? Head to their website <a href="http://www.groq.com" target="_blank">www.groq.com</a> to learn more about the future of computing and review their <a href="https://www.linkedin.com/company/groq/jobs/" target="_blank">multiple career opportunities</a>.<hr><h1>About the sponsor: &nbsp;Groq</h1>Building the computer for the next generation of high performance machine learning. Groq hardware is designed to be both high performance and highly responsive. Groq’s new simplified architecture drives incredible performance at batch size 1. Whether you have one image or a million, Groq hardware responds faster. &nbsp;<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1603879016114-groq++logo+banner.jpg" class="fr-fic fr-dib"> <div language="JavaScript"> </div>

Jonathan Ross, Founder, and CEO of Groq delivered a keynote speech as part of the AI Summit on September 29, 2020, describing insights on the industry, the market, and the unique Groq architecture capable of producing the world’s first PetaOp chip

The shortest path to performance requires a new approach

Today, artificial intelligence — and the computing systems that underlie it — are more than just matters of technology; they are matters of state and society, of governance and the public interest.&nbsp;The choices that technologists, policymakers, and communities make in the next few years will shape the relationship between machines and humans for decades to come.The rapidly increasing applicability of AI has prompted a number of organizations to develop high-level principles on social and ethical issues such as privacy, fairness, bias, transparency, and accountability. Building on those broader principles, the&nbsp;<a href="https://aipolicyforum.mit.edu/" target="_blank">AI Policy Forum</a>, a global effort convened by the MIT Stephen A. Schwarzman College of Computing, will provide an overarching policy framework and tools for governments and companies to implement in concrete ways.“Our goal is to help policymakers in making practical decisions about AI policy,” says Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing. “We are not trying to develop another set of principles around AI, several of which already exist, but rather provide context and guidelines specific to a field of use of AI to help policymakers around the world with implementation.”“Moving beyond principles means understanding trade-offs and identifying the technical tools and the policy levers to address them. We created the college to examine and address these types of issues, but this can’t be a siloed effort. We need for this to be a global collaboration and engage scientists, technologists, policymakers, and business leaders,” says MIT Provost Martin Schmidt. “This is a challenging and complex process for which we need all hands-on deck.”The AI Policy Forum is designed as a yearlong process. Activities associated with this effort will be distinguished by their focus on tangible outcomes — their engagement with key government officials at the local, national, and international level charged with designing those public policies, and their deep technical grounding in the latest advances in the science of AI.&nbsp;The measure of success will be whether these efforts have bridged the gap between these communities, translated principled agreement into actionable outcomes, and helped create the conditions for deeper trust between humans and machines.The global collaboration will begin in late 2020 and early 2021 with a series of AI Policy Forum Task Forces, chaired by MIT researchers and bringing together the world’s leading technical and policy experts on some of the most pressing issues of AI policy, starting with AI in finance and mobility. Further task forces throughout 2021 will convene more communities of practice with the shared aim of designing the next chapter of AI: one that both delivers on AI’s innovative potential and responds to society’s needs.Each task force will produce results that inform concrete public policies and frameworks for the next chapter of AI, and help define the roles that the academic and business communities, civil society, and governments will need to play in making it a reality. Research from the task forces will feed into the development of the AI Policy Framework, a dynamic assessment tool that will help governments gauge their own progress on AI policy-making goals and guide application of best practices appropriate to their own national priorities.On May 6–7, 2021, MIT will host — most likely online — the first AI Policy Forum Summit, a two-day collaborative gathering to discuss the progress of the task forces towards equipping high-level decision-makers with a deeper understanding of the tools at their disposal — and trade-offs to be made — to produce better public policy around AI, and better AI systems with concern for public policy. Then, in fall 2021, a follow-on event at MIT will bring together leaders from across sectors and countries and, built atop the leading research from the task forces, the forum will provide a focal point for work to move from AI principles to AI practice, and serve as a springboard to global efforts to design the future of AI.

Convened by the MIT Schwarzman College of Computing, the AI Policy Forum will develop frameworks and tools for governments and companies to implement concrete policies.

A global collaboration to move artificial intelligence principles to practice

“The foundation of the scientific method is that research results must be testable by others. Testability is even more important in clinical applications because we need a high level of confidence in our methods before they are used with patients,” says Harvard professor&nbsp;John Quackenbush, co-author of the new piece.&nbsp;“In [AI] applications&nbsp;this requires that the models, software code, and data are available for independent validation. Transparency will accelerate research, advance patient care, and will build confidence among scientists and clinicians.”The <a href="https://www.nature.com/articles/s41586-020-2766-y" target="_blank">article</a>, co-authored by more than two dozen researchers from around the world, was published online in Nature on October 14, 2020.Quackenbush and several colleagues organized the commentary in response to a January 2020 study led by researchers at Google Health in which the researchers claimed that an AI system they developed was, in certain settings, better at screening for breast cancer than trained radiologists. The Google Health study also claimed that the AI system improved the speed and reliability of breast cancer screenings. The study enjoyed wide media coverage at the time of its publication.Researchers not involved with the original study, however, have been unable to reproduce the findings due to a lack of details about the methods and algorithm code. AI methods run the risk of “overfitting,” or working only with the specific dataset being tested. This can only be addressed by understanding and testing the methods outside of the original study. The lack of reproducibility impedes cancer research and could lead to unwarranted and even potentially harmful clinical trials, according to the commentary.The authors of the commentary wrote that “transparency in the form of the actual computer code used to train a model and arrive at its final set of parameters is essential for research reproducibility.” They also raised concern that the Google Health study relied on two large datasets that are under license and cannot be easily accessed by outside researchers.While there are numerous obstacles to overcome in order to improve transparency and reproducibility when applying AI methods in medicine, the commentary noted there is a growing number of effective frameworks and platforms to share code, overcome software challenges of large-scale machine learning applications, and ensure patient privacy.“Transparency and reproducibility in artificial intelligence,” Benjamin Haibe-Kains, George Alexandru Adam, Ahmed Hosny, Farnoosh Khodakarami, Massive Analysis Quality Control (MAQC) Society Board of Directors, Levi Waldron, Bo Wang, Chris McIntosh, Anna Goldenberg, Anshul Kundaje, Casey S. Greene, Tamara Broderick, Michael M. Hoffman, Jeffrey T. Leek, Keegan Korthauer, Wolfgang Huber, Alvis Brazma, Joelle Pineau, Robert Tibshirani, Trevor Hastie, John P. A. Ioannidis, John Quackenbush, Hugo J. W. L. Aerts, Nature, online October 15, 2020, doi: 10.1038/s41586-020-2766-y

Scientists working at the intersection of AI and cancer care need to be more transparent about their methods and publish research that is reproducible, according to a new commentary co-authored by CSAIL's Tamara Broderick.

Op-ed: medical AI needs more transparency, reproducibility

Artificial intelligence has arrived in our everyday lives—from search engines to self-driving cars. This has to do with the enormous computing power that has become available in recent years. But new results from AI research now show that simpler, smaller neural networks can be used to solve certain tasks even better, more efficiently, and more reliably than ever before.An international research team from MIT CSAIL, TU Wien (Vienna) and IST Austria has developed a new artificial intelligence system based on the brains of tiny animals, such as threadworms. This novel AI-system can control a vehicle with just a small number of artificial neurons.&nbsp;<iframe src="https://www.youtube.com/embed/8KBOf7NJh4Y?&amp;wmode=opaque" frameborder="0" allowfullscreen="" class="fr-draggable" style="width: 789px; height: 437px;"></iframe>The team says that the system has decisive advantages over previous deep learning models: It copes much better with noisy input, and, because of its simplicity, its mode of operation can be explained in detail. It does not have to be regarded as a complex “black box”, but it can be understood by humans. This new deep learning model has now been published in the journal&nbsp;Nature Machine Intelligence.<h2>Learning from nature</h2>Similar to living brains, artificial neural networks consist of many individual cells. When a cell is active, it sends a signal to other cells. All signals received by the next cell are combined to decide whether this cell will become active as well. The way in which one cell influences the activity of the next determines the behavior of the system—these parameters are adjusted in an automatic learning process until the neural network can solve a specific task.“For years, we have been investigating what we can learn from nature to improve deep learning,” says Prof. Radu Grosu, head of the research group “Cyber-Physical Systems” at TU Wien. “The nematode&nbsp;C. elegans, for example, lives its life with an amazingly small number of neurons, and still shows interesting behavioral patterns. This is due to the efficient and harmonious way the nematode’s nervous system processes information.”“Nature shows us that there is still a lot of room for improvement,” says Prof. Daniela Rus, director of MIT CSAIL. “Therefore, our goal was to massively reduce complexity and develop a new kind of neural network architecture.”“Inspired by nature, we developed<a href="https://arxiv.org/abs/2006.04439" target="_blank">&nbsp;new mathematical models</a> of neurons and synapses,” says Prof. Thomas Henzinger, president of IST Austria.“The processing of the signals within the individual cells follows different mathematical principles than previous deep learning models,” says Dr. Ramin Hasani, postdoctoral associate at the Institute of Computer Engineering, TU Wien and MIT CSAIL. “Also, our networks are highly sparse—this means that not every cell is connected to every other cell. This also makes the network simpler.”[YK3]&nbsp;<h2>Autonomous Lane Keeping</h2>To test the new ideas, the team chose a particularly important test task: self-driving cars staying in their lane. The neural network receives camera images of the road as input and is to decide automatically whether to steer to the right or left.“Today, deep learning models with many millions of parameters are often used for learning complex tasks such as autonomous driving,” says Mathias Lechner, TU Wien alumnus and PhD student at IST Austria. “However, our new approach enables us to reduce the size of the networks by two orders of magnitude. Our systems only use 75,000 trainable parameters.”The new system consists of two parts: “The camera input is first processed by a so-called convolutional neural network,” says Alexander Amini, a PhD student at MIT CSAIL. “It only processes the visual data to extract structural features from incoming pixels. It has nothing to do with the actual steering of the vehicle.”&nbsp;This network decides which parts of the camera image are interesting and important, and then passes signals to the crucial part of the network – a “control system” that then steers the vehicle.Both subsystems are stacked together and are trained simultaneously. Many hours of traffic videos of human driving in the greater Boston area were collected, and are fed into the network, together with information on how to steer the car in any given situation—until the system has learned to automatically connect images with the appropriate steering direction and can independently handle new situations.The control part of the system (called&nbsp;neural circuit policy, or NCP), which translates the data from the perception module into a steering command, only consists of 19 neurons. “This is orders of magnitude smaller than what would have been possible with previous state-of-the-art models,”&nbsp;[MOU5]&nbsp;says Mathias Lechner.<h2>Causality and Interpretability</h2>The new deep learning model was tested on a real autonomous vehicle. “Our model allows us to investigate what the network focuses its attention on while driving. Our networks focus on very specific parts of the camera picture: The curbside and the horizon. This behavior is highly desirable, and it is unique among artificial intelligence systems,” says Ramin Hasani. “Moreover, we saw that the role of every single cell at any driving decision can be identified. We can understand the function of individual cells and their behavior. Achieving this degree of interpretability is impossible for larger deep learning models.”<h2>Robustness</h2>“To test how robust NCPs are compared to previous deep models, we perturbed the input images and evaluated how well the agents can deal with the noise,” says Mathias Lechner. “While this became an insurmountable problem for other deep neural networks, our NCPs demonstrated strong resistance to input artifacts. This attribute is a direct consequence of the novel neural model and the architecture.”“Interpretability and robustness are the two major advantages of our new model,” says Ramin Hasani. “But there is more: Using our new methods, we can also reduce training time and the possibility to implement AI in relatively simple systems. Our NCPs enable imitation learning in a wide range of possible applications, from automated work in warehouses to robot locomotion. The new findings open up important new perspectives for the AI community: The principles of computation in biological nervous systems can become a great resource for creating high-performance interpretable AI—as an alternative to the black-box machine learning systems we have used so far.”

