AI Weekly: AI beats race car drivers in a video game and a U.S. bill promises to regulate social media

Join online with today’s leading executives at the Data Summit on March 9th. Register here.


In AI this week, Sony – the Japanese electronics giant that still sells luxury MP3 players – announced Gran Turismo (GT) Sophy, an AI system that outperformed human drivers in a head-to-head competition in the video game series Gran Turismo. Can. Alphabet-owned AI research lab DeepMind revealed that YouTube is now using AI that specializes in chess to compress streaming videos. And in other news, the US Sen. Amy Klobutcher (D-MN) and Cynthia Lumis (R-WY) introduced a bill that would be the first major step for Congress to address, as The Virgin “Algorithmic amplification of losses,” says McKenna Kelly.

In at least some circles, the GT Sophy has moved away from AI field video games as a measure of progress. But one expert said Wired He described it as a “landmark achievement for AI” because the technologies used to develop it could be applied to real-world self-driving cars. While autonomous cars currently rely heavily on manually-written software for control, the GT Sophie – whose details are detailed in a paper published in the journal Architecture. Nature This week – AI benefits for a better effect.

Sony researchers worked with Polyphony Digital, the studio behind Gran Turismo, to train the system for hours and hours. Through a technique called Reinforcement Learning, which “rewards” or “punishes” the system for making beneficial or unfavorable decisions, GT Sophie learned to beat professional esports drivers without imposing penalties for improper play.

The GT Sophie could pave the way for games featuring AI players that are more alive today than ever before. But beyond that (and potentially promoting real-world self-driving cars), the GT Sophie can help drivers improve their skills. “Sophie takes on some racing lines the human driver could never have imagined,” said Kazunori Yamau, creator. Gran Turismo And a real life race car driver, said Wired“I think a lot of textbooks on driving skills will be rewritten.”

Video-compressing AI

In December 2020, Deepmind published a paper in the journal Nature Muzero details an algorithm that chooses the rules of games like chess, shogi and go as he plays. At the time, lab researchers said they were planning to explore potential commercial applications, and now, they seem to have found one: compressing YouTube videos.

Deepmind claims that MuZero was able to reduce the average amount of data YouTube needed to stream users by 4%, without causing significant damage to video quality. Deepmind says – the system is now actively used in most videos on YouTube – but not all – mostly working to improve Google’s open source video compression codec VP9.

MuZero treats video compression as a “game”, trying to compete with its own previous attempts to compress video and improve certain quality and bitrate metrics. Like the GT Sophie, it is a reinforcement education system, but MuZero is not given any past examples of effective strategies, meaning he should learn through his own experience.

MuZero has led to surprising insights, for example ignoring the standard video compression rule that the bitrate should be the maximum for the first frame in the scene and then after 10 frames. Muzero found that, for many video sequences, as long as the bitrate is maximized for one of these two frames, the other does not require much bandwidth.

MuZero isn’t perfect, though. It competes with slide-style videos, allocating more bandwidth to transition frames when skimming on slides. According to Deepmind, YouTube engineers discovered this and fixed it with hard-coded rules for that type of video.

Deepmind claims that Muziro can help people in countries with limited broadband to see what they would otherwise struggle to see. “If you can compress videos and make them smaller, you can save a lot on all Internet traffic,” Dave Silver, who heads the reinforcement education group at Deepmind, told VentureBeat in an earlier interview. “This is something we can apply our learning algorithms to and has a lot of real world features because you never know what you’re going to see next in the video.”

Algorithm regulation

Sensation. Klobutcher and Lumis’s bill is proposed before Congress to regulate a very different type of algorithm than GT Sophie: who recommends content on social media. The bill, known as the Social Media NUDGE Act, would direct agencies, including the National Science Foundation, to identify “content neutral” methods for slowing the spread of harmful content and misinformation, and would mandate the Federal Trade Commission to lay down social media platforms. These different methods in practice.

Notably, the NUDGE Act will not amend Section 230 of the Communication Density Act, which protects web hosts and platforms from legal liability for content uploaded to their services. Last March, Reps. Anna Ashu (D-CA) and Tom Malinowski (D-NJ) introduced a similar algorithmic regulation bill, which protects Americans from the Dangerous Algorithms Act, but one that would put hosts and platforms on the hook for “amplified content.” Violation of civil rights.

“For a long time, tech companies have said, ‘Trust us, we’ve got this,'” Klobucher said in a statement. “But we know that social media platforms often profit people, algorithms push dangerous content that hooks users and spreads misinformation.”

The NUDGE Act would go further than the EU’s proposed AI legislation, “The AI ​​Act”, which prohibits certain uses of AI and drastically regulates “high-risk” uses such as public benefits, credit scoring and eligibility determining systems. Is. Delivery of emergency services. While the AI ​​Act requires AI providers and users to adhere to the rules on data governance, documentation, transparency, accuracy and security, it does not consider the algorithms used in social media to be high risk (unless a regulator considers it. -Base based).

Public Knowledge’s Greg Guis says Facebook’s own report found that its algorithms have enabled it to spread false information and misinformation, with bills such as the NUDGE Act being overdue. “Public knowledge supports this law because it encourages informed decision-making to promote a known problem: misinformation,” he said in a statement. “Most importantly, Bill does all of this without complying with Article 230 immunity.”

However, only time will tell whether the NUDGE Act will be able to attract the support of fellow delegates before the next midterm elections, which will change the balance of power in Congress. If that doesn’t happen, it’s still a way to get more involved with the social media, given the political differences over the issue.

For AI coverage, send news tips to Kyle Wiggers – and be sure to subscribe to the AI ​​weekly newsletter and bookmark our AI channel, The Machine.

Thanks for reading,

Kyle Wiggers

AI Senior Staff Writer

Venturebeat’s mission Digital Town Square is set to become a place for technical decision makers to gain knowledge about the changing enterprise technology and practices. Learn more

Similar Posts

Leave a Reply

Your email address will not be published.