Conversation

Notices

1. Embed this notice
   Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:28:19 JST Paul Cantrell

   • Richard McElreath 🦔

   This delightful brain-poke from @rlmcelreath is worth considering in earnest:
   https://nerdculture.de/@rlmcelreath/110938545075497824

   There is no single consensus definition for “intelligence” in living organisms! How the heck can we define •artificial• intelligence if can’t even define •natural• intelligence?

   My definition: AI is anything that

   1. humans have consistently done better than computers in the recent past

   2. and now computers are doing (or approximating)

   3. and our culture allows us to be impressed by it.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:33:42 JST Paul Cantrell

     in reply to

     Arithmetic? Not AI because computers have always done it better than humans.

     Winning checkers? Was AI, but now it’s just algorithms because computers have been good at it for too long. Same for resolving logic problems, parsing well-formed English sentences, soon for chess.

     Sweating, sneezing, farting? Not AI when a computer does it, because not culturally valued. But standing, walking? AI.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:38:31 JST Paul Cantrell

     in reply to

     Automated theorem provers make a nice example. I heard a talk a while back about speeding up Coq by having an LLM feed it a better set of starting points for its solution search.

     In that talk, the speaker referred to this as “using AI to assist a theorem prover.” Time was that a theorem prover •itself• was AI! But now •that• is just algorithms; it’s only the LLM that counts as “AI.” The line is always shifting.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:40:32 JST Paul Cantrell

     in reply to

     TL;DR “Artificial intelligence” is a term whose meaning depends entirely on historical and social context; it has no precise or stable meaning at all. AI is in the eye of the beholder. Thank you for coming to my TED thread.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:52:42 JST Paul Cantrell

     in reply to

     • ambersz

     @ambersz Well, if a computer does it, I’m impressed whatever we call it

   • Embed this notice
     ambersz@mastodon.gamedev.place's status on Tuesday, 29-Aug-2023 01:52:44 JST ambersz

     in reply to

     @inthehands How about farting "in a creative, musical, or amusing manner"? 🤭
     https://en.wikipedia.org/wiki/Flatulist

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 01:54:45 JST Paul Cantrell

     in reply to

     • Zeewater

     @Rana
     “It's nothing new but since we haven't seen that particular entity do it before, some will see it as impressive”

     I think that’s apt, and I think we’re right to be impressed: tasks writing and tying shoes are complex, it’s amazing that •any• entity can do them, growth is exciting, and there’s always the promise of further learning.

     The question of how the learning does & doesn’t generalize is tricky for both humans and machines.

   • Embed this notice
     Zeewater (rana@mastodon.nl)'s status on Tuesday, 29-Aug-2023 01:54:47 JST Zeewater

     in reply to

     @inthehands Maybe it's a bit like being impressed that a child learned to write a name or tie shoes
     It's nothing new but since we haven't seen that particular entity do it before, some will see it as impressive

     Though while a child grasps the basic understanding of doing something, an AI utterly fails when put in a different situation

     The difference of citing something vs knowing why and being able to repurpose that knowledge

     AI are 1trick ponies

     (Very simplified ofc, I'm no AI expert)

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 03:14:47 JST Paul Cantrell

     in reply to

     • Righteous Hazard
     • Richard McElreath 🦔

     @bertwells @rlmcelreath This sounds like the “Righteous Hazard Test” and I’m all for it

   • Embed this notice
     Righteous Hazard (bertwells@mathstodon.xyz)'s status on Tuesday, 29-Aug-2023 03:14:49 JST Righteous Hazard

     in reply to

     • Richard McElreath 🦔

     @inthehands @rlmcelreath Perhaps you could propose a Cantrell AI Test: a machine possesses an AI in a field if it can convince an expert interviewer in that field to beat the shit out of whoever programmed it.

   • Embed this notice
     Paul Cantrell (inthehands@hachyderm.io)'s status on Tuesday, 29-Aug-2023 03:18:57 JST Paul Cantrell

     in reply to

     • Zeewater

     @Rana I think it’s an open question how different we humans really are in our ability to adapt, or how we can tell. Regardless, the word “intelligence” is culturally loaded without having a clear meaning, and that confuses the issue substantially.

     Every time we try to draw a clear, simple, empirical line around human-like thought, machines cross it — yet every time we’ve thought computers were on the cusp of human-like thought, we’ve been wrong.

   • Embed this notice
     Zeewater (rana@mastodon.nl)'s status on Tuesday, 29-Aug-2023 03:19:00 JST Zeewater

     in reply to

     @inthehands I hope one day AI technology will get to the point where it can use previously acquired knowledge to improvise and show creativity

     The AIs I've seen so far either are just really good at doing one specific thing but utterly failing in a similar yet different task (think playing videogames) or function somewhat like social chameleon where it just tells us what it calculated we want to get back from it (think chat bots, art generation, npcs in games, etc)