Gigaom AI Minute – May 24

Transcript

• A new AI system from startup AtomWise aims to significantly shorten the amount of time and money it takes to bring new medicinal drugs to market. The software, called AtomNet, uses deep learning to identify molecular patterns to screen a million compounds a day versus months the traditional way.

• Clinical experts at Harvard Hospitals are teaming up with engineers at GE on a 10-year AI effort to improve medical care. Ambitions include using AI to analyzing medical images, building AI applications for genomics, population health, and a wide-ranging scope of healthcare use cases.

• The largest maker of graphics and AI chips, Nvidia, has unveiled a new processor for AI applications that can transfer data at 300 gigabits per second. That's 20 times faster than other processors today. The Tesla V100, as it's called, also has 12X the processing speed to power deep learning algorithms than last year's Pascal-based processor.

• Using just a minute of voice audio, now anyone's voice can be copied. That's what Lyrebird, a Montreal-based startup has made possible through the launch of its new API. The service synthesizes speech without the speaker actually speaking; it can learn and even add intonations afterward from just analyzing a recording.

• Baidu AI has programmed a robot to teach another robot to read. The bot learned the real meaning of words through simple positive and negative feedback, even without previous context or input defining the problems. The idea is that the breakthrough could help future AI understand language exactly the same way as people.

• A machine learning algorithm is using sentiment analysis to make stock price predictions with 76 percent accuracy. Triumph Asset Management, an investment management firm is using the algorithm to parses large volumes of news articles for "market moving" indicators that can help traders know whether or not it's a good time to buy or sell.