Category Archives: Ai

Moritz Stefaner has long been obsessed with food. As a designer, he has even used meals to visualize data about everything from ethnic diversity to scientific funding. So when he encountered an AI that generates realistic pictures from words, he went hog wild.

[Image: Midjourney]Stefaner typed words like Michelin star chef, deconstructed, and amuse-gueule into the generator, hoping to evoke the intricate plating of fine dining establishments like Eleven Madison Park or Alinea. And, suffice to say, his plan worked. The images he created are completely convincing plates that you could imagine being served at any 3-Michelin-starred restaurant. That is, until you look a bit closer and you realize that the individual components on the dish often dont even exist in real life.

[Image: Midjourney]Squinting, I see the bones of a soft-boiled egg, seaweed, microgreens, sauces, and gels. In one frame, I swear I see the same candied moss I actually ate at the Chicago fine dining establishment Elizabeth. In another, I see a chocolate dessert sitting on a mound of coffee groundsa stones throw from a dish I ate at Dominique Ansel Kitchen. But for the most part, its superbly convincing fictionand what happens when an AIs style vastly outpaces its substance.

[Image: Midjourney]The most surprising thing to me was how well the system deals with really poetic descriptions, writes Stefaner via email. It goes way beyond just capturing objects in certain styles, towards capturing a whole vibe.

[Image: Midjourney]The AI was built by Midjourney, a self-ascribed research lab focused on expanding the imaginative powers of the human species. Much like GLIDE, Disco Diffusion, and Dall-E, Midjourneys AI model generates images from words. But Midjourney makes the process particularly simple. You dont have to understand code or set up anything special to use this AI. Instead, it is hosted on a private Discord, so creating an image is literally as easy as typing it.

[Image: Midjourney]The input to these networks are short texts called prompts. The art of the prompt is really becoming a key skill in interacting with these models. Similar to learning how to nudge a search engine to surface the right results, the prompt artist learns to use the right combination of words to achieve the desired effects, Stefaner says. For instance, one can add drawn by Picasso or in the style of Keith Haring to evoke style modifiers that mimic an artists style.

Stefaner focused on words like fine dining, and stylistically he tagged many of his prompts with dof, which stands for depth of field, and refers to the partially in-focus images that appear to be shot by a traditional camera and are a hallmark of fine dining photography.

No doubt, the way these images are framedthe angles at which the camera seems to be taking themhelps them seem convincing. Thats key because most of these foods are at least a little alien; they are almost-foods, if you will. Pasta dishes look more like thinly sliced banana peels. Fish looks like salmon, if its skin were marbled into its flesh. These oddities arent always as gross as they might sound. One anonymous fine dining plate looks like a cross between prawn shells and flowers. Its downright beautiful, and just the sort of meticulous surprise you hope to encounter when dropping hundreds of dollars on a tasting menu.

I wish I could say the same about what appears to be sea scallops. Seared on top, they melt like a Salvador Dal painting into the plate. Are they made of ice cream? Might they be a foam? My brain tries to make sense of it all until I remember, theres no sense to be made. Im looking at an AI hallucination.

[Image: Midjourney]Its almost like an alien life form observed us and tried to imitate and blend in the best it could, without really understanding what is going on, Stefaner says. This strangely familiar unfamiliar feeling is a bit unsettling, but can also really trigger creativity. We are pattern-seeking animals, always searching for meaning, so we really try to figure out what these dishes could be, what they could taste like, even though they dont quite make sense to us.

[Image: Midjourney]These hallucinations, of course, are trained into the software, which was fed countless labeled images to understand how to draw the objects. And there is no better window into the AIs superficial logic than in Mortizs fine dining high end Michelin star closeup burgeron top, its a pile of rare ground beef, capped with a shiny brioche bun. But on the bottom, where the other bun should be? Thats a coral-like pile of something vaguely edible. In other words, the burger starts at Red Robin and ends with Noma. And while its funny, this image also demonstrates how little these AI models understand about the content they generate.

[Image: Midjourney]In any case, Stefaners images are captivating to behold. They also push us to ask, Whats next? Thus far, weve seen art imitate life. But next, we might see life imitate art.

Id love to do creative sessions with ambitious chefs to generate inspiring images, based on new prompts (or their existing menus!) and then see if we can together reverse-engineer them into successful dishes, Stefaner says. Its a new type of agent you can inject in your design process to generate completely new, oblique ideas.

Read this article:

These 3-Michelin-starred plates were invented by AI. The food doesn't even exist - Fast Company

Lilt, a provider of AI-powered business translation software, today announced that it raised $55 million in a Series C round led by Four Rivers, joined by new investors Sorenson Capital, CLEAR Ventures and Wipro Ventures. The company says that it plans to use the capital to expand its R&D efforts as well as its customer footprint and engineering teams.

Lilt [aims to] build a solution that [will] combine the best of human ingenuity with machine efficiency, CEO Spence Green told TechCrunch via email. This new funding will [reduce our] unit economics [to make] translation more affordable for all businesses. It will also [enable us to add] a sales team to our existing production team in Asia. We are in three regions the U.S., Europe, the Middle East and Africa (EMEA) and Asia and look to have both sales and production teams in each of these regions.

San Francisco, Calfornia-based Lilt was co-founded by Green and John DeNero in 2015. Green is a former Northrop Grumman software engineer who later worked as a research intern on the Google Translate team, developing an AI language system for improving English-to-Arabic translations. DeNero was previously a senior research scientist at Google, mostly on the Google Translate side, and a teaching professor at the University of California, Berkeley.

15 years ago, I was living in the Middle East, where you make less money if you speak anything other than English. I have never been exposed to that kind of disparity before and the disadvantage was extremely frustrating, Green told TechCrunch. I then returned to the States, went to grad school and started working on Google Translate, where I met [DeNero]. Our mission is to make the worlds information accessible to everyone regardless of where they were born or which language they speak.

To translate marketing, support and e-commerce documents and webpages Lilts principal workload Lilt uses a combination of human translators and tools including hotkeys, style guides and an AI translation engine. Green says that the platform supports around 40 languages and offers custom term bases and lexicons, which show translators a range of possible translations for a given word.

