While much of Meta is working on Mark Zuckerberg’s wish for the metaverse, the company’s head of artificial intelligence is quietly building a “roadmap” for “autonomous” AI.
Noted computer scientist Yann LeCun recently published an article in which he describes that he describes the lack of “common sense” in the current AI effort and sets the stage for future iterations that “learn as efficiently as humans and animals” as they become more and more popular more and more self-sufficient.
Common sense, as described by LeCun, is a set of “models of the world” that allow humans and animals to predict whether events are probable or improbable, plausible or implausible, possible or impossible.
“An autonomous driving system for cars may require thousands of reinforcement learning trials to figure out that driving too fast around a bend will cause something bad and learning to slow down to avoid skidding,” the researcher wrote.
“On the other hand, humans can use their intimate knowledge of intuitive physics to predict these outcomes and largely avoid fatal actions when learning a new skill. »
To bridge the gap between the many iterations of trial and error required to train neural networks and the “intuitive” nature of organic knowledge, LeCun proposes revamping the training methodology of algorithms to learn more efficiently and thus develop a composition of common ground that humans take for granted.
While that may not sound so appealing, something akin to intuition will likely be needed to take the AI from its current state – and admittedly impressive, albeit in narrower realms – to something more. close to human intelligence.
“It’s a practical problem, because we really want machines with common sense,” LeCun said during a late September speech in Berkeley, US. “We want self-driving cars, home robots, smart virtual assistants. »
As a scientist, Meta’s AI manager’s next-generation algorithm training architecture involves several moving parts, such as a system that replicates short-term memory, another that teaches self-criticism of neural networks and the implementation of a “configurator” module, which synthesizes all the inputs into useful information.
Together, these components are meant to help AI replicate the processes of the human mind – a prospect as fascinating as it is terrifying.
The fact that Meta’s senior AI researcher is silently circulating an article about how to turn AI into autonomous “thinkers” is in itself a hugely intriguing story, and given what he hopes to accomplish, it might even be very beneficial for the tech giant. Technology.
Featured Image: metamorworks/
Have you watched the new videos on Youtube digital look? Subscribe to the channel!