Artificial intelligence (AI) can become more efficient and reliable if it is made to mimic biological models. New approaches in AI research are hugely successful in experiments.

New deep learning models require fewer neurons

What can Artificial Intelligence (AI) and robotics do to improve the recycling rate for large pieces of waste? This is the central question in the cooperation project “Smart Recycling” between the leading German Research Center for Artificial Intelligence (DFKI) and the Institute for Energy and Circular Economy (IEKrW) at the University of Bremen. The project partners want to develop appropriate concepts by mid-December. Both technical and economic challenges are taken into account.&nbsp;Although the solutions are primarily intended for the circular economy, they should also be able to be used in other areas.&nbsp;Examples: disaster control, nature conservation through automated waste collection, sustainable agriculture, environmentally friendly mining, maintenance and care of infrastructures or Industry 4.0.&nbsp;In addition to the DFKI and the IEKrW, the Hamburg University of Applied Sciences and Nehlsen Stoffstrom GmbH &amp; Co. KG are involved in the project, which will run until December 15, 2020.&nbsp;The “Smart Recycling” project is funded by the Federal Environment Ministry within the funding priority “AI beacons for the environment, climate, nature and resources”. With its efficient Kl- and robotics-based solution, the "Smart Recycling" project supports, among other things, the goals set in the German Resource Efficiency Program to increase the recycling rate for municipal waste to over 65 percent, the recycling rate of plastic waste and the use of recycled building materials. A significant ecological benefit is expected from the more efficient recycling of large pieces of waste, especially in the construction industry: Of the around 412 million tons of waste in Germany, 220 million tons are construction and demolition waste (data from 2017, Federal Statistical Office).&nbsp;The sorting of waste plays a decisive role in increasing resource efficiency.&nbsp;This is where the Smart Recycling project comes in: the use of modern sensor technology, CL methods and robotics is intended to improve the sorting of waste and thus increase resource efficiency.&nbsp;The project will determine the use of the techniques mentioned and suggest appropriate approaches to increase efficiency.<h2>Background:</h2>The circular economy is about using products and raw materials for as long as possible. If that is no longer possible, raw materials are broken down into their raw materials and recycled. Automated separation of these substances is particularly efficient and therefore environmentally friendly. However, this separation is currently only possible for small pieces of waste or waste that has already been shredded, so-called flakes. Waste that is in large pieces, such as building rubble or bulky waste, must first be shredded for automated sorting. This is usually done manually or with manually operated technology and is therefore slow and cost-intensive. The recycling rate is correspondingly low, so that many valuable raw materials are disposed of together with residual waste or, at best, are recycled with little value.<hr>(Press release from the University of Bremen)

Without recycling processes, waste such as building rubble is processed further unsorted. Source: AdobeStock

What can Artificial Intelligence (AI) and robotics do to improve the recycling rate for large pieces of waste?

Artificial Intelligence And Robotics For A Sustainable Circular Economy

Regardless of which field of profession you are in, you are most definitely using a smart device connected to the internet on a regular/semi-regular basis. In this day and age, you'd have to be living in the woods to be cut off from the digital mesh that encompasses the world. If you're on a University or Office network, then you're part of a LAN(Local Area Network) that inturn connects to many other such smaller branches and finally back to the internet. But do you know if you're the only person with the access to the contents on your device? Are you sure? Think again!This article is merely meant to be an introductory tool to get you warmed up about cyberthreats, cybersecurity and how AI based cybersecurity is trying to address this. I am not an expert on all the different branches of cybersecurity, so take this with a grain of salt. In cybersecurity terms, a person with malicious intent trying to access a secure network is called an attacker&nbsp;or more colloquially a <a href="https://en.wikipedia.org/wiki/Hacker" rel="noopener noreferrer" target="_blank">hacker</a>. A dummy system that looks and functions similar to the real systems on the network is called a <a href="https://en.wikipedia.org/wiki/Deception_technology" rel="noopener noreferrer" target="_blank">decoy</a>. One such decoy that intends to trap and store the details about the attacker that can be used to traceback his/her identity is called a <a href="https://en.wikipedia.org/wiki/Honeypot_(computing)" rel="noopener noreferrer" target="_blank">honeypot</a>. A security tool that tries to prevent any external actor from entering a secure network without proper authentication is called a <a href="https://en.wikipedia.org/wiki/Firewall" rel="noopener noreferrer" target="_blank">firewall</a>. Every input-output combination that happens within a computer, causes the computer to send-receive several internal communication signals called <a href="https://en.wikipedia.org/wiki/System_call" rel="noopener noreferrer" target="_blank">system calls</a>. A perfectly objective entity that follows all the mathematical rules of logic is called a <a href="https://en.wikipedia.org/wiki/Rational_agent" rel="noopener noreferrer" target="_blank">rational actor</a> and any influence of the world like relationships, need for materialistic possessions, etc makes the entity's logic imperfect and these imperfections are called <a href="https://en.wikipedia.org/wiki/Bias" rel="noopener noreferrer" target="_blank">biases</a>. A <a href="https://en.wikipedia.org/wiki/State_(computer_science)" rel="noopener noreferrer" target="_blank">state</a>&nbsp;is the comprehensive description of all relevant details that are needed to make an informed, rational decision about a problem. We will be using these terms throughout this article.So how do we know if there is an attacker in our network? How do we catch him in the act or better yet, prevent him from entering the network altogether? Attackers are highly skilled actors who have intricate knowledge about internal system/network architecture and often engage in research about the specific network they are trying to enter for several days or even weeks using scouting operations before they even attempt an attack. Also an attack can be performed for any number of reasons ranging from establishing a constant presence within a secure network to damaging/stealing sensitive data. So preventing the attacker from ever entering a network, although would be an ideal scenario, isn't entirely practicable if you think about it. Then can we atleast detect and prevent the attacker before they gain control over the network? For that we first have to learn about the different types of possible attacks that are common in a network. These are explained more comprehensively in this <a href="https://www.cisco.com/c/en/us/products/security/common-cyberattacks.html#~types-of-cyber-attacks" rel="noopener noreferrer" target="_blank">article</a> published by cisco so I'm just going to give you a simple one line idea of some important ones:<ol><li>Malware attack - Installs malicious software on your computer that can later be used as an access point.</li><li>Phishing - Deceiving the end user by using a platform that looks like a known, trustworthy source to gain access.</li><li>Man-in-the-Middle - When the attacker eavesdrops on a secure communication line between two parties to gain resources.</li><li>Denial of Service - When the attacker floods the network with rubbish signals to use up the bandwidth otherwise needed by the end user.</li><li>SQL injection - Using malicious code to exploit vulnerabilities in the network and gain access to restricted information.&nbsp;</li></ol>So how do we detect each of these attacks? The most straightforward way would be to constantly monitor the resources used within the network and map it back to the kind of attack that would need that unique set of resources. This is where we can look at system calls. A system calls log would give us a frozen in time screenshot of all the processes that were actively communicating at the time of logging. By using pattern recognition based on years of research done in the field, we can determine the kind of &nbsp;and the level of attack that's currently underway. But is it really that simple? No. Firstly, millions of system calls happen every minute in a single computer, let alone on a network of computers, so logging them and parsing them is no easy task. Secondly, storing such huge log files and doing pattern recognition on a humongous clutter of data is extremely complicated. So how can we do it? This is where an AI can help.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1599878984215-TiADD.png" style="width: 300px;" class="fr-fic fr-dib">Artificial Intelligence algorithms have been trying to solve problems in all domains for the past few decades now and they've clearly outperformed and outsmarted any single human actor. Additionally, AI algorithms are designed to model and learn from copious data without any human intervention or having to explicitly program the behavior. Since I am not an expert on all the methods currently at the cutting edge of cybersecurity, I will stick to what researchers at THINC Lab are trying to implement. At THINC Lab, we use a tool that's quite popular in classical AI called the <a href="https://en.wikipedia.org/wiki/Partially_observable_Markov_decision_process" rel="noopener noreferrer" target="_blank">Partially Observable Markov Decision Process</a>(POMDP). Without going into the gory details of what that means, it is a tool that helps create a recursive model of reasoning. It tries to model the decision making of the attacker(s) for every counter action that the security software performs to cripple them. This way it can figure out the most optimal action to perform at any given timestep and environment state.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1599879016382-deception_dynamics.png" style="width: 300px;" class="fr-fic fr-dib">In order to feed the details of the state into the POMDP to make a rational decision, we need to do the datamining of system logs that I mentioned earlier. For this, we use a <a href="https://en.wikipedia.org/wiki/Convolutional_neural_network" rel="noopener noreferrer" target="_blank">Convolutional Neural Network</a> to learn the pattern and give us key features about the statespace. Following this, we can use decoys and honeypots to either mislead the attacker into thinking he won or give up less sensitive data while prioritizing protecting the most sensitive files. After testing on simulated attacks, this tool is now ready to be tested on more complex platforms. Eventually we intend to release a commercial version that can be installed just like an antivirus software and used seamlessly on any device.&nbsp;One caveat with this tool is that POMDPs in its current form can only model fully rational actors which means it assumes that the human attacker always performs the most optimal action. This is not true. Humans are sub-optimal agents who are bound to have biases influencing their decisions. This is the current research topic at THINC lab. We are trying to make the decision making more realistic by trying to model the kind of biases an attacker can have based on the actions he has performed so far. That ends the short tour on cyberthreats, cyber security and AI based cybersecurity models THINC lab is working on. If you want to learn more about what we do, check out our <a href="http://thinc.cs.uga.edu" rel="noopener noreferrer" target="_blank">website</a>! I hope you are walking out of this one level more aware than you came in. Thanks for taking this trip on my train of thought!

Is your device/network safe? Source: https://pixabay.com/photos/hacker-hack-anonymous-hacking-5332676/

No need to worry, AI based cybersecurity can help you!

Are hackers in your network right now?