The aforementioned AI engine, meanwhile which is regularly trained on fresh data, including feedback from Lilts translators analyzes translation data to make recommendations. But the translators have the final say.

AI and machine learning are helping automating the process around enterprise translation, but you cant automate it all thats why we are a human-in-the-loop process, Green said. We are leaving creative and emotional elements of translation to humans while automating the tedious and repetitive elements. This helps with the unit economics of our business and allows for businesses to apply translation across all customer touch points.

Using Lilts platform, customers can assign translators and reviewers, track due dates and keep tabs on ongoing translation job progress. After signing an annual contract with Lilt, customers can use the services API and connectors to funnel text for translation from platforms including Slack, WordPress, GitHub, Salesforce, Zendesk and Adobe Marketing Cloud.

Translators are paid a fixed hourly rate, negotiated individually. They must earn at least $20 for rendering linguistic services through the platform to cash out, which can include review as well as translation. Lilt tracks hours automatically, counting only time spent actively translating and reviewing content and not time spent conducting external research beyond Lilts standard work limits (30 seconds for translation per segment and 50 seconds for review per segment).

According to a recent Salesforce survey, the average consumer now uses ten channels including social media and SMS to communicate with businesses. Yet the average company supports relatively few languages, with a July 2020 study from Stripe finding that 74% of European e-commerce websites hadnt translated their checkout pages into local languages.

This is where Green sees opportunity, despite competition from rivals like Unbabel. Grand View Research anticipatesthe machine translation market will be worth $983.3 million by 2022.

Already, Lilt which has a workforce of over 150 people claims to have customers in Intel, Emerson, Juniper Networks and Orca Security and others across the education, crypto, technology, defense and intelligence sectors. Its only becoming more vital to ensure an end-to-end multilingual customer experience, Green added.

Existing investor Sequoia Capital, Intel Capital, Redpoint Ventures and XSeed Capital also participated in Lilts Series C round. It brings the startups total capital raised to $92.5 million.

More here:

Lilt raises $55M to bolster its business-focused AI translation platform - TechCrunch

Listen to this article

Recently out of stealth mode, Waabi is building the next generation of self-driving truck technology. Auto Futures caught up with Raquel Urtasun, Founder and CEO of Waabi, at NVIDIAs GTC conference. She reveals how Waabi World and Waabi Driver are set to accelerate self-driving safety and commercialisation.

Urtasun is best known in the mobility space as Chief Scientist and Head of R&D at Uber ATG before it was sold to Aurora. She also serves as a Full Professor in the Department of Computer Science at the University of Toronto. She is a co-founder of the Vector Institute for AI. She founded Waabi in June 2021.

The Mayor of Toronto, John Tory, called her an international star and an extraordinary talent. She was named one of The Top 25 Women of Influence for 2022.

Urtasun defined the company and its mission. Waabi is an AI company where we are building the next generation of self-driving technology for trucking focusing on an L4 (Level Four) hub-to-hub autonomous solution. Which basically means that the majority of our driving is going to be on highways.

This is deliberate because driving on highways is simpler than driving in cities. And, at the same time, Waabi can automate a large variety of operational domains while having similar capabilities, she says.

From the point of view of business, it makes a lot of sense because theres a chronic shortage of drivers. And with the pandemic, it is getting worse. People do not want to be truck drivers because they have to be away from their families.

Truck driving is one of the most dangerous jobs in North America. Automation can have a significant impact on moving cargo that is having a significant impact on all of us, reports Urtasun.

We have a very collaborative approach. In particular, we are collaborating with OEM partners to integrate our solution into their redundant truck platforms. Fleets and carriers are our prime customers. With our solution, we would basically increase the operation efficiency, safety and cost of moving their freight, she says.

We are a company where we are very focused on the commercialisation of our products. We are not focused on building demos we are building a real product.

Urtasun notes that self-driving commercialisation or deployment is being done in limited domains. She explains why.

The process of driving is actually really hard. People underestimate the difficulty of this problem. If you think about the decisions that you make as youre driving theyre actually pretty complex and nuanced because a small change in the environment might totally change the manoeuvre that you should do. And this is difficult to do. On top of this, there are potentially many situations that might arise that the vehicle needs to handle, she explains.

She points out that, even worse is the fact that many of these situations happen rarely. You might see them only once in millions of miles.

Urtasun says that todays simulation technology cannot test and verify at scale. It is capital intensive. A new approach is needed that is less expensive.

Waabi has created a new paradigm that combines the existing approaches to self-driving with the Waabi integrated AI-first autonomy stack that features high-fidelity simulation. It is scalable and more affordable than current self-driving approaches, she says.

Urtasun showed a promotional video that spells out how Waabi Driver and Waabi World work.

It states it could take thousands of self-driving vehicles driving for millions of miles for thousands of years to experience everything necessary to drive safely. Some things happen very rarely. Waabi Worlds high-fidelity driving simulator is the ultimate school for self-driving vehicles.

Waabi World reconstructs from real-life sensor data. It can reconstruct objects such as cars, SUVs or trucks. It can digitally recreate reality to create an endless number of diverse virtual worlds. It can be done across different sensor configurations as though the driver were in a car or truck, automatically and at scale.

In Waabi World, the Waabi Driver can see and behave exactly as it would in the real world. Waabi World then can create traffic scenarios to test the Waabi Driver. Waabi World generates variations of streets and traffic patterns while the Waabi Driver reacts in real-time, then traffic reacts. Waabi World can multiply and evolve scenarios infinitely.

It can evaluate how the driver performs in simulation and use AI to automatically generate challenging and realistic scenarios.

Waabi World does not just test the driver to its limits but also helps it learn new skills. Ultimately, the Waabi Driver will learn on its own to drive safely in any vehicle in any scenario anywhere in the world.

At the end of the day, what you can do with this type of approach is develop this technology at a fraction of the cost at a much faster speed, says Urtasun.

Some think that Waabi World is a competitor to simulation tools such as IPG Carmaker, Cognata, Tass Prescan and Applied Intuition. Urtasun reveals how the company will use its Waabi World simulation.

