Why Meta and Twitter’s AI and ML layoffs matter | The AI ​​beat

Ten days ago, as part of massive Twitter layoffs, the company’s entire ethical artificial intelligence (AI) team — which worked to make Twitter’s algorithms more transparent and fair — was let go. The team, called ML Ethics, Transparency, and Accountability, was led by Rumman Chowdhury, who is known for her leadership in applied algorithmic ethics.

Meanwhile, Meta’s layoffs last week of 11,000 employees, or 13% of the company’s workforce, included a full 50-person research team focused on machine learning (ML) infrastructure called Probability. The probability team consisted of 19 people doing Bayesian modeling, 9 people doing ranking and recommendations, 5 people doing ML efficiency, 17 people doing AI for chip design and compilers, as well as managers, according to a researcher on the team.

Both sets of layoffs are significant, experts say, as they signal a shift in the landscape of even the most sought-after AI and ML talent, as well as a reckoning for big tech and corporate firms in terms of how they respond to their own. responsibility. AI efforts.

Georgios Gousios, head of research at software company Endor Labs and associate professor at Delft University of Technology in the Netherlands, told VentureBeat by email that Meta’s Probability team was the “equivalent of an elite military tactical unit.”

Gousios, who worked on the Probability team from October 2020 to February 2022, said that while Facebook had many developers working on different parts of the tech stack and the company, Probability was doing work “that is orthogonal to day-to-day software production.” , focused on inventing and applying new tools/methods that would make the other teams more efficient in their day-to-day work.”

This included, he explained, probabilistic programming (writing programs where variables are represented by distributions rather than single values), differentiable programming (making neural networks more efficient), and software engineering applications such as tools that use ML to help engineers to help both write code faster with fewer bugs, and debug unavoidable problems faster.

“The quality of the team was extremely high,” he said. “Many of us (myself included) came from years of academic research; many had decades of industrial research experience in places like Microsoft Research or Bell Labs. I think more than 60% had a PhD”

Many in the AI ​​and ML field were surprised at the layoffs, given the high rating of the Probability team.

According to Nantas Nardelli, senior research scientist at climate technology AI firm Carbon Re, these were some of the best in the field, but not as well known as other researchers.

“They tend to produce work that may be less ostentatious but could become the backbone of ML products in 5-10 years,” he told VentureBeat in a LinkedIn post.

Their ML work, he explained, is “well applicable” to problems involving low or medium amount of data, high domain knowledge, and where it is important to estimate uncertainty. “This expertise is generally challenging to acquire, and fewer and fewer people are specializing in it these days,” he said.

Ethical AI layoffs on Twitter offer lessons to companies

Triveni Gandhi, responsible AI lead at data science and ML platform Dataiku, said she was not surprised by the ethical AI layoffs at Twitter.

“My instinctive reaction was, of course, that they are the first to be fired, because of the way the current leadership at Twitter has expressed what they think about questions of ethics, trust and security,” she told VentureBeat.

But as a responsible AI leader, she added, she also started thinking about what the news meant for her enterprise customers: “Are they going to think, well, we don’t need this stuff?”

However, she said she realized that the public reaction to the layoffs was an indication of how important and respected ethical AI has become.

“I think other companies are seeing this very public reduction of that particular team, and they’re thinking, ‘I don’t want to go down the same road,'” she said. “I don’t want to create mistrust among the consumers of my AI products.”

Among her clients, she added, she sees a “sense of determination” emerging from the news of AI’s ethical layoffs on Twitter. “They say, ‘We can be better than that, we’re going to make it possible’ [responsible AI teams] and start putting things into practice,” she said. “Like, let’s get away from the thought leadership on this and get rubber on the road.”