Last Week in AI #154: Light-based AI chips, DeepMind AI for competitive coding, EU and US AI regulations, and more!
Celestial AI lands funding to develop photonic AI accelerator chips, DeepMind's new research on AI that solves competitive programming questions, EU and US aligning on AI regulations.
Photonic chips may help speed up common workloads by AI algorithms. These chips use light, instead of electricity, to perform computation and send signals. In theory, they produce less heat and can operate faster. In practice, however, existing prototypes are too big and hard to manufacture, and they still rely on electronic chips for some controls. Many startups, like Celestial AI, are trying to realize the promise of photonic chips. Celestial is developing an architecture that can scale to multiple chips, allowing faster parallel computations. Other companies have already started selling accelerators that include photonic components. Experts estimate that only 20% of the potential benefits of photonic chips have been unlocked to date, so there is plenty of development room to go.
Our take: There’s been a lot of concern in recent years about how Moore’s law is ending. While the rate of improvement for general-purpose chips has slowed, the improvement of application-specific chips, like ones that perform AI workloads, is just getting started. There are plenty of untapped innovations, like photonic chips, that can really shine with the highly parallel, relatively low-precision workloads of AI algorithms, and their benefits will only become more apparent as AI becomes more common in everyday compute.
AlphaCode achieved an estimated rank within the top 54% of participants in programming competitions by solving new problems that require a combination of critical thinking, logic, algorithms, coding, and natural language understanding.
The approach is largely similar to other recent works focused on generating text; a large Transformer is trained with a large dataset scraped from the internet, like GitHub Copilot. This results in an impressive performance on generating programs to solve specific problems, a difficult problem for AI. While better than average, the performance still lives much to be desired:
On CodeContests, given up to a million samples per problem, AlphaCode solved 34.2% of problems. And on Codeforces, DeepMind claims it was within the top 28% of users who’ve participated in a contest within the last six months in terms of overall performance.
Our take: Frankly, this does not seem that exciting. The approach taken is not particularly interesting and does not provide insights on addressing other important problems. Still, this does represent progress on this particular task, which in itself is good.
Many are calling for the US to take lessons from the EU in developing a more proactive approach to AI regulation. While the EU developed a comprehensive but flawed AI Act, the US's approach to AI governance remained a patchwork with little coherence. But Brookings claims these gradual developments "are aggregating into a meaningful approach to AI regulation" in line with the EU's approach. Actions such as bringing experts into the Biden administration and the Platform Accountability and Transparency Act suggest an increasing pace of regulatory development. Brookings concludes that the US would do well to establish proactive regulatory cooperation with the EU, such as working towards a common definition of AI and establishing regulatory sandboxes.
Our take: There is still disagreement regarding the compatibility of regulation and fast-paced innovation. Notably, France has expressed its disagreement with the EU's regulation-first approach. It is hard to know how certain regulations will affect the pace of development without some empirical evidence, but there are a number of competing interests at play for AI regulations, like international competition. Ideas like regulatory sandboxes are also interesting. However, the US needs to set out plainly what its priorities are with respect to the development of emerging technologies and develop regulations coherent with those goals. China, for instance, has cracked down on recommendation systems while supporting companies like SenseTime. This results from its clearly specified priorities that value “grounded” technologies in spaces like security over attention aggregators.
Artificial intelligence system rapidly predicts how two proteins will attach - "The machine-learning model could help scientists speed the development of new medicines."
Reimagining Chess with AlphaZero - "The combination of human curiosity and a powerful reinforcement learning system allowed us to reimagine what chess would have looked like if history had taken a slightly different course."
Chinese scientists create AI nanny to look after babies in artificial womb - "Researchers in Suzhou have developed an AI system able to monitor and take care of embryos as they grow into fetuses in the lab"
Researchers Introduce A Novel Human-Like Driving And Decision Making Framework Designed For Autonomous Vehicles (AVs) - "In the last years, we assisted to a great hype around autonomous vehicles (AVs). However, if we want to see streets where AVs and human drivers live side by side, AVs must be able to dive into our transportation system."
Solving (Some) Formal Math Olympiad Problems - "We built a neural theorem prover for Lean that learned to solve a variety of challenging high-school olympiad problems, including problems from the AMC12 and AIME competitions, as well as two problems adapted from the IMO."
Machine learning helps improve the flash graphene process - "Scientists at Rice University are using machine-learning techniques to fine-tune the process of synthesizing graphene from waste through flash Joule heating."
Machine Learning Gets a Quantum Speedup - "Two teams have shown how quantum approaches can solve problems faster than classical computers, bringing physics and computer science closer together."
Scientists are building an AI-powered robot to take apart your old phone - "Think Apple’s iPhone recycling robot but for any brand"
How video games could be used to generate AI training data - "Games have been used to crowdsource data in the recent past, particularly in domains like protein molecule folding, RNA behavior, and complex genome sequencing."
AI-generated Valentine's Cards - "what if the cards were generated by a glitchy AI?"