Our simulation is an internal product and not something that we will license to other companies to use. Our product is the Waabi Driver, which is our solution to L4 trucking (no human driver). Our simulation technology is very different from those companies. Think about it as a next-generation simulation. It is built to scale from day one. It is immersive, reactive and super high fidelity. The domain gap is nearly zero, which is very different from current simulators.

Waabi teams work in Toronto and San Francisco. Its self-driving vehicles have not been seen on roads yet.

What is in the future for Waabi? When will the Waabi Driver drive on public roads? asks Auto Futures.

You will see Waabi testing on roads very soon. We have very exciting things to show. We are building an L4 solution for trucking, meaning that we can choose where and when to operate. So we dont need to solve the super hard snow storm problem from the first day, replies Urtasun.

The rest is here:

How Waabi AI is Fuelling the Next Generation of Self-Driving Trucks: CEO Raquel Urtasun - Auto Futures

There is widespread agreement that responsible artificial intelligence requires principles such as fairness, transparency, privacy, human safety, and explainability. Nearly all ethicists and tech policy advocates stress these factors and push for algorithms that are fair, transparent, safe, and understandable.1

But it is not always clear how to operationalize these broad principles or how to handle situations where there are conflicts between competing goals.2 It is not easy to move from the abstract to the concrete in developing algorithms and sometimes a focus on one goal comes at the detriment of alternative objectives.3

In the criminal justice area, for example, Richard Berk and colleagues argue that there are many kinds of fairness and it is impossible to maximize accuracy and fairness at the same time, and impossible simultaneously to satisfy all kinds of fairness.4 While sobering, that assessment likely is on the mark and therefore must be part of our thinking on ways to resolve these tensions.

Algorithms also can be problematic because they are sensitive to small data shifts. Ke Yang and colleagues note this reality and say designers need to be careful in system development. Worrying, they point out that small changes in the input data or in the ranking methodology may lead to drastic changes in the output, making the result uninformative and easy to manipulate.5

Algorithms also can be problematic because they are sensitive to small data shifts.

In addition, it is hard to improve transparency with digital tools that are inherently complex. Even though the European Union has sought to promote AI transparency, researchers have found limited gains in consumer understanding of algorithms or the factors that guide AI decisionmaking. Even as AI becomes ubiquitous, it remains an indecipherable black box for most individuals.6

In this paper, I discuss ways to operationalize responsible AI in the federal government. I argue there are six steps to responsible implementation:

There need to be codes of conduct that outline major ethical standards, values, and principles. Some principles cut across federal agencies and are common to each one. This includes ideas such as protecting fairness, transparency, privacy, and human safety. Regardless of what a government agency does, it needs to assure that its algorithms are unbiased, transparent, safe, and capable of maintaining the confidentiality of personal records.7

But other parts of codes need to be tailored to particular agency missions and activities. In the domestic area, for example, agencies that work on education and health care must be especially sensitive to the confidentiality of records. There are existing laws and rights that must be upheld and algorithms cannot violate current privacy standards or analyze information in ways that generate unfair or intrusive results.8

In the defense area, agencies have to consider questions related to the conduct of war, how automated technologies are deployed in the field, ways to integrate intelligence analytics into mission performance, and mechanisms for keeping humans in the decisionmaking loop. With facial recognition software, remote sensors, and autonomous weapons systems, there have to be guardrails regarding acceptable versus unacceptable uses.

As an illustration of how this can happen, many countries came together in the 20th century and negotiated agreements outlawing the use of chemical and biological weapons, and the first use of nuclear weapons. There were treaties and agreements that mandated third-party inspections and transparency regarding the number and type of weapons. Even at a time when weapons of mass destruction were pointed at enemies, adversarial countries talked to one another, worked out agreements, and negotiated differences for the safety of humanity.

As the globe moves towards greater and more sophisticated technological innovation, both domestically and in terms of military and national security, leaders must undertake talks that enshrine core principles and develop conduct codes that put those principles into concrete language. Failure to do this risks using AI in ways that are unfair, dangerous, or not very transparent.9

Some municipalities already have enacted procedural safeguards regarding surveillance technologies. Seattle, for example, has enacted a surveillance ordinance that establishes parameters for acceptable uses and mechanisms for the public to report abuses and offer feedback. The law defines relevant technologies that fall under the scope of the law but also illustrates possible pitfalls. In such legislation, it is necessary to define what tools rely upon algorithms and/or machine learning and how to distinguish such technologies from conventional software that analyzes data and acts on that analysis.10 Conduct codes wont be very helpful unless they clearly delineate the scope of their coverage.

Employees need appropriate operational tools that help them safely design and deploy algorithms. Previously, developing an AI application required detailed understanding of technical operations and advanced coding. With high-level applications, there might be more than a million lines of code to instruct processors on how to perform certain tasks. Through these elaborate software packages, it is difficult to track broad principles and how particular programming decisions might create unanticipated consequences.

Employees need appropriate operational tools that help them safely design and deploy algorithms.

But now there are AI templates that bring sophisticated capabilities to people who arent engineers or computer scientists. The advantage of templates is they increase the scope and breadth of applications in a variety of different areas and enable officials without strong technical backgrounds to use AI and robotic process automation in federal agencies.

At the same time, though, it is vital that templates be designed in ways where their operational deployment promotes ethics and fights bias. Ethicists, social scientists, and lawyers need to be integrated into product design so that laypeople have confidence in the use of these tools. There cannot be questions about how these packages operate or on what basis they make decisions. Agency officials have to feel confident that algorithms will make decisions impartially and safely.

Right now, it sometimes is difficult for agency officials to figure out how to assess risk or build emerging technologies into their missions.11 They want to innovate and understand they need to expedite the use of technology in the public sector. But they are not certain whether to develop products in-house or rely on proprietary or open-source software from the commercial market.

One way to deal with this issue is to have procurement systems that help government officials choose products and design systems that work for them. If the deployment is relatively straightforward and resembles processes common in the private sector, commercial products may be perfectly viable as a digital solution. But if there are complexities in terms of mission or design, there may need to be proprietary software designed for that particular mission. In either circumstance, government officials need a procurement process that meets their needs and helps them choose products that work for them.

