|
smoka is by no means the only one to make the ontological-privilege argument against AI roger penrose thinks there's some quantum element to both intelligence and consciousness that means you're not going to get either of those from a Turing machine personally I don't buy it: even if there are quantum elements involved, simulation of a biological system would just get more difficult, not be impossible, though depending on just how difficult that could be a distinction without a difference
|
# ¿ Sep 8, 2017 09:29 |
|
|
# ¿ May 14, 2024 05:31 |
|
what disappoints me is that there isn't anywhere near the same focus on symbolic reasoning that there used to be symbolic reasoning systems like Cyc are actually rather good now and do actually do the "learn how to learn new things" bit however they don't result in billions of dollars of sales of specialized vector hardware, and can be hard to parallelize and use at a large scale too, so they haven't had the same level of investment that neural nets have been getting in this latest resurgence in interest in AI
|
# ¿ Sep 8, 2017 09:33 |
|
it's worth reading Lenat's "The Nature of Heuristics" to see what the symbolic-reasoning people were up to also "Theory Formation by Heuristic Search" and "Eurisko: A Program That Learns New Heuristics and Domain Concepts" not a neural net in the bunch, just a whole lot of frames in a custom language atop Lisp eschaton fucked around with this message at 08:36 on Sep 9, 2017 |
# ¿ Sep 9, 2017 08:27 |
|
echinopsis posted:we install kill switches and also make sure they cant gently caress and reproduce because of course a super intelligence wouldn't be able to subvert its own programming and disable any software kill switch needs to be a locally-triggered physical power cut, carried out by someone who can't be persuaded not to
|
# ¿ Sep 10, 2017 12:17 |
|
Shinku ABOOKEN posted:no but seriously is there any effort to make an ai that generates programs, possible better ais? read the links I posted earlier in the thread, they're about an old design for what's effectively a self-improving AI and there's not much difference between an AI improving itself or improving a copy of itself
|
# ¿ Sep 12, 2017 08:27 |