Tell this AI your story's themes, and it'll write the first paragraph for you - "It's fun to make computers say silly things"
Airbnb says AI program thwarts thousands of party houses in Florida - "We believe it worked. Those (holiday) weekends were generally quiet."
Artist uses AI to design terrifying, faceless, Bosch-style tarot cards. You okay, hun? - "Welcome to the uncanny valley, where artificial intelligences like Looking Glass take snippets of what we know of this world, and turn them into forms almost recognisable, but with just enough absurdity to make you feel weird inside."
AI-driven personalized content grows in importance for marketers - "A new survey from AI creative platform Persado underlines increasing interest in natural language processing and generation (NLP/NLG) among marketers, driven by the demand for personalized messaging and the need to extract maximum value from first-party data."
AI traffic lights could reduce waiting times at intersections - "Ordinarily, traffic lights change color in response to predetermined rules and induction loop sensors embedded in the road. According to new research, however, making those lights artificially intelligent may keep traffic flowing faster and smoother."
AI could cut hiring biases as companies make push to find workers, proponents say - "While the potential for bias in AI has been a concern, some say well-crafted algorithms can help companies be fairer in their hiring decisions"
How machine learning can clean up our cities - "Computational design can identify greater efficiencies across the built environment, enabling us to create more sustainable cities."
Superpedestrian gets $125M to expand tech that corrects unsafe scooter riding - "Shared e-scooter company Superpedestrian is gearing up to deploy its Pedestrian Defense safety system at scale, a feature that allows the company’s e-scooters to detect and correct unsafe riding behaviors, like riding on sidewalks or the wrong way down a one-way street, in real time."
Machine learning the hard way: IBM Watson's fatal misdiagnosis - "When clinical trials were published, Watson came up short every time. It didn't matter what field it was in, it consistently scored less well than human clinicians – sometimes under 50 per cent – and demonstrated some alarming blind spots in suggested treatments."
Scale AI awarded $250M AI contract by Department of Defense - "Scale AI, a tech startup that focuses on the data that powers artificial intelligence, has won a nearly $250 million blanket purchasing agreement that aims to give all federal agencies access to its tech."
Electric Sheep Robotics raises $21.5M for autonomous lawn mowers - "Electric Sheep Robotics, a developer of autonomous navigation technology for lawn mowers, $21.5 million in Series A funding. It also announced the general availability of its Dexter system that turns existing commercial lawn mowers, both gas and electric, into autonomous mowers."
Self-driving tech firm Cruise to raise $1.35 bln from SoftBank as it readies robotaxis - "Self-driving tech firm Cruise on Tuesday said investor SoftBank Group Corp’s (9984.T) Vision Fund has agreed to invest an additional $1.35 billion through a second tranche, as it is working to launch commercial ridehailing service in San Francisco."
Two of Google's Ethical AI Staffers Leave to Join Ousted Colleague’s Institute - "Google’s Ethical AI research group lost two more employees, adding to the turmoil at the unit studying an area that is vitally important to the technology giant’s business future and political standing."
AI and Machine Learning Salaries Drop - "But average U.S. tech salaries climbed nearly 7 percent in 2021"
Tesla agrees to fix "rolling stop" software feature over safety fears - "The company said the update would disable the controversial feature, which allowed some cars to roll through intersections with stop signs instead of braking completely."
US Treasury considers alternatives to ID.me biometrics for IRS online access - "An Internal Revenue Service contract being served by ID.me for online tax returns is continuing to generate controversy based on the use of facial recognition and biometric authentication, prompting a search for alternative identity verification measures."
AI Insurance Company Faces Class Action for Use of Biometric Data - "After a tweeting mishap, Lemonade Inc., an AI-based insurance company, faces a class action for allegedly violating New York laws against the use of biometric data without consent by using facial recognition technology to analyze videos submitted in the claims process."
Why ‘the future of AI is the future of work’ - "In a new book about how technology will affect workers, MIT experts explain how artificial intelligence is far from replacing humans — but still changing most occupations."
Human Borgs: How Artificial Intelligence Can Kill Creativity And Make Us Dumber - "The research, published in late 2021, uncovers risks, consequences, and solutions to over reliance on AI in business and creative decisions."
Commentary: How homeowners defeated Zillow’s AI, which led to Zillow Offers’ demise - "Now that the dust has settled, we can ask how could the smart and savvy team at Zillow Group, armed with cutting-edge AI methods and mountains of data, lose half a billion dollars buying homes in the second half of 2021."
The state of AI ethics: The principles, the tools, the regulations - "The institute’s State of AI Ethics Reports, published semi-annually, condense the top research & reporting around a set of ethical AI subtopics into one document."
Democratic lawmakers take another stab at AI bias legislation - "Democrats in Congress on Thursday renewed a push to hold tech companies accountable for bias in their algorithms."
CFPB's scrutiny of AI in banking could help bring much-needed clarity - "The Consumer Financial Protection Bureau (CFPB) is signaling that it will clamp down on artificial intelligence (AI) practices in banking that can be used to discriminate against people, per American Banker."
Copyright © 2022 Skynet Today, All rights reserved.