We also need to keep humans in some types of AI decisionmaking loops so that human oversight can overcome possible deficiencies of automated software. Carnegie Mellon University Professor Maria De-Arteaga and her colleagues suggest that machines can reach false or dangerous conclusions and human review is essential for responsible AI.12

However, University of Michigan Professor Ben Green argues that it is not clear that humans are very effective at overseeing algorithms. Such an approach requires technical expertise that most people lack. Instead, he says there needs to be more research on whether humans are capable of overcoming human-based biases, inconsistencies, and imperfections.13 Unless humans get better at overcoming their own conscious and unconscious biases, manual oversight runs the risk of making bias problems worse.

In addition, operational tools must be human-centered and fit the agency mission. Algorithms that do not align with how government officials function are likely to fail and not achieve their objectives. In the health care area, for example, clinical decisionmaking software that does not fit well with how doctors manage their activities are generally not successful. Research by Qian Yang and her colleagues documents how user-centered design is important for helping physicians use data-driven tools and integrating AI into their decisionmaking.14

Finally, the community and organizational context matter. As argued by Michael Katell and colleagues, some of the most meaningful responsible AI safeguards are based not on technical criteria but on organizational and mission-related factors.15 The operationalization of AI principles needs to be tailored to particular areas in ways that advance agency mission. Algorithms that are not compatible with major goals and key activities are not likely to work well.

To have responsible AI, we need clear evaluation benchmarks and metrics. Both agency and third-party organizations require a means of determining whether algorithms are serving agency missions and delivering outcomes that meet conduct codes.

One virtue of digital systems is they generate a large amount of data that can be analyzed in real-time and used to assess performance. They enable benchmarks that allow agency officials to track performance and assure algorithms are delivering on stated objectives and making decisions in fair and unbiased ways.

To be effective, performance benchmarks should distinguish between substantive and procedural fairness. The former refers to equity in outcomes, while the latter involves the fairness of the process, and many researchers argue that both are essential to fairness. Work by Nina Grgic-Hlaca and colleagues, for example, suggests that procedural fairness needs to consider the input features used in the decision process, and evaluate the moral judgments of humans regarding the use of these features. They use a survey to validate their conclusions and find that procedural fairness may be achieved with little cost to outcome fairness.16

Joshua New and Daniel Castro of the Center for Data Innovation suggest that error analysis can lead to better AI outcomes. They call for three kinds of analysis (manual review, variance analysis, and bias analysis). Comparing actual and planned behavior is important as is identifying cases where systematic errors occur.17 Building those types of assessments into agency benchmarking would help guarantee safe and fair AI.

A way to assure useful benchmarking is through open architecture that enables data sharing and open application programming interfaces (API). Open source software helps others keep track of how AI is performing and data sharing enables third-party organizations to assess performance. APIs are crucial to data exchange because they help with data sharing and integrating information from a variety of different sources. AI often has impact in many areas so it is vital to compile and analyze data from several domains so that its full impact can be evaluated.

Technical standards represent a way for skilled professionals to agree on common specifications that guide product development. Rather than having each organization develop its own technology safeguards, which could lead to idiosyncratic or inconsistent designs, there can be common solutions to well-known problems of safety and privacy protection. Once academic and industry experts agree on technical standards, it becomes easy to design products around those standards and safeguard common values.

An area that would benefit from having technical standards is fairness and equity. One of the complications of many AI algorithms is the difficulty of measuring fairness. As an illustration, fair housing laws prohibit financial officials from making loan decisions based on race, gender, and marital status in their assessments.

One of the complications of many AI algorithms is the difficulty of measuring fairness.

Yet AI designers either inadvertently or intentionally can find proxies that approximate these characteristics and therefore allow the incorporation of information about protected categories without the explicit use of demographic background.18

AI experts need technical standards that guard against unfair outcomes and proxy factors that allow back-door consideration of protected characteristics. It does not help to have AI applications that indirectly enable discrimination by identifying qualities associated with race or gender and incorporating them in algorithmic decisions. Making sure this does not happen should be a high priority for system designers.

Pilot projects and organizational sandboxes represent ways for agency personnel to experiment with AI deployments without great risk or subjecting large numbers of people to possible harm. Small scale projects that can be scaled up when preliminary tests go well protect AI designers from catastrophic failures while still offering opportunities to deploy the latest algorithms.

Federal agencies typically go through several review stages before launching pilot projects. According to Dillon Reisman and colleagues at AI Now, there are pre-acquisition reviews, initial agency disclosures, comment periods, and due process challenges periods. Throughout these reviews, there should be regular public notices so vendors know the status of the project. In addition, there should be careful attention to due process and disparate analysis impact.

As part of experimentation, there needs to be rigorous assessment. Reisman recommends opportunities for researchers and auditors to review systems once they are deployed.19 By building assessment into design and deployment, it maximizes the chance to mitigate harms before they reach a wide scale.

The key to successful AI operationalization is a well-trained workforce where people have a mix of technical and nontechnical skills. AI impact can range so broadly that agencies require lawyers, social scientists, policy experts, ethicists, and system designers in order to assess all its ramifications. No single type of expertise will be sufficient for the operationalization of responsible AI.

For that reason, agency executives need to provide funded options for professional development so that employees gain the skills required for emerging technologies.20 As noted in my previous work, there are professional development opportunities through four-year colleges and universities, community colleges, private sector training, certificate programs, and online courses, and each plays a valuable role in workforce development.21

Federal agencies should take these responsibilities seriously because it will be hard for them to innovate and advance unless they have a workforce whose training is commensurate with technology innovation and agency mission. Employees have to stay abreast of important developments and learn how to implement technological applications in their particular divisions.

Technology is an area where breadth of expertise is as important as depth. We are used to allowing technical people to make most of the major decisions in regard to computer software. Yet with AI, it is important to have access to a diverse set of skills, including those of a non-technical nature. A Data and Society article recommended that it is crucial to invite a broad and diverse range of participants into a consensus-based process for arranging its constitutive components. 22 Without access to individuals with societal and ethical expertise, it will be impossible to implement responsible AI.

Thanks to James Seddon for his outstanding research assistance on this project.

The Brookings Institution is a nonprofit organization devoted to independent research and policy solutions. Its mission is to conduct high-quality, independent research and, based on that research, to provide innovative, practical recommendations for policymakers and the public. The conclusions and recommendations of any Brookings publication are solely those of its author(s), and do not reflect the views of the Institution, its management, or its other scholars.