Biomedical engineers at Duke University have shown that different strains of the same bacterial pathogen can be distinguished by a machine learning analysis of their growth dynamics alone, which can then also accurately predict other traits such as resistance to antibiotics. The demonstration could point to methods for identifying diseases and predicting their behaviors that are faster, simpler, less expensive and more accurate than current standard techniques.The results appear online on August 3 in the Proceedings of the National Academy of Sciences.For most of the history of microbiology, bacteria identification has relied on growing cultures and analyzing the physical traits and behaviors of the resulting bacterial colony. It wasn’t until recently that scientists could simply run a genetic test.Genetic sequencing, however, isn’t universally available and can often take a long time. And even with the ability to sequence entire genomes, it can be difficult to tie specific genetic variations to different behaviors in the real world.For example, even though researchers know the genetic mutations that help shield/protect bacteria from beta-lactam antibiotics—the most commonly used antibiotic in the world—sometimes the DNA isn’t the whole story. While a single resistant bacteria usually can’t survive a dose of antibiotics on its own, large populations often can.<a href="https://bme.duke.edu/faculty/lingchong-you" target="_blank">Lingchong You</a>, professor of biomedical engineering at Duke, and his graduate student, Carolyn Zhang, wondered if a new twist on older methods might work better. Maybe they could amplify one specific physical characteristic and use it to not only identify the pathogen, but to make an educated guess about other traits such as antibiotic resistance.“We thought that the slight variance in the genes between strains of bacteria might have a subtle effect on their metabolism,” You said. “But because bacterial growth is exponential, that subtle effect could be amplified enough for us to take advantage of it. To me, that notion is somewhat intuitive, but I was surprised at how well it actually worked.”How quickly a bacterial culture grows in a laboratory depends on the richness of the media it is growing in and its chemical environment. But as the population grows, the culture consumes nutrients and produces chemical byproducts. Even if different strains start with the exact same environmental conditions, subtle differences in how they grow and influence their surroundings accumulate over time.In the study, You and Zhang took more than 200 strains of bacterial pathogens, most of which were variations of&nbsp;E. coli, put them into identical growth environments, and carefully measured their population density as it increased. Because of their slight genetic differences, the cultures grew in fits and starts, each possessing a unique temporal fluctuation pattern. The researchers then fed the growth dynamics data into a machine learning program, which taught itself to identify and match the growth profiles to the different strains.To their surprise, it worked really well.“Using growth data from only one initial condition, the model was able to identify a particular strain with more than 92 percent accuracy,” You said. “And when we used four different starting environments instead of one, that accuracy rose to about 98 percent.”Taking this idea one step further, You and Zhang then looked to see if they could use growth dynamic profiles to predict another phenotype—antibiotic resistance.The researchers once again loaded a machine learning program with the growth dynamic profiles from all but one of the various strains, along with data about their resilience to four different antibiotics. They then tested to see if the resulting model could predict the final strain’s antibiotic resistances from its growth profile. To bulk up their dataset, they repeated this process for all of the other strains.The results showed that the growth dynamic profile alone could successfully predict a strain’s resistance to antibiotics 60 to 75 percent of the time.“This is actually on par or better than some of the current techniques in the literature, including many that use genetic sequencing data,” said You. “And this was just a proof of principle. We believe that with higher-resolution data of the growth dynamics, we could do an even better job in the long term.”The researchers also looked to see if the strains exhibiting similar growth curves also had similar genetic profiles. As it turns out, the two are completely uncorrelated, demonstrating once again how difficult it can be to map cellular traits and behaviors to specific stretches of DNA.Moving forward, You plans to optimize the growth curve procedure to reduce the time it takes to identify a strain from 2 to 3 days to perhaps 12 hours. He’s also planning on using high-definition cameras to see if mapping how bacterial colonies grow in space in a Petri dish can help make the process even more accurate.This research was conducted in collaboration with groups of Deverick J. Anderson, Joshua T. Thaden and Vance G. Fowler from the Duke University School of Medicine, and Minfeng Xiao from BGI Genomics.This research was partially supported by the National Institutes of Health (LY, R01GM098642, R01GM110494, 1A1125604), the Army Research Office (LY, W911NF-14-1-0490), the David and Lucile Packard Foundation, the Shenzhen Peacock Team Plan grant (MX, No. KQTD2015033117210153), the Centers for Disease Control and Prevention (DJA, U54CK000164), AHRQ (DJA, R01-HS23821), NIH (VGF, R01-AI068804), and the National Science Foundation’s Graduate Research Fellowship (CZ, HRM).“Temporal encoding of bacterial identity and traits in growth dynamics.” Carolyn Zhang, Wenchen Song, Helena R. Ma, Xiao Peng, Deverick J. Anderson, Vance G. Fowler Jr, Joshua T. Thaden, Minfeng Xiao, and Lingchong You. PNAS, 2020. DOI: 10.1073/pnas.2008807117

A new method for identifying strains of bacteria and guessing their resistance to antibiotics uses an AI model to analyze their growth dynamics in culture.

Machine learning algorithm uses high-temporal-resolution growth curves to identify pathogens with 98% accuracy and predict antibiotic resistance just as well as genetic-based methods

AI May Offer a Better Way to ID Drug-Resistant Superbugs

<h2 data-selectable-paragraph="">Introduction</h2>In a small meeting room on a Tuesday morning, I showed a short video we created to the people in the room. After it finished I glanced over to find the first and only journalist to win four Pulitzer Prizes with his head in his hands, contemplating the end of truth and reality as we know it. How did we get to this place?<iframe src="https://www.youtube.com/embed/i7QNUZWS6VE?&wmode=opaque&enablejsapi=1&origin=https://www.wevolver.com" frameborder="0" allowfullscreen="" class="fr-draggable" style="width: 789px; height: 442px;" data-gtm-yt-inspected-9="true" data-lf-form-tracking-inspected-kn9eq4roawkarlvp="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" data-lf-vimeo-playback-inspected-kn9eq4roawkarlvp="true"></iframe>A new version of the deepfake video we initially shared with The NYT is now available on YouTube.&nbsp;Earlier this summer, our company Dessa became known around the world for making <a href="https://medium.com/dessa-news/real-talk-speech-synthesis-5dd0897eef7f" rel="noopener" target="_blank">the most realistic AI voice to date</a>, which replicated one of the most famous voices in podcasting, <a href="http://podcasts.joerogan.net/" rel="noopener nofollow" target="_blank">Joe Rogan</a>. The AI voice, which was built using a text-to-speech synthesis system we developed called RealTalk, could say any words we wanted it to say, even if Rogan had never said them before.<blockquote>We didn&rsquo;t just do this for kicks. One of our main objectives with releasing RealTalk was to get the public to take notice of just how realistic deepfakes were becoming, and to understand the tremendous implications at hand for society as a result.</blockquote>It worked.Since we released RealTalk in May, we have been shocked by the huge wave of public reception to it. More than 2 million people have viewed our video featuring clips of the <a href="https://www.youtube.com/watch?v=DWK_iYBl8cA&t=11s" rel="noopener nofollow" target="_blank">faux Rogan voice on YouTube</a>. Awareness about the work also quickly spread around the world, with journalists from over 20 different countries covering it (you can find a selection of the news coverage <a href="https://www.dessa.com/newsroom" rel="noopener nofollow" target="_blank">here</a>).One of the journalists who became interested in the work was David Barstow, the journalist we mentioned earlier, who emailed us shortly after we shared RealTalk publicly. At first, I had to do a double take to see whether or not the email was a prank. Here was someone who had taken on some of the world&rsquo;s biggest stories (most recently, revealing <a href="https://www.nytimes.com/interactive/2018/10/02/us/politics/donald-trump-tax-schemes-fred-trump.html" rel="noopener nofollow" target="_blank">Trump&rsquo;s tax evasion schemes</a>), getting in touch with us, wanting to work with us on a story about deepfakes.Not nervous or heart-attacky at all, I agreed to have a call with him the same day. David said he had first become interested in learning more about deepfakes after seeing his son showed him a video of a now legendary deepfake that <a href="https://www.vox.com/2018/4/18/17252410/jordan-peele-obama-deepfake-buzzfeed" rel="noopener nofollow" target="_blank">Jordan Peele had created of Obama</a>, and was thinking of their implications especially as the 2020 American election drew nearer.During the call, David asked me if we had thought about pairing synthetic audio with synthetic video to create a fully fledged deepfake (one that would be nearly impossible to discern as fake), and likened such a feat to an Edison moment. If the team were to attempt such a project, he wondered, would we also be interested in being filmed during the process, and ultimately becoming the subject of a documentary on The Weekly?We had to pinch ourselves after that initial phone call. But we were confident that if anyone could distill why it was important for the world to know about deepfakes, it was David Barstow. And the truth was, it was like he had read our minds. Shortly after releasing RealTalk, and before we had that call, we had already started to think about merging audio and video as a next phase of the project, certain it would help us in our mission to spread even more awareness about the power and danger of deepfakes.After that call with David, we quickly decided that opening our doors to The New York Times in the way we did was a no-brainer. Having the work showcased in a documentary by The Times would mean that millions more people would become aware of deepfakes and their implications. And we would have the credibility of The New York Times behind us. It really seemed like the ultimate soapbox for our message.At the same time, along the way we were starkly aware of the risk it simultaneously posed to our reputation. What if they took our work, but chose not to share the intentions we had behind it? There was a huge risk that we would come across as evil geniuses or, less worse, idealistic tech bros. But ultimately, we determined that this was a risk worth taking.Six months later, we&rsquo;re excited to share this work with the world.It may not seem like it, but the timeline we set aside to create this work was a very short one (especially since we were juggling many other projects simultaneously throughout this period). To maximize our control of the message, however, we knew we would have to act fast, since we were confident many others with machine learning expertise would be working on similar projects at the same time we were.The unfortunate reality is that at sometime in the near future, deepfake audio and video will be weaponized. Before that happens, it&rsquo;s crucial that machine learning practitioners like us who can spread the word do so proactively, helping as many people as possible become aware of deepfakes, and the potential ways they can be misused to distort the truth or harm the integrity of others. Ultimately that is why we decided we had to be the first to reach the finish line.At this point, it&rsquo;s also worth noting that we will not be sharing the model, code or data used to create the deepfake publicly. You might be thinking that this sounds extreme, unfair to ML researchers, or just a rehashing of <a href="https://www.theverge.com/2019/11/7/20953040/openai-text-generation-ai-gpt-2-full-model-release-1-5b-parameters" rel="noopener nofollow" target="_blank">Open AI&rsquo;s decision to do a staged release</a> of the GPT-2 language processing model (the AI that was &ldquo;too dangerous to release&rdquo;) first announced earlier this year.It isn&rsquo;t, because the reality is that tools for generating synthetic audio and video (and especially for generating them in tandem) are a lot more dangerous. If we were to open source the tools we used to make RealTalk, they could actually be used to scam people as individuals, or disrupt democracy on a mass scale, and it wouldn&rsquo;t be that hard. In contrast, for an individual to use a model like GPT-2 to spread disinformation, they would need to somehow make tons of fake news websites, and also get traffic to them.That said, there are select pieces of information we can share about the technical process of creating the work we feel comfortable sharing. Here is a brief overview of the building blocks we used to create our hyper-realistic deepfake video of Joe Rogan announcing the end of his podcast.<h2 data-selectable-paragraph="">Technical Overview</h2>Audio:&nbsp;To synthesize the audio used in our final version of the video, we used the same RealTalk model we developed to recreate Joe Rogan&rsquo;s voice with AI back in May. RealTalk uses an attention-based sequence-to-sequence architecture for a text-to-spectrogram model, also employing a modified version of <a href="https://deepmind.com/blog/wavenet-generative-model-raw-audio/" rel="noopener nofollow" target="_blank">WaveNet</a> that functions as a neural vocoder.The final dataset our machine learning engineers used to replicate Rogan&rsquo;s voice consists of eight hours of clean audio and transcript data, optimized for the task of text-to-speech synthesis. The eight-hour dataset contains approximately 4000 clips, each consisting of Joe Rogan saying a single sentence. The clips range from 7&ndash;14 seconds long respectively. You can find a more in-depth blog post about the technical underpinnings of RealTalk we released earlier this summer <a href="https://medium.com/dessa-news/realtalk-how-it-works-94c1afda62f0" rel="noopener" target="_blank">here</a>.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNTk3MTU5NTY1NjQwLTBfaEszakdFVFNTUnBHYmRFdy5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19">The following spectrogram was generated by the RealTalk model to showcase a sample from the text-to-spectrogram model output. Areas highlighted in red designate normal, deep, short and long breaths respectively. In the training data, the speaker&rsquo;s speech is much more flat than the following sample, so the engineers used the features discussed earlier to add extra intonations to the speech output.&nbsp;<h2 data-selectable-paragraph="">Video</h2>For the video portion of the work, we used a FaceSwap deepfake technique that many in the machine learning community are already familiar with. The main reason we chose the FaceSwap instead of other techniques is that it looked realistic and was pretty reliable, without requiring many hours of training data to employ properly. This was an important factor for us due to our expedited timelines. To use the FaceSwap technique, we would only need one high-quality and authentic video of Joe Rogan to use as source data, and an actor who resembled him for the video we would use to swap Joe Rogan&rsquo;s face onto. We ultimately picked Paul, shown below, because his physique closely resembled Joe Rogan&rsquo;s.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNTk3MTU5NTk4MDExLTFfaVQ5UUNfcmFyeFdTazJpOTMxXzd2dy5wbmciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19">The image of our actor is on the left, used to generate the deepfake of Joe Rogan seen on the right.&nbsp;Combining Audio &amp; Video:&nbsp;Combining the synthetic audio and synthetic video together is the most difficult part of the process, because there are multiple disconnected steps in the process that make it hard to perfect. First off, the fake audio is generated separately from the video. The fake audio replicates the tone, breathing and pace at which the target speaks (in this case, Joe Rogan). For the video to look convincing, the actor posing as Rogan needed to match the same expressions and speed as the synthesized audio.This presented us with a huge challenge, as the actor had to focus not only on copying the original audio exactly to ensure a lip sync, but also copying the expressions of the audio, as well as moving and behaving in a natural way. The first time either of the actors did manage to match the lip-sync, they basically just stared into one spot unblinkingly while copying the lip movements. Unless we were replicating a robot from an 80s sci-fi movie, this likely wouldn&rsquo;t be very convincing.Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/the-pandemic-of-deepfakes-a-rising-threat-bigger-than-identity-theft" rel="noopener noreferrer" target="_blank">The pandemic of deepfakes : A rising threat bigger than identity theft</a>In short, building a deepfake convincing enough to fool everyone is thankfully something that for now is still really hard to do. But this won&rsquo;t be the case forever.<h2 data-selectable-paragraph="">Conclusion</h2>Why Building Awareness Is CrucialDessa may have achieved a deepfake merging audio and video at this level of high fidelity first, but we will not be the only ones to do it. Many others are working on the same technology, and future versions of this will be much faster and easier to use than ours. As mentioned above, the reality is that deepfakes can (and are already are) being used for malicious purposes.We need to make sure as many people are aware of how advanced this technology is, and we need to do it fast. Malicious actors will move way faster than the rest of us, because they don&rsquo;t care about getting things perfect and don&rsquo;t have bureaucracy slowing them down. They will simply do whatever it takes to demean innocent individuals, steal money, or disrupt democracy.Next steps: Contributing to the field of deepfake forensicsOur machine learning engineers&rsquo; development of hyperrealistic deepfakes has also led us to examine what we can do to help make detecting them easier and more reliable in the wild. Earlier today, we shared a <a href="https://www.dessa.com/post/deepfake-detection-that-actually-works" rel="noopener nofollow" target="_blank">new technical blog post</a> from our engineers that shows how a recent dataset released by Google for deepfake detection falls short at reliably identifying fake videos found in the wild on YouTube. After demonstrating where the dataset falls short, our engineers have also provided a solution for making it more robust and functional on real-world data. This work has also been open-sourced for anyone wanting to replicate and expand on our results over on GitHub. The work we did on deepfake detection was featured today in The New York Times by Cade Metz, the newspaper&rsquo;s AI Reporter. <a href="https://www.nytimes.com/2019/11/24/technology/tech-companies-deepfakes.html" rel="noopener nofollow" target="_blank">Find the article here</a>.Earlier this fall, we also released an article and open source tools for building systems for detecting deepfake audio, which you can find respectively on our <a href="https://medium.com/dessa-news/detecting-audio-deepfakes-f2edfd8e2b35" rel="noopener" target="_blank">Medium</a> and <a href="https://github.com/dessa-public/fake-voice-detection" rel="noopener nofollow" target="_blank">GitHub</a>.

