Photo credit: venturebeat.com
François Chollet, renowned for his work as a former Google engineer and creator of the popular Python deep learning framework Keras, has announced the launch of a new AI research and science lab named Ndea. He co-founded this initiative with Mike Knoop, who is best known as a co-creator of Zapier.
In a statement on Ndea’s new website, the founders articulated their vision of integrating advanced pattern recognition, rooted in deep learning, with formal reasoning techniques they refer to as “guided program synthesis.”
Their objective is to transition AI technology towards a state where systems can adapt and innovate independently—moving beyond the limitations of existing, task-specific applications. This approach aims to advance the pursuit of artificial general intelligence (AGI), which many in the AI community define as the capability of machines to outperform humans across a majority of valuable cognitive and economic tasks. They emphasize the need for computational systems that can not only apply existing knowledge but also engage with novel problems and create innovative solutions. As they note:
“We need computers that can pose problems and explore new territory, not just apply known solutions. We need computers that can innovate. The path to AGI is not through incremental improvements to existing methods.”
At this stage, the founders have not disclosed whether they have secured external funding or if the project is being supported solely by their own financial resources.
This announcement follows news earlier this year about Ilya Sutskever, a former co-founder and chief scientist at OpenAI, who launched a startup aimed at developing “Safe Superintelligence” with substantial private funding of $1 billion after a contentious internal conflict with fellow co-founder Sam Altman.
Beyond Deep Learning
Chollet and Knoop argue that, while current deep learning technologies demonstrate impressive capabilities, they are fundamentally limited by their dependence on extensive datasets and the challenge of efficiently adapting to new tasks. They believe that program synthesis represents a powerful solution to these challenges.
Unlike conventional deep learning methods, which interpolate between data points, program synthesis seeks discrete programs that can clarify data—offering improved generalization while requiring significantly fewer datasets. The blending of deep learning’s intuitive strengths with the structured reasoning of program synthesis could pave the way for transformative advances in AI research.
“Ndea’s mission is to operationalize AGI to realize unprecedented scientific progress for the benefit of all current and future generations,” the founders stated.
Building a “Factory for Scientific Advancement”
Ndea is not solely focused on achieving AGI; it aspires to function as a “factory for rapid scientific advancement,” equipped to tackle both known and emerging challenges. The lab envisions itself as a pivotal force in fostering scientific discovery, addressing topics from autonomous vehicle development to sustainable energy solutions.
Chollet expressed that their research could lead to significant breakthroughs, stating on a social media platform: “If we’re successful, we won’t stop at AI. With this technology, we want to address every scientific problem it can tackle. We view accelerating scientific progress as one of AI’s most exciting potential applications.”
According to Chollet, the key to this scientific advancement lies in creating AI that can learn as effectively as humans without encountering systemic barriers.
While he recognizes the unpredictability of achieving success, Chollet underscores the importance of striving for this lofty objective, affirming: “We believe we have a small but real chance of achieving a breakthrough—creating AI that can continually improve with no bottlenecks in sight.”
A New Research Focus for AGI
Program synthesis is a relatively nascent field within AI research, with Chollet comparing its current trajectory to the state of deep learning in 2012. Yet, as its potential becomes increasingly acknowledged by leading AI labs, it remains largely seen as merely one component of what is necessary for AGI.
Ndea, however, positions program synthesis as an equally essential element to deep learning, making it central to their research strategy. The lab is actively recruiting a diverse and talented team of researchers and engineers to establish what it aims to be the most proficient program synthesis team worldwide.
Operating as a fully remote organization, Ndea is seeking candidates with a strong technical background, particularly those adept at translating complex mathematical concepts into practical coding solutions.
Founders with Strong Track Records
Chollet and Knoop bring a wealth of experience to their new venture. During his tenure at Google, Chollet was involved in vital research concerning deep learning and AI strategies, offering insights into existing models’ limitations and areas for enhancement. He has contributed significantly not only to Keras but also to the ARC-AGI benchmark, a notable standard for assessing progress in AGI development.
Chollet is also recognized for his book Deep Learning with Python and has been celebrated as one of Time’s “100 Most Influential People in AI.” Knoop, on the other hand, played a pivotal role in founding Zapier, where he was instrumental in its engineering and product development, including the integration of AI technologies and the establishment of best practices for managing globally distributed teams. Both founders are co-creators of the ARC Prize Foundation, dedicated to fostering open research in AGI.
Future Visions Rooted in Ancient Tradition
Ndea derives its name from the ancient Greek concepts of ennoia (intuitive understanding) and dianoia (logical reasoning), symbolizing the lab’s ambition to unite deep learning with program synthesis. Through the operationalization of AGI, Ndea aims to compress the timeline of scientific progress, potentially achieving milestones in mere decades instead of centuries.
While the pursuit of AGI is fraught with uncertainties and challenges, Chollet and Knoop remain hopeful regarding their strategy. They envision AGI as a gateway to solving critical global issues and unlocking new realms of discovery.
Source
venturebeat.com