Microsoft provides support to The Brookings InstitutionsArtificial Intelligence and Emerging Technology (AIET) Initiative. The findings, interpretations, and conclusions in this report are not influenced by any donation. Brookings recognizes that the value it provides is in its absolute commitment to quality, independence, and impact. Activities supported by its donors reflect this commitment.

Go here to see the original:

Six Steps to Responsible AI in the Federal Government - Brookings Institution

March 30, 2022 The MIT AI Hardware Programis a new academia and industry collaboration aimed at defining and developing translational technologies in hardware and software for the AI and quantum age. A collaboration between the MIT School of Engineering and MIT Schwarzman College of Computing, involving the Microsystems Technologies Laboratories and programs and units in the college, the cross-disciplinary effort aims to innovate technologies that will deliver enhanced energy efficiency systems for cloud and edge computing.

A sharp focus on AI hardware manufacturing, research, and design is critical to meet the demands of the worlds evolving devices, architectures, and systems, says Anantha Chandrakasan, dean of the MIT School of Engineering and Vannevar Bush Professor of Electrical Engineering and Computer Science. Knowledge-sharing between industry and academia is imperative to the future of high-performance computing.

Based on use-inspired research involving materials, devices, circuits, algorithms, and software, the MIT AI Hardware Program convenes researchers from MIT and industry to facilitate the transition of fundamental knowledge to real-world technological solutions. The program spans materials and devices, as well as architecture and algorithms enabling energy-efficient and sustainable high-performance computing.

As AI systems become more sophisticated, new solutions are sorely needed to enable more advanced applications and deliver greater performance, says Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing and Henry Ellis Warren Professor of Electrical Engineering and Computer Science. Our aim is to devise real-world technological solutions and lead the development of technologies for AI in hardware and software.

The inaugural members of the program are companies from a wide range of industries including chip-making, semiconductor manufacturing equipment, AI and computing services, and information systems R&D organizations. The companies represent a diverse ecosystem, both nationally and internationally, and will work with MIT faculty and students to help shape a vibrant future for our planet through cutting-edge AI hardware research.

The five inaugural members of the MIT AI Hardware Program are:

The MIT AI Hardware Program will create a roadmap of transformative AI hardware technologies. Leveraging MIT.nano, the most advanced university nanofabrication facility anywhere, the program will foster a unique environment for AI hardware research.

We are all in awe at the seemingly superhuman capabilities of todays AI systems. But this comes at a rapidly increasing and unsustainable energy cost, says Jess del Alamo, the Donner Professor in MITs Department of Electrical Engineering and Computer Science. Continued progress in AI will require new and vastly more energy-efficient systems. This, in turn, will demand innovations across the entire abstraction stack, from materials and devices to systems and software. The program is in a unique position to contribute to this quest.

The program will prioritize the following topics:

We live in an era where paradigm-shifting discoveries in hardware, systems communications, and computing have become mandatory to find sustainable solutions solutions that we are proud to give to the world and generations to come, says Aude Oliva, senior research scientist in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and director of strategic industry engagement in the MIT Schwarzman College of Computing.

The new program is co-led by Jess del Alamo and Aude Oliva, and Anantha Chandrakasan serves as chair.

Source: MIT News

Read more:

MIT AI Hardware Program Launches to Bolster Innovation in Next-Gen AI Hardware - HPCwire

Paid Feature Over the last several years, the limiting factors to large-scale AI/ML were first hardware capabilities, followed by the scalability of complex software frameworks. The final hurdle is less obvious, but if not overcome could limit what is possible in both compute and algorithmic realms.

This final limitation has less to do with the components of computation and everything to do with cooling those processors, accelerators, and memory devices. The reason why this is not more widely discussed is because datacenters already have ample cooling capabilities, most often with air conditioning units and the standard cold-aisle, hot aisle implementation.

Currently, it is still perfectly possible to manage with air cooled server racks. In fact, for general enterprise applications that require one or two CPUs, this is an acceptable norm. However, for AI training in particular, and its reliance on GPUs, the continued growth of AI capabilities means a complete rethink in how systems are cooled.

Apart from the largest supercomputing sites, the world has never seen the kind of ultra-dense AI-specific compute packed into a single node. Instead of two CPUs, AI training systems have a minimum of two high-end CPUs with an additional four to eight GPUs. The power consumption goes from 500 watts to 700 watts for a general enterprise-class server to between 2,500 watts and 4,500 watts for a single AI training node.

Imagine the heat generated from that compute horsepower then visualize an air conditioning unit trying to cool it with mere chilled air. One thing that becomes clear with that kind of per-rack density of compute and heat is that there is no way to blow enough air to sufficiently cool some of the most expensive, high performance server gear on the planet. This leads to throttling the compute elements or, in extreme cases, shutdowns.

This brings us to another factor: server rack density. With datacenter real estate demand at an all-time high, the need to maximize densities is driving new server innovations but the cooling can only keep up by leaving gaps in the racks (where more systems could reside) to let air try to keep up. Under these conditions, air cooling is insufficient to the task, and it also leads to less compute out of each rack and therefore more waste in server room space.

For normal enterprise systems with single-core jobs on two-CPU servers, the problems might not compound quite as quickly. But for dense AI training clusters, an enormous amount of energy is needed to bring cold air in, capture the heat on the back end, and bring it back to a reasonable temperature. This consumption goes well beyond what is needed to power the systems themselves.

With liquid cooling, you remove the heat far more efficiently. As Noam Rosen, EMEA Director for HPC & AI at Lenovo, explains, when you use warm, room temperature water, to remove heat to cool components, you do not need to cool anything; you dont invest energy to reduce water temperature. This becomes a very big deal as you get the node counts of the national lab and datacenters that do large-scale AI training.

Rosen points to quantitative details to compare general enterprise rack-level power needs versus those demanded by AI training via a lifecycle assessment on the training of several common large AI models. They examined the model training process for natural-language processing (NLP) and found that the NLP training process can emit hundreds of tons of carbon equivalent to nearly five times the lifetime emissions of an average car.

