François Chollet and Lex Fridman disagree about the Turing Test

Full episode with François Chollet (Aug 2020): • François Chollet: Meas...
Clips channel (Lex Clips): / lexclips
Main channel (Lex Fridman): / lexfridman
(more links below)
Podcast full episodes playlist:
• Lex Fridman Podcast
Podcasts clips playlist:
• Lex Fridman Podcast Clips
Podcast website:
lexfridman.com/ai
Podcast on Apple Podcasts (iTunes):
apple.co/2lwqZIr
Podcast on Spotify:
spoti.fi/2nEwCF8
Podcast RSS:
lexfridman.com/category/ai/feed/
François Chollet is an AI researcher at Google and creator of Keras.
Subscribe to this KZbin channel or connect on:
- Twitter: / lexfridman
- LinkedIn: / lexfridman
- Facebook: / lexfridman
- Instagram: / lexfridman
- Medium: / lexfridman
- Support on Patreon: / lexfridman

[@el_chivo99]

definitely agree with Chollet here. One of the primary ideas in Collet's paper is that we need to rely less on testing AI's skills and more on testing AI's ability to learn skills, adapt, and generalize beyond problems that it *or its creator* are aware of. Designing a model to pass the turing test is no different than designing a model to play chess at a world-class level. Are both challenging? Yes. Do either demonstrate AI intelligence? No - they only display the intelligence of the creators.

[@greengoblin9567]

Why does lex keep on thinking that gpt has close to human levdl intelligence

[@BMoser-bv6kn]

I don't understand your assertion here. Passing the Turing test requires comprehension, spatial awareness, long term memory, and at least some rudimentary ability to plan ahead. It is completely different from learning to play chess well - it literally requires the creation of a system able to learn, play, and administer ANY arbitrary game from being told its ruleset in normal conversation. And that's just one aspect of what it takes to pass off as a human being. The enormity of that problem space encompasses essentially "be a human". Is this not obvious? Language is thought.

[@el_chivo99]

@@BMoser-bv6kn Like I said, passing the Turing test is difficult, no doubt. But as a **measure** of intelligence it is poor since the test requires a subjective panel of evaluators who are all biased. In his paper Chollet advocates for empirical measures and tests, and poses the ARC as a good way to quantify intelligence.

[@artukikemty]

@@BMoser-bv6kn learning to play Chess requieres the same thing you said. But it could be a good example of true intelligence, if done without human intervention. Can GPT-4 learn to play chess by it own?

[@artukikemty]

@@BMoser-bv6kn Chess and language are probabilistic, intelligence is not.

[@sagittariusa2008]

Ex Machina is a must watch if you need to stoke that fire.

[@reck0n3r]

Battlestar Galactica too.

[@justjohn193]

That movie is a masterpiece

[@Consural]

The problem with "tricks" being allowed in the Turing Test is: If the AI was the one doing the tricks out of its own will, it would be absolutely acceptable and conducive to progress. However, since we don't have an AI with "its own will" in the first place, it has turned into humans creating systems that can trick the human judges(who more often than not lack the proper capacity to test them.) Which doesn't help progress the field of AI at all, which is the primary concern of Mr. Chollet and he's absolutely correct.

[@4allmo]

The greatest trick of the AI was to make mankind believe there is no AI.

[@z4zuse]

Humans judging AI is similar to a horse judging a car

[@OrochiShaka]

People think AI will rule the world and be invincible, but I tend to think that a "perfect being" like AI cannot come from an imperfect being. So it will have flaws too.

[@samdelahunty1506]

Bender would bribe Fry in the Turing test. I think this discussion unfortunately completely missed the point of Turing paper and Turing Test. I think the points Chomsky made on his podcast went straight over his head. Turing was extremely smart and didn't naïvely set a low bar for measurement, and his paper on the subject wasn't intended for engineers at all. It was a philosophical look at what it is to say a machine can think. Which Turing himself says "is too meaningless to deserve discussion". The same way if you want to say a submarine can swim then it can swim. There is also more interestingly Jaron Lanier's take on it, the fact that the Turing test is a quest for a being to be accepted as one, maybe that's how he felt as a homosexual being treated as a non human.

[@Flamian300]

can't wait to ask "how did you lose your virginity" to these Ai. toughest turing test question :)

[@edwardlewis1963]

I did not.

[@brandongarcia4219]

edward lewis 😂

[@ascensionblade]

@@edwardlewis1963 damnit, ok, we'll let you leave the facility, human.

[@sagittariusa2008]

All AI needs is an internet feed to realize what we are and what we are not. Garbage in garbage out whether its FIFO or LIFO

[@thadfreebourn5138]

Judgement...AI will work when it creates with none; who decides what's intelligent, determines standards...we are in preschool, evolutionarily

[@AnyFactor]

Disagrees with chomsky Disagrees with Lex Do I see a pattern here?

[@kristermister4791]

If your goal is to trick a panel of human judges then yes, you might be incentivised to reply on tricks, but tricks are easily sussed out, so maybe not?

[@oskarjankowski5709]

I don't think it's anything like David Copperfield making an illusion of teleportation. What credibility does a panel of judges who don't closely inspect the method have? If you substitute a computer with a human that misses the entire point of the test. As for all else, I don't have an issue with what Chollet says.

[@chychywoohoo]

Lovelace test is better

[@Rick1234567S]

Perhaps I can help. I have invented a new Turing test. Please spread this test to determine if a respondent is an AI or a real person. I have two questions that will determine if you are a robot AKA a Smithbot or a real human. "LPI is managed by the Universities Space Research Association (USRA), an independent non-profit organization that offers an avenue for government and industry to engage the expertise of the academic community with USRA’s technical leadership, innovative R&D, operational excellence, management of premier facilities, and education programs to advance space- and aeronautics-related sciences and exploration. USRA works across disciplines including biomedicine, planetary science, astrophysics, and engineering and integrates those competencies into applications ranging from fundamental research to facility management and operations. LPI is operated by USRA under a cooperative agreement with the Science Mission Directorate of the National Aeronautics and Space Administration." www.lpi.usra.edu/publications/slidesets/clem2nd/slide_26.html So two questions. First that graph was made by people who have PhDs, 30 of them on staff or similar. So that is an accurate rendition of the data not done by a 3 year old it was done by scientists with PhDs. 1) What are the red lines within the dark circular area in the titanium distribution map? 2) What are the light blue radial lines around the dark area in the titanium distribution map and the iron distribution map?

[@artukikemty]

Shit-GPT

[@johnbouttell5827]

Science is magic. Discuss.