Comments by "Dazzer1234567" (@Dazzer1234567) on "Michael I. Jordan: Machine Learning, Recommender Systems, and Future of AI | Lex Fridman Podcast #74" video.

  1. The first real AGI won't be like a secretary or a bot that can call a restaurant, it will be like a 2 or 3 year old child that can convince you that it has the intelligence and learning capacity of a real  2 or 3 year old child. Anyone who has a lot of experience with children of this age knows that children sometimes say things that make absolutely no sense, leaving you scratching your head. Then you realize that they've conflated two unconnected things, and you laugh about it. This is probably a necessary condition of human learning, searching out the valid connections. So it will also be with the first AGIs......... they'll also make these childlike mistakes. But they'll also be able to recognize an elephant after seeing 50 pictures of elephants without needing to see a data set of 100,000 elephant pictures. These will be the first really convincing AGIs, and from there they'll move up the intelligence scale of children, to adults, and beyond. So the first real AGIs will be useless to us, except for applications like children's toys which make you scratch your head and laugh!
    5
  2.  @TheReferrer72  Hubris perhaps, but until now we know of only one route to AGI. Until these other routes you speak of have been demonstrated, i'll place my money on the child model as an important benchmark!
    1