Machine learning trains computers through trial and error with massive datasets. Humans, on the other hand, are born with innate sensibilities that help them learn rapidly about the world around them. Now, researchers are trying to get AI to learn and think more like babies. Right now, AI has been relying on neural nets that must use big data to build and create associations. Babies don’t use big data, but they have the incredible ability to make associations with only a few experiences.
It turns out, babies or children learn a lot like a combination of machine learning and simple rules-based programs. AI right now is good at one or the other, but not a combination like humans. Researchers believe that if AI masters learning like a child, it will be able to teach robots things like common sense. Early experiments with child-like AI learning has outperformed standard machine learning that relies on big data. With this new technique, robots will even be able to pass those pesky CAPTCHA tests.
Researchers believe that robots of the future will be more human-like because they will start as babies and learn like children. Will researchers ever be able to replicate the imagination of child? Add your voice to the discussion on Vialogues.
Excerpts from the discussion:
@02:36 Rebecca Sullivan: While AI will probably be able to outperform humans in logical, algorithmic knowledge, I don't think AI will ever be able to interact with the world the way children do, since one of the key motivations of children's exploration and imagination is wonder. I'm not sure how the trusting joy of a child could be programed into AI.
@02:38 RuthS: I know there must be a reason, but why should we want robots that experience the world the way we do? I wish we could just consider robots as robots rather than trying to make them as human-like as possible.