If you think this is an image of Joe Rogan, think again. It’s actually a frame from a hyper-realistic deepfake we created, overlaid with an AI facsimile of his voice. Learn more after the jump.

Raising Awareness About The Dangers of Synthetic Media

We Just Created The Deepest Deepfake Yet. Here's Why.

By building the enormous network of wireless sensors that forms the IoT, we’re generating more data than ever before. More data promises better decision making – but only if we don’t drown in information first. Proper data management starts by processing information at the edge to filter, refine, and then prioritize what’s sent to the Cloud. AI is the best tool for the job.<h2>Dealing with information overload</h2>The amount of data generated by the IoT looks to increase exponentially. For example, analyst Gartner predicts that by 2021, the IoT will comprise 25 billion devices, all generating round-the-clock information. How much information? The International Data Corporation (IDC) projects that connected things will generate no less than 79.4 zettabytes (one zettabyte is equal to a billion terabytes) of data within the next five years.Much of this information will be generated by remote wireless sensors with small batteries. Transmitting this amount of raw data to the Cloud is expensive and drains batteries rapidly. Much of the data is of low value, typically information that confirms a system is stable when vital information would be about how things are changing. It’s much better to process the data locally to determine which is critical and time-sensitive and then use (relatively high power) wireless transmission infrequently and for as short a time as possible to send just that information to the Cloud.<h2>Small AI for edge devices, big AI for the Cloud</h2>Determining which data is essential is best done using AI. But it has to be a particularly well-optimized form of AI to match the power, processor, and memory constraints of IoT edge devices.Edge-AI startup Imagimob specializes in the development of this software. The company takes proven AI software and then optimizes it. It can then run on the Arm M-class processors common to battery-powered IoT sensors and edge devices using Nordic Semiconductor solutions. For example, a system developed by the company employs an AI “neural network” to learn and interpret gestures to control headphones without the need for traditional switches and buttons. All the AI processing power is contained within the battery-powered headphones.With optimized AI, edge devices equipped with processors such as the Arm M-class used in Nordic’s IoT solutions can crunch and store sensor data locally and only send only what’s necessary to the Cloud – minimizing transmission costs and power consumption.According to Steve Poulsen, President of U.S. embedded design consultant and Nordic customer, <a href="https://www.nordicsemi.com/News/2020/02/Signetik-cellular-IoT-platform-can-perform-edge-computed-AI-on-Nordic-LTE-M-NB-IoT-multimode-SiP?utm_campaign=2021%20wevolver" rel="noopener noreferrer" target="_blank">Signetik</a>, the AI’s task should be split into two:<ul><li>Resource-constrained edge devices (“small AI”) Will use optimized software to collate and sort sensor data locally</li><li>Powerful Cloud-based servers (“big AI”) Will use superior resources to quickly and deeply analyze the filtered data for significant trends and patterns.</li></ul><h2>The IoT is transforming the world without us noticing</h2>Today the IoT and AI are all around us, but the changes being brought about by the technologies are so subtle that most people overlook them. It’s like we’re all standing on a slowly rising platform without realizing it. But sectors such as healthcare, logistics, transportation, utility metering, retail payments, lighting, VR, and much more besides are being continuously transformed by the IoT and AI. The “AI of Things” is slowly taking shape. Its influence will help us slow the avalanche of data provided by the IoT - and develop new ways of using the insights it uncovers.The IoT and AI are such a natural fit that their convergence is already leading to intelligent networks that are increasingly capable of solving problems without human assistance across a diverse number of industries. The impact on AI will be dramatic; according to analyst Research and Markets, for instance, the IoT will represent 83 percent of all AI chipsets sales by 2023.By then, the AI of Things will underpin a larger technological revolution that’s already started and is making use of parallel advances in fields such as robotics, quantum computing, autonomous vehicles, and genomics.<hr>This article was first published on Nordic's&nbsp;<a href="https://blog.nordicsemi.com/getconnected" rel="nofollow" target="_blank"></a><a href="https://blog.nordicsemi.com/getconnected?utm_campaign=2021%20wevolver" rel="noopener noreferrer" target="_blank">Get Connected Blog</a>.

By building the enormous network of wireless sensors that forms the IoT, we’re generating more data than ever before. More data promises better decision making – but only if we don’t drown in information first.