When training a new model from scratch or adopting a model to a new data set, the process emits even greater carbon due to the duration and computational power required to tune an existing model. As a result, researchers recommend industries and businesses to make a concerted effort to use more efficient hardware that requires less energy to operate.

Rosen puts warm water cooling in stark context by highlighting what one of Lenovos Neptune family of liquid cooled servers can do over the traditional air route. Today it is possible to take a rack and populate it with more than one hundred Nvidia A100 GPUs all in a single rack. The only way to do that is with warm water cooling. That same density would be impossible in an air-cooled rack because of all the empty slots to let the air cool components and even then, it likely could not address the heat from that many GPUs.

Depending on the server configuration, cooling by warm water can remove 85 percent to 95 percent of the heat. With allowable inlet temperatures for the water being as high as 45C, in many cases, energy-hungry chillers are not required, meaning even greater savings, lower total cost of ownership and less carbon emission, Rosen explains.

For customers who cannot, for whatever reason, add plumbing to their datacenter, Lenovo offers a system that features a completely enclosed liquid cooling loop that augments traditional air cooling. It affords customers the benefits of liquid cooling without having to add plumbing.

At this point in AI training with ultra-high densities and an ever-growing appetite for more compute to power future AI/ML among some of the largest datacenter operators on the planet, the only path is liquid and thats just from a datacenter and compute perspective. For companies doing AI training at any scale, the larger motivation should be keeping carbon emissions in check. Luckily, with efficient liquid cooling, emissions stay in check, electricity costs are slashed, densities can be achieved, and with good models, AI/ML can continue changing the world.

Sponsored by Lenovo.

Read more from the original source:

Liquid Cooling Is The Next Key To Future AI Growth - The Next Platform

The time is right for investing in the global natural language processing (NLP) market, projected to grow from $20.98 billion in 2021 to $127.26 billion in 2028 at a CAGR of 29.4% in that forecast period.

To get a sense on NLP user perspectives, this past February, Applause surveyed its global crowdtesting community to gain insight into perceptions around the use of artificial intelligence (AI) voice applications such as chatbots, interactive voice response (IVR), and other conversational assistants. Check out our summary infographic for some highlights. We had over 6,600 responses from around the world. I want to share our findings and call out a few interesting points.

While just over half of respondents reported they prefer to wait for a human agent when calling a company for customer support (51%), 25% said they prefer immediate access to an automated touch tone response system and 22% prefer an automated virtual service representative that responds to voice commands.

Consumers increasingly expect businesses to have automated chatbots and automated voice systems: 31% said they always expect companies to have chatbots, 61% said it depended on the industry. A vast minority, 6.7%, stated they never expect chat functionality on a companys website or app, while 11% dont expect call centers to have IVR systems that greet them. Still, customers expect IVR more often than not: 46% always expect call centers to have IVR systems that greet them while another 40% said their expectations varied by industry.

Webinars

Join voicebot.ai founder Bret Kinsella and Applauses Emerson Sklar as they cover lessons learned through global testing efforts and new models for conversational AI and other AI projects.

Users expect mobile apps to include voice functionality as well: 44% always expect mobile apps to have voice assistants or voice search features while 41% said it depends on the app category.

Of the 5896 respondents (88%) who said they had used chat functionality on a website at least once, 63% said they were somewhat satisfied or extremely satisfied with the experience. Of the 19% who found the experiences dissatisfying, the top three complaints were:

They could not find the answers they were looking for (29%)

The chatbot did not understand what they were asking (25%)

The chatbot wasted users' time (did not add value) before connecting them with an agent (20%)

Customers expect companies to have automated chatbots and automated voice systems to greet them and there is tremendous ROI for companies who get the NLP experience right, such as freeing up customer service reps for higher-value activities and reducing wait time for customers yet developing NLP technologies requires special attention to details that many other digital products may not.

Ebooks

Natural language assistants offer advantages to businesses. Our whitepaper covers these in detail and lists best practices for creating great user experiences.

Want to see more like this?

Share this:

Read more:

Pulsing Perceptions and Use of AI Voice Apps - ARC

BOSTON--(BUSINESS WIRE)--OReilly, the premier source for insight-driven learning on technology and business, today announced the results of its annual AI Adoption in the Enterprise survey. The benchmark report explores trends in how artificial intelligence is implemented, including the techniques, tools, and practices organizations are using, to better understand the outcomes of enterprise adoption over the past year. This years survey results showed that the percentage of organizations reporting AI applications in productionthat is, those with revenue-bearing AI products in productionhas remained constant over the last two years, at 26%, indicating that AI has passed to the next stage of the hype cycle.

For years, AI has been the focus of the technology world, said Mike Loukides, vice president of content strategy at OReilly and the reports author. "Now that the hype has died down, its time for AI to prove that it can deliver real value, whether thats cost savings, increased productivity for businesses, or building applications that can generate real value to human lives. This will no doubt require practitioners to develop better ways to collaborate between AI systems and humans, and more sophisticated methods for training AI models that can get around the biases and stereotypes that plague human decision-making.

Despite the need to maintain the integrity and security of data in enterprise AI systems, a large number of organizations lack AI governance. Among respondents with AI products in production, the number of those whose organizations had a governance plan in place to oversee how projects are created, measured, and observed (49%) was roughly the same as those that didn't (51%).

As for evaluating risks, unexpected outcomes (68%) remained the biggest focus for mature organizations, followed closely by model interpretability and model degradation (both 61%). Privacy (54%), fairness (51%), and security (42%)issues that may have a direct impact on individualswere among the risks least cited by organizations. While there may be AI applications where privacy and fairness arent issues, companies with AI practices need to place a higher priority on the human impact of AI.

While AI adoption is slowing, it is certainly not stalling, said Laura Baldwin, president of OReilly. There are significant venture capital investments being made in the AI space, with 20% of all funds going to AI companies. What this likely means is that AI growth is experiencing a short-term plateau, but these investments will pay off later in the decade. In the meantime, businesses must not lose sight of the purpose of AI: to make people's lives better. The AI community must take the steps needed to create applications that generate real human value, or we risk heading into a period of reduced funding in artificial intelligence.

