In this episode, Byron talks about transcending transfer learning. For more on Artificial Intelligence: https://voicesinai.com https://gigaom.com https://byronreese.com https://ift.tt/31WftGA... Transcript: In our quest to build robots with artificial intelligence, we face five distinct challenges, and I've gone over four of them in prior videos. The first was a robot being able to see and recognize things, The second one was to contextualize them, And the third one would be to contextualize moving things the way life moves, And the fourth is applying that knowledge to other disciplines, And then there is the fifth. And that is that robots cannot improvise. Everyone - every human at every skill level - can improvise in a way far beyond any machine. If you try to open a door and the handle breaks off in your hand, you don't just stand there immobilized and unable to comprehend a world that you never thought about, a world where door knobs don't turn, they break off in your hand. No, you try to figure how to get the door open. You stick it back in and try to get it to engage with the mechanism and do all of that. If you lock yourself out of your house, you’re resourceful, and you improvise and figure out a way to get in. If a gust of wind blows the umbrella away that you’re carrying, you go after it without ever having been taught how to retrieve a blowing umbrella. So even if a computer can see and understand what it sees and derive context from it, and do that with objects in motion, and can use that in other domains, it still isn't creative. We don't just passively perceive the world the way our typical AI does, we react to it in a way that transcends transfer learning.
No comments:
Post a Comment