Why AI is set to reshape the IoT

Voice command will continue to become a natural and broadly accessible method of interacting with technology. As described in our previous article, Voice At The Edge Helps Technology Work For More People, controlling devices with speech offers a huge range of advantages to users and designers.&nbsp;Devices with voice interfaces are <a href="https://www.juniperresearch.com/press/press-releases/digital-voice-assistants-in-use-to-8-million-2023" rel="noopener noreferrer" target="_blank">expected to rise</a> from 3 billion devices worldwide to 8 billion by the end of 2035. These won’t just be more sophisticated versions of the home assistants that we know today, but a whole new range of devices and applications that will offer natural voice interaction.&nbsp;In this article, we dive deeper into how voice command on edge devices works, and take a look at the leading low power devices available for voice command system design by examining a recent <a href="https://www.syntiant.com/post/keyword-spotting-power-comparison" rel="noopener noreferrer" target="_blank">white paper</a> from Syntiant.<h1 dir="ltr">The basics of keyword spotting</h1>Smart devices are always active - listening for the predefined “wake” up instruction, such as "Alexa," "Ok Google," etc. This action is commonly known as keyword spotting (KWS). When a keyword is detected, the device wakes up and its full-scale voice recognition is activated either on the cloud or on the device. In some devices, keywords are also used as voice commands, for example, ‘hello, start, stop’ in a voice-enabled robot vacuum.&nbsp;As a KWS system is always-on, it must have very low power consumption. &nbsp;A KWS system also should detect the spoken keywords with high accuracy and low latency to attain the best user experience. Low power consumption is especially important for portable devices, such as smartphones and wearables.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNjAxMjc4NDg5ODUxLXNvbnkuanBnIiwiZWRpdHMiOnsicmVzaXplIjp7IndpZHRoIjo5NTAsImZpdCI6ImNvdmVyIn19fQ==">Image: SonyThese tough conflicting system requirements, in combination with the increasing amount of voice command applications and audiences, make KWS a hugely relevant research area.&nbsp;Until recently, the only devices capable of performing KWS at edge-compatible power levels were low-power microcontroller units (MCUs) and digital signal processors (DSPs). Processor-designer, Syntiant, has recently introduced the NDP10x series of its Neural Decision Processors (NDPs), a suite of highly efficient chips purpose-built for KWS using deep neural networks that increase the options for system designers.<h1 dir="ltr">Introduction to microcontrollers</h1>Microcontrollers, sometimes known as an embedded controller, &nbsp;are present in applications with and without voice command, such as vehicles, robots, office machines, medical devices, vending machines, and home appliances. They were first developed by Intel in the early 1970s and are now produced by many other manufacturers, including &nbsp;Microchip (PIC microcontrollers), Atmel ( AVR microcontrollers), Hitachi, Phillips, Maxim, NXP, Intel, etc.The need for low power consumption for keyword spotting systems makes microcontrollers a natural choice for deploying KWS in an always-on system. However, deployment of neural network-based KWS on microcontrollers comes with the challenges of limited memory footprint and limited computer resources.&nbsp;<h1 dir="ltr">Introduction to neural decision processors</h1>The SyntiantR NDP100TM microwatt-power <a href="https://www.syntiant.com/" rel="noopener noreferrer" target="_blank">Neural Decision Processor</a>TM is the first and currently the only commercial silicon device designed to put machine learning processing into almost any consumer device. The custom-built chips are specialized to run TensorFlow neural networks. TensorFlow is an open-source software library developed by Google and available for free for dataflow and differentiable programming across a range of tasks.&nbsp;The Syntiant NDP is specifically designed for devices that use voice commands and have ultra-low power requirements making them well-suited to devices with KWS systems.The Syntiant NDP100 and NDP101 chips have been certified by Amazon, which enables the company’s neural processors to be used in Alexa-enabled devices, however, Syntiant NDPs are not limited to Alexa devices only.<img src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNTk3MzE4NzkyNTM2LWNyb3BwZWQgZWFyYnVkcyBzeW50aWFudC5qcGciLCJlZGl0cyI6eyJyZXNpemUiOnsid2lkdGgiOjk1MCwiZml0IjoiY292ZXIifX19">The Syntiant NDP100 microwatt-power Neural Decision Processor&nbsp;<h1 dir="ltr">How do they compare and what do I need for my device?</h1>System designers are faced with a growing number of high-quality options for adding a KWS algorithm to their devices. As previously described, a voice command interface often requires the device to be ‘always on’ and listening for its wake words, thus low power consumption is essential. However, it can be difficult to compare the power consumption of different devices from different manufacturers. Datasheets may describe current consumption in uA/MHz, but it is not generally possible to convert that into the power required for a given task. Additionally, different network topologies, training procedures, etc., can affect system accuracy and power consumption.The critical question system designers must ask is how much power consumption will be added by a solution running a KWS algorithm with sufficient accuracy to satisfy the use case?<h1 dir="ltr">Syntiant White Paper</h1>Two recent research projects have compared device options for KWS. The 2018, “Hello Edge,” paper&nbsp;[1] performed neural network architecture evaluation and exploration for running KWS on resource-constrained microcontrollers. In 2019, Syntiant tested their NDP10x series of Neural Decision Processors against a range of MCU’s suitable for KWS which they outlined in this whitepaper [2]. The rest of this article will be dedicated to examining the results of the Syntiant test, which provides a good overview of some of the leading options for adding KWS to edge devices.<h1>The line-up</h1>The test compared several of the lowest-power options for KWS available, including their own NDP10x, and compared speed, accuracy, and power on a common task: the Google keyword<a href="https://ai.googleblog.com/2017/08/launching-speech-commands-dataset.html" target="_blank">&nbsp;dataset</a> assembled by Pete Warden of Google.&nbsp;Included in the test were the STM32F746 Discovery<a href="http://www.st.com/en/evaluation-tools/32f746gdiscovery.html" target="_blank">&nbsp;board</a>, the STM32L476G, and its <a href="https://www.st.com/en/evaluation-tools/32l476gdiscovery.html" target="_blank">evaluation board</a>, as well as the Ambiq Apollo 3 Blue, using the <a href="https://learn.sparkfun.com/tutorials/using-sparkfun-edge-board-with-ambiq-apollo3-sdk" target="_blank"></a><a href="https://learn.sparkfun.com/tutorials/using-sparkfun-edge-board-with-ambiq-apollo3-sdk" target="_blank">SparkFun Edge Board</a>, a joint effort by Google, Ambiq, and SparkFun, and the Syntiant’s NDP100 solution using their own internal development board.<img class="fr-fic fr-dib" src="https://images.wevolver.com/eyJidWNrZXQiOiJ3ZXZvbHZlci1wcm9qZWN0LWltYWdlcyIsImtleSI6ImZyb2FsYS8xNTk3MzE4ODc1OTA4LWNyb3BwZWQgc3ludGlhbnQgYm94LmpwZyIsImVkaXRzIjp7InJlc2l6ZSI6eyJ3aWR0aCI6OTUwLCJmaXQiOiJjb3ZlciJ9fX0=" style="width: 788px;"><h1 dir="ltr">The test</h1>The researchers modified the code to run a batch of 20 inference cycles on each of the platforms. For the MCU targets, the input audio was taken from a pre-recorded waveform stored in memory to avoid contaminating the measurements with microphone power. The measured power includes computation of<a href="http://practicalcryptography.com/miscellaneous/machine-learning/guide-mel-frequency-cepstral-coefficients-mfccs/" target="_blank">&nbsp;MFCC</a> or filter-bank energy features, but not an ADC or microphone power. The NDP measurements include decimation of <a href="https://en.wikipedia.org/wiki/Pulse-density_modulation" target="_blank">PDM outpu</a>t from a digital microphone, as well as feature extraction and the neural inference energy.&nbsp;MCUs often have several on and off-chip peripherals powered from the same supply, so the Syntiant team subtracted a baseline current measurement made with the MCU powered up but idle. The NDP results are direct measurements with nothing subtracted. The MCU experiments were performed with HW settings as set in the available open-source software linked above.<h1 dir="ltr">The results</h1>The table below summarizes the measurements. Simple ‘side by side’ comparisons are difficult due to the trade-offs between model complexity, accuracy, compute time, and energy. To make reading the results more clearly, the researchers compared across three axes:- Energy per inference is the energy required to process one frame of audio samples and produce a label (e.g., “yes,” "no,” “hello”), including feature extraction and running inference.- Inference time is the time required to process one frame of audio and frame rate is its inverse, indicating the maximum number of audio frames per second that the processor is capable of processing with a given network.&nbsp;- G10 Accuracy is the accuracy result on the Google-10 keyword task.Estimates for the operations and parameter counts for the Small FC and Small DS-CNN networks are taken from the “Hello Edge” paper. For the TinyConv network used on the Ambiq chip, those figures were estimated from the code in the “Speech Commands” TensorFlow example.<img class="fr-fic fr-dib" src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk3MzA5ODAxMjI3LXRhYmxlMSBhcnRpY2xlLmpwZyIsICJlZGl0cyI6IHsicmVzaXplIjogeyJ3aWR0aCI6IDk1MCwgImZpdCI6ICJjb3ZlciJ9fX0=" style="width: 788px;"><h1 dir="ltr">What does this mean?</h1>It’s clear that the purpose-built compute path in the NDP100 provides dramatic energy savings relative to the stored-program architecture of an MCU solution. The closest result in energy, from the Ambiq Apollo 3 Blue, is 27x higher than the NDP energy, operating with a much smaller and less accurate network.&nbsp;The STM32L4786 running the Small DS-CNN network had a comparable accuracy to the NDP but requires 193x the energy.The frame rates are also critical to examine. Production KWS systems typically require at least 20 and often 100 fps in order to avoid “stepping over” a target utterance. The NDP also provides a speed advantage over low-power MCUs thanks to its parallelism and dedicated data path.The Apollo's energy advantage comes with a speed penalty resulting in a maximum frame rate of 3.7fps, which will typically yield much lower real-world precision than the software accuracy of 66 percent &nbsp;(in TensorFlow on pre-aligned data) would predict.&nbsp;Of the tested MCUs, only the STM32F746, from ST’s high-performance F7xx line, exceeds 20 fps, and it does so at an energy cost about 700x higher than the NDP, resulting in a power cost of about 50 mW (milliwatts), in addition to the baseline power required to power up the MCU initially.<h1 dir="ltr">Conclusion</h1>This test clearly demonstrates that a purpose-built compute engine for neural inference can provide energy and speed advantages over stored-program architectures. The continued development of ultra-low-power MCU’s and highly efficient purpose-built chips will bring voice to the edge in a range of applications that have the potential to disrupt the way we interact with technology.&nbsp;<hr><h1 dir="ltr">AI Voice Applications: Call for Projects</h1>Are you developing devices that include a voice interface or another application of voice technology? We would love to hear from you!&nbsp;Wevolver is looking to connect with exciting relevant projects to profile in an upcoming article about AI empowered voice applications. This is a great opportunity to get exposure among the engineering community!&nbsp;Please leave your details via this <a href="https://forms.gle/3HuBqDiRqX3wEHtK8" rel="noopener noreferrer" target="_blank">short form</a>. &nbsp;We will be in touch shortly.<hr><h1 dir="ltr">About the sponsor: Syntiant</h1><h2>Voice at the edge</h2>Semiconductor company Syntiant is pushing forward voice at the edge possibilities with the development of the Neural Decision Processors™ (NDP10x). These high power r devices offer highly accurate wake word, command word, and event detection in a tiny package with near-zero power consumption.&nbsp;Currently, in production with Amazon’s Alexa Voice Service firmware, the NDP10x family makes Always-On Voice interfaces a reality by supporting entirely new form factors designed to wake up and respond to speech rather than touch. These processors will move voice interfaces from the cloud on to devices of all sizes and types, including ultra-portable wearables and devices. <hr><h1 dir="ltr">References</h1>1. Zhang, Y., Suda, N., Lai, L., &amp; Chandra, V. Hello Edge: Keyword Spotting on Microcontrollers. ArXiv, abs/1711.07128. Feb 2018. Available from:<a href="https://arxiv.org/pdf/1711.07128.pdf" target="_blank">https://arxiv.org/pdf/1711.07128.pdf</a>2. Holleman, J. The Speed and Power Advantage of a Purpose-Built Neural Compute Engine, June 2019. Available from: https://www.syntiant.com/post/keyword-spotting-power-comparison

Ultra-low-power MCU’s and highly efficient purpose-built chips are bringing voice to the edge.