Other key findings include:

The complete report is now available for download here: https://get.oreilly.com/ind_ai-adoption-in-the-enterprise-2022.html. To learn more about OReillys AI-focused training courses, certifications, and virtual events, visit http://www.oreilly.com.

About OReilly

For over 40 years, OReilly has provided technology and business training, knowledge, and insight to help companies succeed. Our unique network of experts and innovators share their knowledge and expertise through the companys SaaS-based training and learning platform. OReilly delivers highly topical and comprehensive technology and business learning solutions to millions of users across enterprise, consumer, and university channels. For more information, visit http://www.oreilly.com.

Read more:

As Adoption of Artificial Intelligence Plateaus, Organizations Must Ensure Value to Avoid AI Winter, According to New O'Reilly Report - Business Wire

From weather sensors and autonomous vehicles to electric grid monitoring and cloud gaming, the worlds edge computing is getting increasingly complex but the world of HPC hasnt necessarily caught up to these rapid innovations at the edge. At a panel at Nvidias virtual GTC22 (HPC, AI, and the Edge), five experts discussed how leading-edge HPC applications can benefit from deeper incorporation of AI and edge technologies.

On the panel: Tom Gibbs, developer relations for Nvidia; Michael Bussmann, founding manager of the Center for Advanced Systems Understanding (CASUS); Ryan Coffee, senior staff scientist for the SLAC National Accelerator Laboratory; Brian Spears, a principal investigator for inertial confinement fusion (ICF) energy research at Lawrence Livermore National Laboratory (LLNL); and Arvind Ramanathan, a principal investigator for computational biology research at Argonne National Laboratory.

The edge of a deluge, and a deluge of the edge

Early in the panel, Gibbs who served as the moderator of the discussion termed the 2020s the decade of the experiment, explaining that virtually every HPC-adjacent domain was in the midst of having a major experimental instrument (or a major upgrade to an existing instrument) come online. Its really exciting; but on the other side, these are going to produce huge volumes of rich data, he said. And how we can use and manage that data most effectively to produce new science is really one of the key questions.

Coffee agreed. Im actually at the X-ray laser facility at SLAC, and so I come at this from a short pulse time-resolved molecular physics perspective, he said. And we are facing an impending data deluge as we potentially move to a million-shots-per-second data rate, and so thats pulled me over the last half decade more into computing at the edge. And so where I feed into this is: how do we actually integrate the intelligence at the sensor with whats going on with HPC in the cloud?

One of the major opportunities I see moving forward is: we can look at rare events now, he continued. No one in their right mind is really going to move terabytes per second that just doesnt make sense however, we need the ability to record terabytes per second to watch for the anomalies that actually are driving the science that happens right now.

Spears, hailing from fusion research at LLNL, spoke to his field as a prime example. He pointed to recent success at LLNLs ICF experiment, where the team managed to produce 1.35 megajoules of energy from the fusion reaction almost break-even and that the JET team in Europe had a similar breakthrough in the last few months. But fusion research, he said, depended on data streams off of our cameras for experiments that last for you know, some of the action is happening over 100 picoseconds.

Sharpening AIs edge

So: huge amounts of data at very fast timescales, with the aim being to move from once-daily experiments to many-times-per-second. Spears explained how they planned on handling this. Were going to do things fast; were gonna do them in hardware at the edge; were probably gonna do them with an AI model that can do low-precision, fast compute, but thats going to be linked back to a very high-precision model that comes from a leadership-class institution.

You can start to see from these applications the convergence of the experiment and timescales, he said, driving changes in the way we think about representing the physics and the model and moving that to the edge[.]

AI, then, accelerates this same strategy, helping to whittle down the data that moves from the edge to the larger facilities. You can use AI in terms of guiding where the experiments must go, in terms of seeing what data we might have missed, Ramanathan said.

Bussmann agreed, citing many fields employing a live stream [of data] that will not be recorded forever so we have to make fast decisions and we have to make intelligent decisions, he said. We realize that this is an overarching subject across domains by now because of the capabilities that have become [widespread].

AI provides the capability to put a wrapper around that, train a lightweight surrogate model, and take what I actually think in my head and move it toward the edge of the computing facility, Spears said. We can run the experiments now for two purposes: one is to optimize whats going on with the actual experiment itself so we can be moving to a brighter beam or a higher-temperature plasma but we can also say, I was wrong about what I was thinking, because as a human, I have some weaknesses in my conception of the way the world looks. So I can also steer my experiment to the places where Im not very good, and I can use the experiment to make my model better. And if I can tighten those loops by doing the computing at the edge I can have these dual outcomes of making my experiment better and making my model better.

Just a matter of time

Much of the discussion in the latter half of the panel focused on how these AI and edge technologies could be used to usefully interpolate sparse or low-resolution data. You really need these surrogate models, Ramanathan said, explaining how his drug discovery work operated in ranges of 15 to 20 orders of magnitude and that to tackle it, it was useful to build models that can adaptively sample this landscape without having all of this information: rare event identification.

I can run a one-dimensional model really cheaply I can run 500 million of those, maybe, Spears said. A two-dimensional model is a few hundred or a thousand times more expensive. A three-dimensional model is thousands of times more expensive than that. All of those have advantages in helping me probe around in parameter space, so what a workflow tool allows us to do is make decisions back at the datacenter saying: run interactively all of these 1D simulations and let me make a decision about how much information gain Im getting from these simulations.

And when I think Ive found a region or parameter or design space that is high-value real estate, Ill make a workflow decision to say, plant some 2D simulations instead of 1Ds and Ill hone in on another more precise area of the high-value real estate. And then I can elevate again to the three-dimensional model which I can only run a few times. Thats all high-precision computing thats being steered on a machine like Sierra that we have at Lawrence Livermore National Laboratory.

All of our problems are logarithmically scaled, right? added Coffee. We have multiple scales that we want to be sensitive to it doesnt matter which domain youre in. We all are using computers to help us do the thing that we dont do well, which is swallow data quickly enough.

When you start talking about integrating workflows for multiple domains and they all have a similar pattern of use, doesnt that beg us to ask for an infrastructure where we bind HPC together with edge to follow a common model and a common infrastructure across domains? he continued. I think were all asking for the same infrastructure. And this infrastructure is really, now, not just what happens in the datacenter, right? Its what happens in the datacenter and how its sort of almost neurologically connected to all of the edge sensors that are distributed broadly across our culture.

