Creating Human level AI – Yoshua Bengio at Asilomar Conference on Beneficial AI 2017

It’s thought-provoking that Bengio’s presentation is about creating human level AI (AGI) and yet someone felt it obligatory to ask a query at the end of the video that illustrate how naive the field is regarding AGI. Person asked about the difficult of generating enough data for the machines to learn. Do we even realize how clueless the question sounds within the context of human-level AI?

Think what data do we generate for human babies to learn, it’s proven that babies can catch languages both native and non-native till 6 months and then ignore the non-native language. It’s a simple fact deduced, but its lot complicated for us to create. Just think about what all technologies are needed or demonstrated in the fact, learning for sure, learning with noise elimination without any supervision and many more, remember the babies can’t even talk at this point of time. This bring us to another problem: unsupervised learning, which Bengio correctly identifies are a requirement for human-level AI. Humans have no problem learning to understand the world without symbolic supervision (labels).

Are we waiting for mainstream AI to figure out AGI anytime soon? Waiting is not the option for doers. The mainstream AI community is still doing symbolic AI and completely ignoring advances in other scientific fields but over the decades there are folks who understands the importance of field cognitive science and its emphasis of a collaborative approach to the problem. It’s very important for the different fields involved in AI to collaborate with each other and take steps hand in hand. It’s a must watch video guys.