Keyword Spotting in Edge Device

<h1 dir="ltr">Voice in devices</h1>A voice user interface (VUI) &nbsp;allows you to interact with a device, such as a computer, smartphone, or home appliance by simply talking to it. VUIs allow you to ask Siri the weather, request Alexa to play a song, or tell your smart globe to dim the lights without ever touching anything. &nbsp;The history of voice commands goes back to the early 1950s when Bell Laboratories introduced the &ldquo;Audrey&rdquo; system, which could recognize a single voice speaking ten digits aloud. About a decade later, IBM designed <a href="https://www-03.ibm.com/ibm/history/exhibits/specialprod1/specialprod1_7.html" target="_blank">&ldquo;Shoebox,&rdquo;</a> a computer that understood and responded to 16 words in English.Interestingly, one of the first widely available consumer products with voice recognition was a children&rsquo;s toy, the infamous Julie Doll, which could understand eight vocal sounds as commands.<iframe allowfullscreen="" class="fr-draggable" frameborder="0" height="648" src="https://www.youtube.com/embed/ewu_NBUHePU?enablejsapi=1&origin=https://www.wevolver.com" width="864" data-lf-form-tracking-inspected-kn9eq4roawkarlvp="true" data-lf-yt-playback-inspected-kn9eq4roawkarlvp="true" data-lf-vimeo-playback-inspected-kn9eq4roawkarlvp="true" data-gtm-yt-inspected-9="true" id="271065592" title="The 1987 Voice First Doll: Julie by Worlds of Wonder Commercial"></iframe>In 2011, Apple released Siri, opening the way to a range of other digital assistants in smartphones, stand-alone devices, and laptops.&nbsp;The advent of Siri and her voice assistant colleagues happened thanks to cloud computing which allowed the complex processing required to synthesize speech to occur in a remote data center with massive computing power.&nbsp;We are becoming more familiar with operating devices using our voice especially in mostly private settings such as in our cars or houses. VUI is currently <a href="https://www.dbswebsite.com/blog/trends-in-voice-search/" rel="noopener noreferrer" target="_blank">estimated</a> to be in almost 3 billion devices worldwide and is expected to soar to 8 billion by the end of 2035.<h1 dir="ltr">Processing at the edge</h1>Alexa is constantly waiting patiently for your commands. However, she can&rsquo;t be constantly recording you and analyzing every sound in your house to check if it is a command. This would not only cause extreme crowding in the cloud, but it would pose a significant security issue. Instead, Alexa is only listening for &nbsp;&ldquo;wake words,&rdquo; - in her case &ldquo;Alexa,&rdquo; - which then wakes up her full system, activating her full voice recognition power. When you then ask Alexa to give you a weather report, your voice is recorded and sent over the Internet to Amazon&#39;s Alexa Voice Services, which <a href="https://en.wikipedia.org/wiki/Amazon_Alexa" rel="noopener noreferrer" target="_blank">parses</a> the recording into commands. &nbsp;The system then sends the relevant output back to your device and you can plan your outfit accordingly. Alexa responding to your wake command is known as edge processing, which happens on the processor in the device, while the weather request is cloud processing and occurs in a data center connected to your home via the Internet.Not all voice command devices need to be connected to the Internet. Advances in edge processing mean more complex commands can be processed on the device that are free from a cloud connection, enabling faster computation with less latency and more security. Thanks to the development of ultra-low-power chips that can do neural network processing on the device, voice commands can be added to almost any device, including super small earbuds and wearables.&nbsp;<h1 dir="ltr">Why voice is important</h1>&ldquo;Speech is the fundamental means of human communication,&rdquo; <a href="https://mitpress.mit.edu/books/wired-speech" rel="noopener noreferrer" target="_blank">writes</a> Clifford Nass, Stanford researcher and co-author of Wired for Speech. Nass illustrates how humans are wired for speech and verbal communication, and provides insight into how voice offers a fast, intuitive method of accessing technology and navigating through its functions.&nbsp;Voice isn&rsquo;t just a more intuitive method of communicating with technology, it also offers an opportunity for using devices in new ways under new circumstances.Because speech requires less cognitive abilities than a touchscreen, voice-enabled devices allow a user to continue to concentrate on a task and issue commands rather than reallocate precious thinking resources to solve a tactile interface. Workers in highly concentrated environments, such as surgery, factory floors, and labs, could be assisted by voice-enabled robots to pass them tools, read charts, or monitor their health.&nbsp;These qualities can be extended to all users of voice interface devices, making our interaction with technology more enjoyable, more intuitive, and more personal.&nbsp;At last year&rsquo;s CES, &nbsp;a very expensive <a href="https://www.mediapost.com/publications/article/330448/kohlers-new-smart-toilet-earns-ces-sitting-ovatio.html#:~:text=Marketing%20a%20%E2%80%9Csmart%E2%80%9D%20something%20is%20quite%20the%20trend.&text=Numi%202.0%20intelligent%20toilet%20with%20KOHLER%20Konnect%20is%20Kohler's%20most%20advanced%20toilet." rel="noopener noreferrer" target="_blank">toilet</a> that flushed via voice command was exhibited. While the demonstration seemed like a humorous application of technology in the moment, in today&rsquo;s new age of extra sanitary caution it could be lifesaving.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk2MjAwMTM0NDIzLXZvaWNlIHRvaWxldC5qcGciLCAiZWRpdHMiOiB7InJlc2l6ZSI6IHsid2lkdGgiOiA5NTAsICJmaXQiOiAiY292ZXIifX19">Image: Kohler<h1 dir="ltr">New audiences</h1>Voice command offers opportunities to introduce new and existing technologies to audiences in several ways. First, voice command devices can be easier to learn and more intuitive to use, so increasing technological skills across the globe. Additionally, &nbsp;voice control brings technology to people who have been overlooked by touch interfaces, such as people with <a href="https://www.researchgate.net/publication/275406227_Self_Assistive_Technology_for_Disabled_People_-_Voice_Controlled_Wheel_Chair_and_Home_Automation_System" rel="noopener noreferrer" target="_blank">disabilities</a> and the elderly, as well as to audiences who have been left out of cloud anchored devices due to the lack of a reliable Internet connection. Voice command functionality also opens possibilities for people with neural diversity to interact with devices, such as touch tablets, whose logic can cause confusion and frustration. [1]&nbsp;Users with physical limitations, such as reduced arm and finger dexterity, vision impairment, or blindness, may be able to naturally and easily interact with devices using their voice without additional adaptation to meet their needs. While many <a href="https://www.sesame-enable.com/" rel="noopener noreferrer" target="_blank">tools</a> have been built for users with vision impairment and blindness access, devices that are brought to market with integrated voice commands reduce the gap that can open between product launch and ability adoption. It can also reduce the cost of technology that is made especially for less able audiences. These applications can range from asking your Kindle to read you a book to managing the temperature of your home.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk2MTk4NDE1NDgxLWFsZXhhIG9uIGNsYXNzcm9vbS5qcGciLCAiZWRpdHMiOiB7InJlc2l6ZSI6IHsid2lkdGgiOiA5NTAsICJmaXQiOiAiY292ZXIifX19">Voice interface provides new ways to deliver education. Image: Bob WilliamsEdge processing means sophisticated voice command devices can be introduced into devices for use in remote areas where Internet service is limited. This may bring voice command devices to remote communities that could improve education and training. It could also open possibilities for voice command devices, such as search and rescue robots, archeological equipment, and environmental monitoring systems.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk2MTk4MjgwNjYxLWtpZCBsaXN0ZW5pbmcgdm9pY2UgYXNzaXN0LmpwZyIsICJlZGl0cyI6IHsicmVzaXplIjogeyJ3aWR0aCI6IDk1MCwgImZpdCI6ICJjb3ZlciJ9fX0=" style="width: 788px;">Voice command offers alternative ways to interact with common technology.&nbsp;<h1 dir="ltr">New Applications Due to Edge Processing of AI</h1>Advanced edge computing is available in large part due to the speed and power of the latest AI semiconductors used in voice processing. These chips have ultra-low power consumption, which makes them suitable for use in almost any portable device, from Bluetooth earbuds to smartphones to wearable health devices. The next generation of voice command devices will have the advantages of <a href="https://www.syntiant.com/" rel="noopener noreferrer" target="_blank">on-device processing</a>. This proximity to end-users means designers can use embedded computing resources to build behavioral-based messaging, offers, and experiences, without compromising privacy or relying on cloud connectivity. These devices will be:<h2 dir="ltr">Secure</h2>No cloud requirement means voice command devices can be highly secure, operating without a network connection. Voice-recognition has the chance to become the most secure biomarkers we possess. With more AI-enabled devices on the market, voice can become part of dynamic encryption mechanisms that can provide different levels of devices in different locations. &nbsp;<h2 dir="ltr">Fast with low latency</h2>One of the biggest advantages of on edge processing is the shrinking of latency, the measurement of how long it takes for a data packet to travel from its point of origin to its destination. By resolving more processes closer to the source and relaying less data back to the center of the network, AI-enabled edge devices can greatly improve performance and speed.&nbsp;<h2 dir="ltr">Private</h2>AI-powered edge processing allows for a user&#39;s data to be stored on a personal device rather than in a remote data centre. This not only meets increasing consumer demand for more control and ownership over their personal data, but it also makes it much easier to meet emerging legislation requirements such as the EU&rsquo;s General Data Privacy Regulation (GDPR).&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk2MTk4MDA0MjE5LTE3MjEwNV9Wb2x2b19NaWNyb3NvZnRfQmFuZF8yLmpwZyIsICJlZGl0cyI6IHsicmVzaXplIjogeyJ3aWR0aCI6IDk1MCwgImZpdCI6ICJjb3ZlciJ9fX0=">Voice can be added securely to wearables without the need for a cloud connection. Image: Volvo&nbsp;<h1>Conclusion</h1>Voice control devices have the potential to enhance the way we communicate with current technology and create new ways of interacting with our environment.&nbsp;Highly powerful, low-energy, AI-enabled chips mean voice interfaces can now be added to anything, anywhere - with and without the cloud.Recommended reading:&nbsp;<a href="https://www.wevolver.com/article/neural-network-architectures-at-the-edge-modelling-for-energy-efficiency-and-machine-learning-performance" rel="noopener noreferrer" target="_blank">Neural Network Architectures at the Edge: Modelling for Energy Efficiency and Machine Learning Performance</a>As &lsquo;no-touch&rsquo; rules and etiquette become essential for public health, how can voice command bring us closer and keep us safe? How would our attitude and experience of working closely with machines change if we can communicate with our most instinctive method? And what technologies can be introduced to audiences who have been overlooked by the complexity and necessary dexterity of touch interfaces?<hr><h1 dir="ltr">About the sponsor: Syntiant</h1><h2>Voice at the edge.</h2>Semiconductor company&nbsp;<a href="https://www.syntiant.com/" rel="noopener noreferrer" target="_blank">Syntiant</a>&nbsp;is pushing forward voice at the edge possibilities with the development of the Neural Decision Processors&trade; (NDP10x). These high power devices offer highly accurate wake word, command word, and event detection in a tiny package with near-zero power consumption.&nbsp;Currently, in production with Amazon&rsquo;s Alexa Voice Service firmware, the NDP10x family makes Always-On Voice interfaces a reality by supporting entirely new form factors designed to wake up and respond to speech rather than touch. These processors will move voice interfaces from the cloud on to devices of all sizes and types, including ultra-portable wearables and devices.<img class="fr-fic fr-dib" src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNjAwMTczODc0NzY0LXN5bnRpYW50IGxvZ28uanBnIiwgImVkaXRzIjogeyJyZXNpemUiOiB7IndpZHRoIjogOTUwLCAiZml0IjogImNvdmVyIn19fQ==" style="width: 788px;"> <hr><h1 dir="ltr">References</h1>1. Balasuriya S, Sitbon L, Bayor A, Hoogstrate M, Brereton M. Use of voice activated interfaces by people with intellectual disability. Proceedings of the 30th Australian Conference on Computer-Human Interaction. 2018;.&nbsp;2. Pyae A, Joelsson T. Investigating the usability and user experiences of voice user interface. Proceedings of the 20th International Conference on Human-Computer Interaction with Mobile Devices and Services Adjunct - MobileHCI &#39;18. 2018;.

AI-powered edge processing opens possibilities for voice command on any device.

