|
|
|
|
|
|
by ilaksh
3120 days ago
|
|
|
I don't think we need a genius or a new paradigm for AGI. I think the main ideas of a few different approaches to do it are already out there. There are a lot of tough problems to solve but to me it looks like it's a matter of hard work rather than any totally new approach. Look up for example Yann LeCun's AGI presentation. Now see projects like Ogmaneo or recent Deep Mind papers that do fast online learning or have ways to work around catastrophic forgetting and other issues raised by LeCun. I believe that one straightforward way to get there is to continue to apply neural network research towards attempting to mimic animals. And there are a bunch of people doing that and making progress. |
|
|
That is harder than you might think. For example, the simplest animal with a nervous system, C. elegans worm has been simulated "only 20 to 30 percent of the way towards where we need to get" according to one of the OpenWorm project leaders. That's 302 neurons total.
Using a computer analogy, it's like struggling to build an abacus when our ultimate goal is to build a Core i7 processor.
rather than any totally new approach
It's not clear to me that we have any approach to achieve AGI. People like Ben Goertzel have worked on that for decades without much to show for it. Current deep learning methods have little to do with AGI, they focus on very narrow applications, by design.