That does seem to be a common belief. Whereas I believe that if we could achieve artificial mouse intelligence (AMI), it would be a relatively smaller step to AHI. (And I don't think the term AGI even makes sense).https://twitter.com/nicholdav/status/1203884280050540544 …
-
-
Replying to @TonyZador
We have much better knowledge of human cognition than mouse cognition. (Though better understanding of mouse neural circuits). With knowledge of target, AHI may be easier to achieve than AMI (unless you limit definition of mouse intelligence to what we have already discovered).
1 reply 0 retweets 4 likes -
Replying to @AndrewHires
The symbolic approach to AI, which dominated for the first few decades, drew on human cog and went straight for AHI. I dont know that we've learned much about human cognition since then that'll help I bet the path to AHI goes through AMI (and neural networks/neural circuits)
1 reply 0 retweets 10 likes -
Replying to @TonyZador
Based on the GPT-2 output I seen lately. Seems like there has been much more progress on natural language processing than generative mouse vocalizations. This may be a recurring theme.
2 replies 0 retweets 1 like -
Replying to @AndrewHires @TonyZador
You probably want to go for mammal brain core architecture, while all the while exposing it to and testing it on human-child-like uptake of human behavior and cultural abstractions... mouse tasks may not put a cortex through hard enough challenges to be discriminative of models.
1 reply 0 retweets 0 likes -
(At least the “mouse tasks” we most easily create in silico. As you mention, really hitting all the complex full body motor control and vocalization might indeed be a strong enough selection pressure. A step: https://arxiv.org/abs/1911.09451 )
1 reply 0 retweets 0 likes
(I mean this more as a rhetorical question than an assertion.)
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.