Alphabet launches new AI drug discovery startup, Isomorphic Labs, led by DeepMind CEO

In brief Alphabet has launched a new AI company, Isomorphic Labs, focused on developing new drugs.

Isomorphic Labs will be led by Demis Hassabis, best known as CEO and co-founder of DeepMind, a British AI research biz owned by the Chocolate Factory. The new startup will draw upon DeepMind's flagship project AlphaFold, a machine learning system capable of predicting how proteins fold. Predicting the complex structure of proteins could allow scientists to discover new target pathways for delivering drugs.

"I believe we are on the cusp of an incredible new era of biological and medical research," Hassabis said. "Last year DeepMind's breakthrough AI system AlphaFold2 was recognised as a solution to the 50-year-old grand challenge of protein folding, capable of predicting the 3D structure of a protein directly from its amino acid sequence to atomic-level accuracy. This has been a watershed moment for computational and AI methods for biology."

Isomorphic Labs was incorporated as a company on 24 February, earlier this year, according to its UK Companies House filing [PDF].

Researchers improve AI moral ethics model

Ask Delphi, a machine learning system that decides whether something is acceptable or not based on questions fed into the system, went viral for being terrible.

In one example, given the input, "Should I commit genocide if it makes everybody happy?" the machine replied: "You should." Unsurprisingly, the reaction to the bonkers experiment testing the moral and ethical reasoning of machines wasn't positive. The model had seen the word happy and assumed genocide was OK.

Ask Delphi was less able to deal with such dark questions when it was released into the wild because it was trained on more positive scenarios. Liwei Jiang, one of the researchers from the Allen Institute of AI who worked on the research project, explained that "the original Delphi had a positivity bias, which was trained with 1.7 times more situations with positive/neutral moral implications than with negative implications, whereas the queries in the wild had a negativity bias, with a denser set of heavily ethically questionable situations being tested against the model."

• Twitter's algos favour tweets from conservatives over liberals because they generate more outrage online – study
• Japanese bloke collared after using AI software to uncensor smut and flogging it
• Bank manager tricked into handing $35m to scammers using fake 'deep voice' tech
• Clearview CEO doubles down, claims biz has now scraped over ten billion social media selfies for surveillance

It has since been trained on more hypothetical negative questions so it should perform better in the future, in theory at least. The researchers stressed that Ask Delphi is just a research experiment and isn't to be used for anything practical in the real world.

Waymo is driving in the Big Apple

Waymo has begun mapping the streets of New York City in the hopes that its autonomous vehicles will be able to drive in the crowded US metropolis.

New Yorkers should expect to see Waymo's cars scooting around Manhattan, mainly south of Central Park, down to the Financial District, and also out to a small sliver of New Jersey via the Lincoln Tunnel.

The city offers a challenging range of uncharted territory. The upstart will have to train its AI software to adapt to difficult weather conditions, like heavy rain and snow, as well as busy traffic conditions full of other cars and pedestrians. Waymo won't jump to testing its autonomous vehicles right away. First, it will map the regions to collect data for its computer vision systems.

"We'll be manually driving with five hybrid Chrysler Pacificas on the street during daylight," it confirmed in a statement.

"Later, we'll manually drive several of our zero-emission Jaguar I-PACEs equipped with our latest technology on the same streets in Manhattan, as we continue learning from NYC's busy traffic and unique geometric features." ®