Bipedal Robot Makes Historic 5KM Run by Using Machine Learning

Cassie, developed by engineers at Oregon State University is the first bipedal robot to use machine learning to control a running gait on outdoor terrain. It completed the five-kilometers run untethered around the campus in just over 53 minutes on a single charge. This includes 6.5 minutes of resets following 2 falls, which are due to overheated computer and being asked to execute a turn at too high a speed. It is the first time a robot has learned to walk and run, and successfully do that outside over human terrain. Since Cassie’s introduction in 2017, engineers and students have been exploring machine learning options for the robot. Cassie has knees that bend like an ostrich’s, and it can teach itself to run with a deep reinforcement learning algorithm. Running requires dynamic balancing – the ability to maintain balance while switching positions or otherwise being in motion – and Cassie has learned to make infinite subtle adjustments to stay upright while moving. In a related project, Cassie has become adept at walking up and down stairs. As for potential application, bipedal robots with intelligence have capabilities to help people in more scenarios beyond logistics work.

Duolingo IPO: Negative Impacts of EU Proposals to Regulate AI on Innovation

Founded in 2011, Duolingo is an American language learning company offering a website, mobile app and assessment platform that assist over 300 million global users in learning up to 40 languages. On July 28th, it went public on NASDAQ and was valued at approximately $5 billion by the closing price. However, EU proposals for regulating AI, which was released in April, threaten the use of one of Duolingo’s niftiest innovations, the English Test, in its current form. The English Test is qualified by over 3,000 educational institutions to demonstrate proficiency. But this is also a high-risk AI system according to the EU proposal. This label applies because the test uses AI, both for personalisation — questions appropriate to the taker’s skill level are generated on the fly — and for grading. The requirements for high-risk AI systems and the obligations placed on their providers take up 10 pages in the proposal, which is expensive and time consuming. This might hinder the innovation for AI-using entrepreneurs. Although we recognise the importance of regulations on data protection and privacy law, restricting the field of potential innovators just to those who can afford high upfront costs is not a good idea for long term development.

Google is Using AI to Design New Generation of AI Chips Fasters than Humans

As Google said in a paper in the journal Nature, algorithm’s designs are “comparable or superior” to those created by humans but can be generated much faster. Work that takes months for humans can be accomplished by AI in under six hours. Google has been working on how to use machine learning to create chips for years, but this recent effort seems to be the first time its research has been applied to a commercial product: an upcoming version of Google’s own TPU (tensor processing unit) chips, which are optimized for AI computation. Google’s engineers note that this work has major implications for the chip industry. It should allow companies to explore the possible architecture space more quickly for upcoming designs customize chips for specific workloads more easily. This could help offset the forecasted end of Moore’s Law. AI won’t necessarily solve the physical challenges of squeezing more transistors onto chips, but it could help find other paths to increasing performance at the same rate. The virtuous cycle of AI designing chips for AI looks like it is just getting started. Google itself has explored using AI in other parts of the process such as architecture exploration, and rivals like Nvidia are looking into other methods to speed up the workflow.