Voice At The Edge Helps Technology Work For More People

The <a href="https://www.bbc.com/news/world-middle-east-49710934" rel="noopener nofollow" target="_blank">drone attack</a> claimed by Yemeni rebels on key Saudi Arabian oil refineries that took place on September 14, 2019 has brought the powerful technology back into the news. Unfortunately, the strikes that disrupted roughly 5% of the world’s oil supply has also contributed more ammunition to the overarching negative connotations the word “drone” conjures.“Drone” is a very broad term. Colloquially, drones are usually thought of as remote-piloted flying devices used by militaries for surveillance and offensive tactics or by civilians for recreational or business purposes. Merriam-Webster defines it as “an unmanned aircraft or ship guided by remote control or onboard computers.” Essentially, though, the term can refer to any machine that is controlled without being in direct contact with a human.While drone machines that are remotely controlled are effectively just tools — such as flying drones used to inspect commercial property roofs thus increasing workplace safety — drones that run themselves require some level of artificial intelligence to guide their actions.There is actually a drone in use by some Japanese companies called T-Frend which is “designed to reduce overtime by flying around the office after hours, playing loud music and taking pictures of any staff who are still working and reporting them to management” (<a href="https://www.forbes.com/sites/tromero/2018/06/07/can-workplace-drones-help-to-combat-japans-overtime-culture/#38c47fd5509c" rel="noopener nofollow" target="_blank">Forbes</a>).Currently, most drones are operated by human pilots. This occurs with military machines like the “Predator”, and by the aforementioned building inspectors, and by marine biologists seeking elusive life forms a mile below the surface of the Pacific ocean. Indeed, a current <a href="https://www.popularmechanics.com/technology/robots/a19139755/ocean-xprize-round-2-finalists/" rel="noopener nofollow" target="_blank">XPrize competition</a> is being held to encourage development of deep sea exploration drones.However, all of our current examples of drone usage do not involve vast scale. With remote-controlled drones, one pilot is usually required per drone unit. This is scalable to a degree, but is completely reliant on the available number of trained pilots.The Uran-9 robotic drone “fire support vehicle” was completed in 2016 and used by Russia and deployed to Syria for a combat test. As reported in <a href="http://defence-blog.com/army/combat-tests-syria-brought-light-deficiencies-russian-unmanned-mini-tank.html" rel="noopener nofollow" target="_blank">Defence Blog</a>, the unmanned ground vehicle performed horribly. It lost contact with its remote operators at distances of only 300+ meters; was unable to move efficiently; demonstrated inferior optics; and was not sufficiently armored enough to last very long without constant access to a repair crew.All of this might be explained by a rushed development and production cycle, which, in turn, can be blamed on what is fast becoming a new kind of cold war as the world’s richest countries try to outdo one another in drone technology.When it comes to flying drones, however, the uses and cost savings are huge:<ul><li data-selectable-paragraph="">Oil companies use them to monitor pipelines for damage</li><li data-selectable-paragraph="">Drones are used to capture footage for films, TV and sporting events that couldn’t be done any other way</li><li data-selectable-paragraph="">Drones deliver donated food, medicine and other supplies to remote territory and people in need during disasters</li><li data-selectable-paragraph="">Wildlife management officials count birds, deer and other animals</li><li data-selectable-paragraph="">Police are replacing helicopters with drones to search for and track suspects and criminals</li><li data-selectable-paragraph="">Surveyors and engineers employ drones to map sites and elevations</li><li data-selectable-paragraph="">Departments of Transportation are beginning to use drones to monitor traffic conditions</li><li data-selectable-paragraph="">Drones are patrolling our borders</li></ul><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1595301765360-miguel-angel-hernandez-okUm2FBNKSQ-unsplash.jpg">Photo by <a href="https://unsplash.com/@miguelherc96?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText" target="_blank">Miguel Ángel Hernández</a> on <a href="https://unsplash.com/?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText" target="_blank">Unsplash</a>&nbsp;And we can’t overlook the impact of a new technology often has on the realm of entertainment and sports: the <a href="https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=20&cad=rja&uact=8&ved=2ahUKEwjAwbacm97lAhVumuAKHSQpA64QFjATegQIBRAH&url=https%3A%2F%2Fwww.youtube.com%2Fchannel%2FUCiVmHW7d57ICmEf9WGIp1CA&usg=AOvVaw11PWzneUJlL3UmDrej7eKH" rel="noopener nofollow" target="_blank">Drone Racing League</a> is growing by leaps and bounds and there will likely be many other variations and extensions of drones-as-entertainment in the near future.Commercial use of drones, though, is currently illegal by FAA regulations. In the future — once drones can be lawfully operated by public and private business entities — we can expect to see:<ul><li data-selectable-paragraph="">Drones delivering packages (via Amazon or the USPS)</li><li data-selectable-paragraph="">Unmanned cargo transport via large aerial drones or ocean vessels</li><li data-selectable-paragraph="">Drones unloading fertilizers and pesticides on crops for commercial farmers</li><li data-selectable-paragraph="">Drones delivering pizza (as one example that is rather low on the priorities list)</li></ul>For many of the above-mentioned applications, the drones are (or will be) remotely piloted. Again, this makes the number of trained pilots into a limiting factor. This, however, is not necessarily a bad thing. The Association for Unmanned Vehicle Systems International (<a href="http://www.auvsi.org/auvsiresources/economicreport" rel="noopener nofollow" target="_blank">AUVSI</a>) states that:<blockquote>“…in the first three years of (commercial unmanned aircraft integration) integration more than 70,000 jobs will be created in the United States with an economic impact of more than $13.6 billion. This benefit will grow through 2025 when we foresee more than 100,000 jobs created and economic impact of $82 billion.”</blockquote>Every year that this integration is delayed almost definitely costs the United States billions in economic growth. Dozens of Fortune 500 companies and hundreds of startups and peripheral business have drone technology queued up and “ready to launch”. Most of the initial planned uses are those requiring remote pilots.But the full potential of drones will only be reached with vast scaling, and training a billion people to be full-time drone pilots won’t be happening anytime soon. Machines controlled by artificial intelligence will absolutely be needed to:<ul><li data-selectable-paragraph="">Operate schools of drones that scour oceans and waterways to clean them</li><li data-selectable-paragraph="">Drive robotic taxis / shared autonomous vehicles</li><li data-selectable-paragraph="">Run complete robo-factories, from assembling of products to shipping them via air and roads</li><li data-selectable-paragraph="">Explore the galaxy beyond the reach of timely communications from Earth in swarms of small space probes</li></ul><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1595301816676-dose-media-DiTiYQx0mh4-unsplash.jpg">Photo by&nbsp;<a href="https://unsplash.com/@dose?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText" target="_blank">Dose Media</a> on&nbsp;<a href="https://unsplash.com/s/photos/ai-drones?utm_source=unsplash&utm_medium=referral&utm_content=creditCopyText" target="_blank">Unsplash</a>&nbsp;Many of these uses will disrupt entire industries, such as autonomous vehicles and the commuter and trucking sectors of transportation. Our world will be transformed rather rapidly, between now and 2050, into an Earth that actually begins to resemble many of the science fictions imagined in the late 20th century.There will certainly be a number of unintended consequences. Consider the wide adoption of other revolutionary technologies such as smartphones, nuclear power plants and automobiles.There are easy-to-see “first level” effects: Roughly <a href="https://www.asirt.org/safe-travel/road-safety-facts/" rel="noopener nofollow" target="_blank">1.25 million people</a> die in car-related deaths every year. Smartphone usage has irreversibly changed the psychology of generations born after 2000. Nuclear accidents like the Chernobyl and Fukushima disasters have leaked deadly and mutating radiation into their local biomes for decades. And then there are secondary effects that only become clear after many years, including the continual erosion of extended families as the auto lets people easily move great distances from their places of birth.We already know some of the negatives that widespread drone adoption will facilitate, from easier spying to criminal and terrorist uses. However, the hope is that the positive changes will outweigh those negatives. Despite how terrible some dark sides of previous new technologies have proven to be, it does seem that, overall, things like the automobile, jet planes, genetic modification, and nuclear power have helped humanity prosper to a greater degree than they have harmed us. However, harm to the environment (including burning of fossil fuels) by some of these technologies has definitely had a deleterious effect on the planet as a whole.But drones stand to possibly be one of the better new technologies. They combine electric power with efficient use of transportation ability, and should ultimately help reduce CO2 emissions and improve many services.AI drones will be responsible for giving many of us the future we’ve only ever dreamed of.

As we replace and enhance human effort with that of machines, how much will society change?

Drones, Artificial Intelligence, and the Future

If you are reading this, you've perhaps come across these discussions on the internet: "Is AI going to take away your jobs?", "Will AI replace humans by becoming better than them in every task?", and also some exaggerated, melodramatic versions like: "AI armageddon!!" and "Evil Robots will take over!", etc. To add fuel to fire, Hollywood paints a vivacious picture of just that. I haven't yet come across a robot/AI movie that ends well, have you? All this said and done, have you ever wondered where we are in that path? Is everything Hollywood saying true? Will I indeed lose my job? Will robots take over the world? Whoaaa... It's ok, take a breather! You're going to be fine! The world will be fine!In this article, I will give you my take on where we are possibly headed and what could be a plausible reality we can prepare for. We will also look at the contribution of THINC Lab, where the best minds come together to create AI solutions to enrich and enhance human life. But first, let's just pause and mull over the remarkable progress mankind has made since we stepped on the planet... From not knowing how to light a fire, we've created rockets that blast off into space using fire, from not knowing how to verbally communicate with each other, we've sent out voyager satellites hoping to communicate with other intelligent beings out there! and throughout this incredible journey, mankind has gone through a phenomenal evolution, but the one thing we've always been very good at is survival! Everything we do, we do to survive and grow, emotionally, socially, culturally, and ofcourse, intellectually. So why are we creating something that might threaten our existence one day? The short answer, WE'RE NOT! At least that's not the intent!&nbsp;I personally feel there's a lot of miscommunication or just a lack of communication between the actual architects who work on building these systems and the stakeholders/consumers who are on the receiving end of it. You might've heard of people referring to AI as a double-edged sword, but aren't we humans building the swords in the first place? Couldn't we alter its design with time as needed? Absolutely. Everytime a new scientific invention is made, people fear the loss of jobs and a detrimental shift in paradigm. It doesn't have to be that way as long as the Engineers and Scientists building it are cognizant of the journey we are on. But then people point me to the industrial revolution that caused a massive loss of jobs and turned people's lives topsy turvy, I say look at the world now, people adapted, developed and switched to better domains that demanded the innate humanness. Change is always messy and it takes time for the chaos to reach a steady-state, but that fear of change can't hamper our curiosity and imagination, for like Einstein said, "Knowledge is limited, but imagination encircles the world".But hey, I still haven't told you where we are on this journey. Although there's some debate about it, the common consensus is that AI began sometime in the early 1950s. From the time of Alan Turning (also called the Father of AI, and there is an award attributed to his name - you guessed it, the Turing award), we have made some significant strides in the field of AI and with Quantum Computers coming up, the power of AI seems limitless. We live in the age of Neural Networks, where a bunch of neurons (yes, that is an analogy of the human brain adapted in the NN architecture) crunch data using the fastest computers and learn "patterns". Using these "patterns" or optimized objective function model(s), they can make predictions about the future or recognize other objects that belong to the class of data they were trained on. Some eerily advanced applications include <a href="https://en.wikipedia.org/wiki/Deepfake" rel="noopener noreferrer" target="_blank">Deep Fakes</a>, <a href="https://machinelearningmastery.com/impressive-applications-of-generative-adversarial-networks/" rel="noopener noreferrer" target="_blank">Generative Networks</a>, etc. Also in the field of Robotics, we've handsomely advanced to create robots that can <a href="https://blog.aboutamazon.com/transportation/whats-next-for-amazon-scout" rel="noopener noreferrer" target="_blank">deliver packages</a> to your house, <a href="https://www.tesla.com/models" rel="noopener noreferrer" target="_blank">cars</a> that can drive themselves and we're also in the process of building <a href="https://www.bostondynamics.com/" rel="noopener noreferrer" target="_blank">military robots</a>. So, if AI and robots can do all this, what promise do we have that they will not outperform humans and take over jobs? Honestly, we don't. It would be foolish to think that some jobs would not be displaced/restructured by AI and Robotics.&nbsp;But since we can see it coming now, we can prepare for it, adapt, and make a smooth transition into this new era. But for the most part, the intent of AI/Robotics is to work alongside humans to aid and improve human life. One such example is the project our researchers at THINC Lab are involved in. At THINC, we strive to create cutting-edge AI solutions that can be seamlessly integrated into industries and workplaces. A real-life example of such a solution is the one we are currently developing with the funding provided by the National Science Foundation to create intelligent robots that work alongside humans in an onion sorting plant to sort onions quickly and efficiently.&nbsp;<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk1MTc1MjIwNDUwLUlNR18yMDE5MDQyMl8xNzU0NTEwNjIuanBnIiwgImVkaXRzIjogeyJyZXNpemUiOiB7IndpZHRoIjogOTUwLCAiZml0IjogImNvdmVyIn19fQ==">Director of THINC Lab Dr.Doshi examining the robot during setup.<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk1MTc1NjQyOTA2LXNhd3llci5KUEciLCAiZWRpdHMiOiB7InJlc2l6ZSI6IHsid2lkdGgiOiA5NTAsICJmaXQiOiAiY292ZXIifX19">What the sorting looks like within a computer simulation.<img src="https://images.wevolver.com/eyJidWNrZXQiOiAid2V2b2x2ZXItcHJvamVjdC1pbWFnZXMiLCAia2V5IjogImZyb2FsYS8xNTk1MTc1ODkwNzM4LVNjcmVlbnNob3QgZnJvbSAyMDIwLTA1LTI4IDE4LTM1LTMyLnBuZyIsICJlZGl0cyI6IHsicmVzaXplIjogeyJ3aWR0aCI6IDk1MCwgImZpdCI6ICJjb3ZlciJ9fX0=">Image showing how the robot perceives the world around it. You may say, "Hey, how hard can that be?", but think about this, humans possess generations of evolutionary intelligence combined with a unique set of sensory apparatus that help us feel, see and understand if an onion is good/bad for consumption. However, all a Robot sees is a mathematical representation of the world around it, and teaching it to do this task involves significant challenges, especially when you expect it to collaborate safely with humans. We are currently developing a Neural Network - <a href="https://arxiv.org/abs/1905.04380" rel="noopener noreferrer" target="_blank">SA-Net</a> for seeing and interpreting the visual information and combining it with our unique in-house recipe of a classical AI algorithm that would tell the robot what to do and how to do it. If you are interested to read more about what we do, check out our <a href="http://thinc.cs.uga.edu" rel="noopener noreferrer" target="_blank">website</a>. Thanks for taking this trip on my train of thought!

