At times, it is significant to walk the road not taken to avenge new theories and break the barriers to think out of the box. So proves the works of Yann LeCun, the Chief Scientist at Meta's AI lab and one of the most influential AI researchers in the world. 

A recent study published by the MIT Technology Review looks back at the contributions of Yann LeCun. LeCun has been "trying to give machines a basic grasp of how the world works—a kind of common sense—by training neural networks to predict what was going to happen next in video clips of everyday events," writes Will Douglas Heaven and Melissa Heikkilä. However, he hit a wall, as guessing future frames of a video pixel by pixel was a complex task. 

But the story did not end here. After months of hard work, Yann LeCun found the missing pieces of his puzzle and thus was born a "bold new vision for the next generation of AI". A draft of his thoughts was shared with the MIT Technology Review, where he sketched out an approach that would give the machines the power to think like a man and navigate the world. 

Old school meets the future

While researchers worldwide attempted to develop new technologies, Yann LeCun detoured. He stepped aside from the hottest trends in machine learning, opening the dusted conventional ideas that have gone out of fashion. According to Will and Melissa, proposals could be the first steps on the path to building machines, what many call 'artificial general intelligence or AGI'. Here is a catch - LeCun does not know how to build what he describes. According to the Technology Review, his approach may raise more questions than answers. 

The novel approach revolves around the neural network and focuses only on those features in a scene relevant to the task at hand. LeCun proposes pairing these core networks with another called the Configurator. It will determine what level of detail is required and tweak the overall system accordingly. 

"For LeCun, AGI will be a part of how we interact with future tech. His vision is colored by that of his employer Meta, which is pushing a virtual-reality metaverse", states Will and Melissa in their study.

Significance of Common Sense 

LeCun had remarked, "Getting machines to behave like humans and animals have been the quest of my life". He thinks that animal brains run a kind of simulation of the world- this is the way that animals make a good grip on the world around them. For instance, seeing a dropped ball fall a handful of times is enough to give a child a sense of how gravity works. Likewise, knowing the world as three-dimensional or that things do not disappear after going out of sight are all instances where common sense comes to play. It aids in classifying the possible with the impossible, foreseeing the consequences of actions, making plans and ignoring the irrelevant. 

LeCun views common sense as a vision of intelligence, which is why he and a team of researchers are utilizing videos to train their models. However, their attempts are no piece of cake, which brings a neural network that focuses only on the relevant aspects of the world. 

Building the model 

MIT Technology Review states that LeCun has built an early version of a world model that can do essential object recognition, and he is now working on training to make predictions. But the working of the Configurator is still a mystery. He still believes in cooking up a new recipe to make this work. 

In LeCun's imagination, the world model and Configurator are two key pieces in a larger system known as cognitive architecture. It also includes a perception model that senses the world and a model that uses rewards to motivate the AI to explore. Each neural network is roughly analogous to parts of the brain, according to LeCun. Even though concepts of cognitive architecture inspired by the brain have been around for decades, his ideas own different level of detail. For LeCun, revisiting out-of-fashion ideas is essential because he believes that two dominant approaches in modern AI are dead ends. 

Experts Speak 

According to the MIT study, the research world has mixed opinions about Yann LeCun's theories. David Silver at DeepMind, whose approach LeCun has ignored, stated, "It's an exciting new proposal for how a world model could be represented and learned". Melanie Mitchell, an AI researcher at the Santa Fe Institute, is excited about LeCun's approach. However, Google Brain's Natasha Jaques thinks language models should still play a role. On the other hand, Abhishek Gupta, the Montreal AI Ethics Institute founder, calls LeCun's proposal ideas practical application. 

Read the original story here.