See the original post here:

At GTC22, HPC and AI Get Edgy - HPCwire

LONDON, March 30, 2022 (GLOBE NEWSWIRE) -- According to The Business Research Companys research report on the artificial intelligence in healthcare market, AI-driven surgical robots are gaining prominence among the artificial intelligence in healthcare market trends. Various healthcare fields have adopted robotic surgery in recent times. Robot-assisted surgeries are performed to remove limitations during minimally invasive surgical procedures and to improve surgeons' capabilities during open surgeries. AI is widely being applied in surgical robots and is also used with machine vision to analyze scans and detect complex cases. While performing surgeries in delicate areas of the human body, robotic surgeries are more effective than manually performed surgeries. To meet healthcare needs, many technology companies are providing innovative robotic solutions.

For example, in 2020, Accuracy Incorporated, a US-based company that develops, manufactures, and sells radiotherapy systems for alternative cancer treatments, launched a device called the CyberKnife S7 System, which combines speed, advanced precision, and AI-driven motion tracking for stereotactic radiosurgery and stereotactic body radiation therapy treatment.

Request for a sample of the global artificial intelligence in healthcare market report

The global artificial intelligence in healthcare market size is expected to grow from $8.19 billion in 2021 to $10.11 billion in 2022 at a compound annual growth rate (CAGR) of 23.46%. The global AI in healthcare market size is expected to grow to $49.10 billion in 2026 at a CAGR of 48.44%.

The increase in the adoption of precision medicine is one of the driving factors of artificial intelligence in the healthcare market. Precision medicine uses information about an individual's genes, environmental and lifestyle changes to design and improve the diagnosis andtherapeutics of the patient. It is widely used for oncology cases, and due to the rising prevalence of cancer and the number of people affected by it, the demand for AI in precision medicine will increase. According to research published in the Lancet Oncology, the global cancer burden is set to increase by 75% by 2030.

Major players in the artificial intelligence in healthcare market are Intel Corporation, Nvidia Corporation, IBM Corporation, Microsoft Corporation, Google Inc., Welltok Inc., General Vision Inc., General Electric Company, Siemens Healthcare Private Limited, Medtronic, Koninklijke Philips N.V., Micron Technology Inc., Johnson & Johnson Services Inc., Next IT Corporation, and Amazon Web Services.

The global artificial intelligence in healthcare market is segmented by offering into hardware, software; by algorithm into deep learning, querying method, natural language processing, context aware processing; by application into robot-assisted surgery, virtual nursing assistant, administrative workflow assistance, fraud detection, dosage error reduction, clinical trial participant identifier, preliminary diagnosis; by end-user into hospitals and diagnostic centers, pharmaceutical and biopharmaceutical companies, healthcare payers, patients.

As per the artificial intelligence in healthcare industry growth analysis, North America was the largest region in the market in 2021. Asia-Pacific is expected to be the fastest-growing region in the global artificial intelligence in healthcare market during the forecast period. The regions covered in the global artificial intelligence in healthcare market report are Asia-Pacific, Western Europe, Eastern Europe, North America, South America, the Middle East, and Africa.

Artificial Intelligence In Healthcare Market Global Market Report 2022 Market Size, Trends, And Global Forecast 2022-2026 is one of a series of new reports from The Business Research Company that provide artificial intelligence in healthcare market overviews, analyze and forecast market size and growth for the whole market,artificial intelligence in healthcare market segments and geographies, artificial intelligence in healthcare market trends, artificial intelligence in healthcare market drivers, artificial intelligence in healthcare market restraints,artificial intelligence in healthcare market leading competitors revenues, profiles and market shares in over 1,000 industry reports, covering over 2,500 market segments and 60 geographies.

The report also gives in-depth analysis of the impact of COVID-19 on the market. The reports draw on 150,000 datasets, extensive secondary research, and exclusive insights from interviews with industry leaders. A highly experienced and expert team of analysts and modelers provides market analysis and forecasts. The reports identify top countries and segments for opportunities and strategies based on market trends and leading competitors approaches.

Not the market you are looking for? Check out some similar market intelligence reports:

Robotic Surgery Devices Global Market Report 2022 By Product And Service (Robotic Systems, Instruments & Accessories, Services), By Surgery Type (Urological Surgery, Gynecological Surgery, Orthopedic Surgery, Neurosurgery, Other Surgery Types), By End User (Hospitals, Ambulatory Surgery Centers) Market Size, Trends, And Global Forecast 2022-2026

Artificial Intelligence (AI) In Drug Discovery Global Market Report 2022 By Technology (Deep Learning, Machine Learning), By Drug Type (Small Molecule, Large Molecules), By Therapeutic Type (Metabolic Disease, Cardiovascular Disease, Oncology, Neurodegenerative Diseases), By End-Users (Pharmaceutical Companies, Biopharmaceutical Companies, Academic And Research Institutes) Market Size, Trends, And Global Forecast 2022-2026

Precision Medicine Global Market Report 2022 By Technology (Big Data Analytics, Bioinformatics, Gene Sequencing, Drug Discovery, Companion Diagnostics), By Application (Oncology, Respiratory Diseases, Central Nervous System Disorders, Immunology, Genetic Diseases), By End-User ( Hospitals And Clinics, Pharmaceuticals, Diagnostic Companies, Healthcare And IT Firms) Market Size, Trends, And Global Forecast 2022-2026

Interested to know more about The Business Research Company?

The Business Research Company is a market intelligence firm that excels in company, market, and consumer research. Located globally it has specialist consultants in a wide range of industries including manufacturing, healthcare, financial services, chemicals, and technology.

The Worlds Most Comprehensive Database

The Business Research Companys flagship product, Global Market Model, is a market intelligence platform covering various macroeconomic indicators and metrics across 60 geographies and 27 industries. The Global Market Model covers multi-layered datasets which help its users assess supply-demand gaps.

Go here to see the original:

Companies In The Artificial Intelligence In Healthcare Market Are Introducing AI-Powered Surgical Robots To Improve Precision As Per The Business...