A conversation with Sussman on AI and asynchronous programming • Hacker News Board

Re: A conversation with Sussman on AI and asynchronous programming

#31

Post by edwardkmett » Thu, Oct 15, 2015, 9:32 AM UTC

nwf wrote: post #24

edwardkmett wrote: post #22

Talk about timely, I recently took the opportunity to unmothball the old propagators idea and have been running a bunch of ideas past Sussman's former student Alexey Radul on a fairly constant basis for the last 3 weeks. I started a project at https://github.com/ekmett/propagators which at least gets the basic execution of them right, and have been working on a larger project. Now, Sussman and Radul manage propagatio…

You may be interested in reading, if you haven't, our 2011 position paper on Dyna. (The 2012 paper on what I think you would call propagator networks was fun, but is nothing you don't know already, I'm sure.) http://dyna.org/wiki/index.php/Publications

Indeed. Our old discussions about "omega-continuous semiring homomorphisms" as the way to try to make something half-way between Dyna and the datalog bits I was working on have been very much present in my mind lately. =)

Re: A conversation with Sussman on AI and asynchronous programming

#32

Post by edwardkmett » Fri, Oct 16, 2015, 2:21 AM UTC

Animats wrote: post #26

That's interesting. It takes me back to the '80s, when the logic-based AI people were in charge. The "propagator" paper would be more interesting if there was a use case in the paper. The basic idea is to have a graph of computational elements, but insist that the elements be commutative, associative, and idempotent. Under those restrictions, you get the same answer despite race conditions, which leads to a simple mo…

Without the propagators themselves being monotone they don't have enough to ensure determinism. They also don't have a guarantee of termination. e.g. repeatedly taking Heron steps on a rational interval approximation to compute the sqrt of 2 will continue to ascend the lattice indefinitely, as it isn't complete.

That said, they have many of the elements of the solution. With monotonicity this starts to resemble the more recent work on Lasp, which is being used explicitly to tackle the domain you mention: (strong) eventual consistency: or Kuper's work on LVars, where she drops the idempotent condition fairly early on in the thesis to get closer to CmRDTs, but which then burdens reads in a way that makes them automatically monotone.

Re: A conversation with Sussman on AI and asynchronous programming

#33

Post by david_ar » Fri, Oct 16, 2015, 7:52 AM UTC

TeMPOraL wrote: post #27

panic wrote: post #20

Earlier quoted context omitted.

You can usually come up with an explanation for why you did something. You don't have to "get to the bottom" of your own thought processes to do this: you just need to be able to reconstruct the symbolic manipulation part. This seems like a good thing for an AI to be able to do -- especially a truly "hard" AI that you'd trust to run things at a high level.

Humans often come up with an explanation for why they did something that's simple, consistent and also completely untrue. They sometimes even believe their own explanations. We may run into the same problem with an AI here - it may be, purposefully or not, providing us with untrue explanations.

Indeed, people even offer explanations when tricked into believing they had chosen the opposite outcome to the one they actually picked [1].

[1] http://researchgate.net/publication/6745688

> participants fail to notice mismatches between their intended choice and the outcome they are presented with, while nevertheless offering introspectively derived reasons for why they chose the way they did

Re: A conversation with Sussman on AI and asynchronous programming

#34

Post by edwardkmett » Fri, Oct 16, 2015, 5:26 PM UTC

edwardkmett wrote: post #22

Talk about timely, I recently took the opportunity to unmothball the old propagators idea and have been running a bunch of ideas past Sussman's former student Alexey Radul on a fairly constant basis for the last 3 weeks. I started a project at https://github.com/ekmett/propagators which at least gets the basic execution of them right, and have been working on a larger project. Now, Sussman and Radul manage propagatio…

(Sorry, Lindsey, not Lindsay.)

Re: A conversation with Sussman on AI and asynchronous programming

#35

Post by deckard1 » Fri, Oct 16, 2015, 5:54 PM UTC

panic wrote: post #21

deckard1 wrote: post #15

Earlier quoted context omitted.

My thoughts as well. This is why I feel some people will never "let" AI happen. They want 100% certainty. Neural nets scare them. Despite the fact that real , human intelligence carries no certainty. You either have intelligence, and creativity, and uncertainty, and accidents . Or you have computers and rigid logic. The interesting thing there is, when a space rocket malfunctions, when a car wrecks, or the stock mark…

You either have intelligence, and creativity, and uncertainty, and accidents. Or you have computers and rigid logic. Why does it have to be such a strict dichotomy? What are the specific tradeoffs being made between creativity and rigid logic? Is there really no way to pick something in the middle? The interesting thing there is, when a space rocket malfunctions, when a car wrecks, or the stock market goes crazy... a…

My point wasn't to start a philosophical debate on the term "intelligence". My point was, the intelligence that was recognize in humans (our only real guidepost at all, in terms of "AI") is closer to neural nets than the symbolic and expert systems of the AI winter. And that human intelligence is fuzzy and you have to accept the good and the bad.