And how are AI scientists involved in this?

How far along are we in the AI-Human "race"?

We usually think of <a href="https://woktossrobot.com/2020/05/31/note-on-artificial-general-intelligence/" rel="noopener" target="_blank">intelligence</a> as a quality that individuals possess. However, many of our largest and most complex problems are often solved by groups of individuals. If we think of intelligence as an ability to solve problems, groups can also act intelligently. We call this intelligence ‘<a href="https://woktossrobot.com/2020/05/12/systems-superminds-and-collective-intelligence/" rel="noopener" target="_blank">Collective intelligence</a>’ — the result of a group of individuals acting together in ways that seem intelligent. Collective intelligence is something that makes systems effective, productive, adaptable, or even resilient.<blockquote>‘Collective intelligence’ — the result of a group of individuals acting together in ways that seem intelligent.</blockquote>Although the individuals in the system might not have the same goals, most systems do work towards achieving a particular outcome. A company’s goal could be increasing its profits or market cap while a government’s goal could be improving the well-being of its citizens. This desired outcome can simply be described as the impact of the system. Increasing collective intelligence can help increase the efficiency, productivity, or output of the system, and thereby its impact.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1593690578558-1_dAlwXfVrtYspdCzmC8iADw.png">Jobs in AI&nbsp;The collective intelligence of a system can be increased through three different ways:<h2 data-selectable-paragraph="">1. Addition</h2><ul><li data-selectable-paragraph="">Adding more humans into the system. Eg. Hiring more</li><li data-selectable-paragraph="">Adding more machines into the system. Eg. Purchasing more equipment</li></ul><h2 data-selectable-paragraph="">2. Improvement</h2><ul><li data-selectable-paragraph="">Improving the machines. Eg. Hardware or software upgrades</li><li data-selectable-paragraph="">Improving humans in the system. Eg. Learning a new skill, reading a book</li></ul><h2 data-selectable-paragraph="">3. Connection</h2>HUMAN-MACHINE COLLABORATION<ul><li data-selectable-paragraph="">Making machines help humans. E.g. Spellcheck</li><li data-selectable-paragraph="">Making humans help machines. E.g. Writing software programs</li></ul>OTHER<ul><li data-selectable-paragraph="">Making humans help humans. E.g restructuring organisations</li><li data-selectable-paragraph="">Making machines help machines. E.g Software that automates tasks for the program</li></ul><img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1593690647422-0_nJ74TtEps58KUpt2.png">Source: <a href="https://woktossrobot.com/" rel="noopener" target="_blank">https://woktossrobot.com/</a>&nbsp;<h1 data-selectable-paragraph="">Using AI to increase the intelligence of groups</h1>When we decide to introduce technologies like Artificial Intelligence into the system, we are inherently trying to increase the system’s collective intelligence. AI increases the collective intelligence of the system by:<ol><li data-selectable-paragraph="">Improvements in the machines themselves</li><li data-selectable-paragraph="">Making machines help other machines</li><li data-selectable-paragraph="">Improving human-machine collaboration</li></ol><h2 data-selectable-paragraph="">AI improving machines</h2>Imagine a machine that boils eggs. You put in some eggs, set the timer and you get boiled eggs. Pretty soon you run into some obvious problems with this machine. It only boils chicken eggs well. What about duck eggs? What about Quail? Let’s make a smarter egg boiler. It uses an AI algorithm to analyze different types of eggs, temperature, humidity and whole host of other parameters to boil any kind of egg perfectly. Instead of creating a new product altogether, they simply build an attachment for the original egg boiler that converts it into a smart boiler. In a similar fashion, a software update to an electric boiler can make it more efficient, humane and appear smarter. This is an example of AI improving existing machines. This can happen either through improvements in the hardware or the software.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1593690679728-0_-6mYTn37eMQQ-RG6.png">Source:&nbsp;<a href="https://woktossrobot.com/" rel="noopener" target="_blank">https://woktossrobot.com/</a>&nbsp;<h2 data-selectable-paragraph="">AI making machines help other machines</h2>There is a factory that produces pizza. The factory consists of two main machines: one that makes the base and one that makes the topping. However these machines do not talk to each other. There is a human operator that moves the pizza bases to the the topping machine where the pizza gets built. The company makes some simple software updates and some hardware changes to make the transition between these machines seamless and without the need of a human operator. This is very effective but not very smart. Some engineers tweak the software of these machines and plug it with an AI based sentiment detector. The system predicts the demand for pizza in the locality where the company has customers as well as identifies more areas for supply. It discovers new products like plain pizza base or topping sauce and makes the machines collaborate together to meet demand. Such a system would greatly increase the efficiency of the system. This is an example of making machines help other machines. The AI in this case acts like a manager of the machines.<h2 data-selectable-paragraph="">Cobots: Improving human-machine collaboration</h2>Often humans and machines are seen as adversaries fighting for each other’s <a href="https://woktossrobot.com/2020/05/31/ai-jobs/" rel="noopener" target="_blank">jobs</a>. This view however neglects the powerful opportunities in collaboration between the two sides. Now as our tools become more interconnected and intelligent, machines and humans start looking like <a href="https://woktossrobot.com/2020/05/16/a-cognitive-division-of-labor/" rel="noopener" target="_blank">symbiotic partners</a> in the system. This symbiotic relationship can enable <a href="https://woktossrobot.com/2020/05/31/ai-strategy-checklist/" rel="noopener" target="_blank">new business models</a> and <a href="https://woktossrobot.com/2020/05/31/ai-jobs/" rel="noopener" target="_blank">create jobs</a> that never existed. And yes it will impact jobs too. Imagine a flexible, nimbler robot that works alongside humans to help perform generic, repetitive tasks. Or an algorithm that produces a variety of design options based on the designers preferences and goals. The designer can simply select the most optimum option and start using it as a reference or straight up modify it. This idea is sometimes also referred as ‘Organic AI,’ where machines are flexible and easily partner with humans to perform tasks. Such teams can easily adapt to new data and novel market conditions. The machine can become an extension of the human mind and body.<img src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1593690717576-0_DSMQoY8lab0eBy6P.png" style="width: 788px;" class="fr-fic fr-dib">Although improving machines and helping them talk to each other sound like great opportunities, so far we do not have human systems that are fully automated and operate without any person overseeing them. The opportunities that will truly increase the impact of systems lies somewhere in the middle, where&nbsp;<a href="https://woktossrobot.com/2020/05/31/organising-human-ai-relationships/" rel="noopener" target="_blank">humans and AI collaborate</a> and push each other to achieve higher levels of performance.

We usually think of intelligence as a quality that individuals possess. However, many of our largest and most complex problems are often solved by groups of individuals. If we think of intelligence as an ability to solve problems, groups can also act intelligently.

Artificial Collective Intelligence

Discussions surrounding the latest robotic technology tend to focus primarily on the prospect of creating fully autonomous robots. Despite significant advancements in artificial intelligence and machine learning, autonomous robots are still not quite ready to handle the “edge cases” that the world presents—particularly in unstructured and unpredictable environments like the outdoors.Keeping humans in the loop is critical to any successful robotics implementation. Despite what science fiction movies portray, humans are still very much needed for decision-making, error handling and error recovery in unstructured environments. This is especially true among applications that rely heavily on human intellect or physical capabilities, such as those in the health care, aerospace, or energy industries.As the robots of today move beyond the factory floor and into the outside world, the spectrum of robotic technology has expanded. On one end of the spectrum, teleoperation has made it possible for operators to be in complete control while controlling robots from a safe distance. With teleoperation, a person is responsible for 100 percent of the decision-making and the direct actions of the robot through the use of a remote controller.On the other end of the spectrum, there are fully autonomous robots, which keep humans completely out of the loop. Fully autonomous robots are responsible for all decision-making and actions, as well as all error handling and recovery. They must adapt to varying environments and recover from errors entirely on their own. Even though popular culture wants us to believe that AI-driven robots are now primed to take over all of humanity, it will be a long time—perhaps several decades—before we see fully autonomous, outdoor mobile robotic systems, simply because robots cannot yet fully think, act or respond to adaptions on their own, especially in unstructured or changing environments.<img class="fr-fic fr-dib" src="https://s3-us-west-1.amazonaws.com/wevolver-project-images/froala%2F1592916512228-TheAutonomySpectrum+%281%29.png" style="width: 464px;">In between these two extremes — and readily available for implementation today — is “supervised autonomy.” This type of autonomy has two distinct levels: in the first level, a person is making the decisions and the robot is responsible for the direct action. In the second level, the robot is responsible for the high-level decision-making and the direct actions, and operator remains in the loop solely for error handling and recovery.Supervised autonomy can allow robots to perform duties that are otherwise hazardous to humans, while still giving the robot operators to full control over specific tasks. This technology also can serve as a force multiplier when there are labor shortages. In addition, because this technology transfers physically demanding tasks to the robot, older workers can stay on the job longer while using their expertise to optimize control of the robots.Unlike full autonomy, supervised autonomy is ready now for environments that are both controlled (not open to the public) and populated only by trained personnel.For instance, industries such as aviation, construction, energy, and oil and gas, which all share common requirements for trained personnel, controlled environments, resilience to external factors, and the cognitive ability to learn and adapt, can especially benefit from supervised autonomy.Although this technology has the potential to <a href="https://www.resquared.com/industries" rel="noopener noreferrer" target="_blank">benefit many industries</a>, the energy industry could find it particularly useful. The maintenance of high-voltage power lines is, without a doubt, one of the most hazardous manual labor jobs across any industry. Using robots to conduct dangerous inspection or repair tasks could protect workers from serious or even fatal injuries from falls, burns and electrocutions. At first, robots could be teleoperated by experienced operators. Once the system has learned from the experiences and data collected during teleoperation tasks, it could transition to a supervised autonomous system. Since workers would only be involved in a supervisory capacity to assist with error handling and recovery, they would be kept completely out of harm’s way.Overall, the time is right for businesses to consider using supervised robotic autonomy to augment their staff and operations. Whether on the factory floor, in the operating room, or 40 feet in the air on a power line, supervised autonomy can help to keep workers out of harm’s way while improving productivity.<hr>Author Bio:&nbsp; <a href="https://www.resquared.com/jorgenpedersen" target="_blank">Jorgen Pedersen</a>, President and CEO, founded RE2 Robotics in July of 2001. Jorgen has contributed to the growth of the unmanned systems market for the past 20 years, beginning his career at Carnegie Mellon's National Robotics Engineering Center (NREC). Jorgen has held several leadership positions within the Robotics industry. He served as the Chairman of the Robotics Division of the National Defense Industrial Association (NDIA). Jorgen has also served in key roles such as a member of the Board of Trustees for NDIA and the Board of Directors for the National Advanced Mobility Consortium. Jorgen is the 2016 Carnegie Science Start-up Entrepreneur of the Year recipient. Jorgen was presented with the 2008 Army SBIR Achievement Award for the development of the Small Robot Toolkit for unmanned ground vehicles and the 2011 Tibbetts Award for SBIR Excellence. Jorgen holds a M.S. degree in Robotics and a B.S. degree in Electrical and Computer Engineering from Carnegie Mellon University. Connect with Jorgen on <a href="https://www.linkedin.com/in/jorgenre2/" target="_blank">LinkedIn</a> &amp; <a href="https://twitter.com/JorgenRE2" target="_blank">Twitter</a>.

 In this article, CEO Jorgen Pedersen discusses how supervised autonomy can allow robots to perform duties that are otherwise hazardous to humans, while still giving the robot operators full control over specific tasks. 

artificial intelligence

Latest Posts