Meta VP Jerome Pesenti Leaves as Company Adopts Decentralized AI

Source: pixabay

Meta’s VP for AI, Jerome Pesenti, is leaving the company middle of this month. The announcement was made in a blog post by Meta’s CTO, Andrew Bosworth. The post focused on the company’s move towards decentralized AI and how it intends to merge Pesenti’s AI team with other product teams. Pesenti has been leading Facebook’s parent company’s AI work for four years and helped steer Meta’s direction from a centralized AI approach to distributing ownership of the company’s AI systems back to specific product teams.

“We believe that this will accelerate the adoption of important new technology across the company while allowing us to continue to push the envelope. The teams tasked with driving AI advancements and best practices into the products they support will be known as AI Innovation Centers,” Bosworth said.

The reorganization includes the merging of the Responsible AI group with the company’s Social Impact team. The AI for Product team will move to the engineering team while the AI4AR team will join the XR team in Reality Labs. The company’s AI research team, FAIR, will become a new pillar within Reality Labs Research. Also part of the reorganization is creating a new cross-functional AI leadership team led by Joelle Pineau, the lead of Meta’s AI research lab.

Accelerated Decision-Making with Decentralized AI

To say AI is huge is an understatement. AI is practically everywhere now, from the apps we use to order food delivery to major industries like healthcare, education, and manufacturing where AI is used to run entire supply chains.

Meta said it itself—AI is at the heart of the work they do. It’s also the technology that will power what the company envisions for its Metaverse—complete with advanced VR/AR lenses that will supposedly blur the lines between the real world and the virtual.

The move to decentralized AI is one that the company hopes will give it the edge over its competitors. “Whether for personalization or protection, whether to improve existing services or create entirely novel ones, our future depends on our ability to leverage the newest AI technology at scale. To support this mission, today we are announcing a new decentralized organizational structure for Meta AI,” explained Bosworth.

Centralized AI is like closed source software or propriety software. For decades, companies tried to limit ownership of software products to have an advantage over intellectual property rights. Fast forward to now, we know the advantages of open-source where you decentralize the code so that talented engineers all over the world can contribute and continuously improve the original product. Many SaaS software products, for example, are designed and built on open-source platforms and tools and have APIs that allow users to modify them according to their requirements; thus providing more value to users.

We can apply the same concepts to decentralized AI. One of the advantages is accelerated decision-making. Meta can leverage accelerated decision-making brought about by data available across multiple AI agents versus a single, centralized AI agent. In the real world, data can’t really be centralized in one data center mainly due to privacy reasons. Meta’s decentralized AI approach can improve the speed of AI decision-making by making the most out of all the generated data available to it.

Decentralized AI is indeed creating a lot of excitement in Applied AI. Its potential benefits to training and optimizing AI models can have economic and societal impacts that many experts are still debating on. In the meantime, Meta has clearly made its decision and we’ll just have to wait and see how it will impact the future of AI.