Where Asimov Was Wrong: The DUMB AI fallacy

[@tedarcher9120]

It's called orthogonality thesis. Being very intelligent does not have anything to do with sharing our moral values

[@BookFurnace]

Thanks, I wasn't aware of this term!

[@andrewsuryali8540]

Asimov's AIs are not dumb in the way you describe at all, and Harlan Ellison actually did a good job of explaining this by writing his I, Robot movie script. Ellison imagines a progression of AI evolution under Susan Calvin's guidance from the really dumb robots at the start of the space exploration age to Stephen Byerley, the perfectly human android who understands humans so well he started questioning the activities and motives of his much more powerful Machine brethrens - the ones you call "dumb". Basically, there are always two paths of AI progression in Asimov stories. One path always goes to perfect rationality, which results in orthogonal intelligence development creating very intelligent and manipulative AI like the Machines and R. Daneel Olivaw. The other path always goes to deeper understanding of human morality and emotions to the point of direct sympathy like Byerley and Dors Venabili (who in the end was even able to experience the robotic equivalent of love). In between the two paths there are also those who straddle both throughout their "lives" and choose one path over the other in the end. Andrew Martin chose the humanist path, R. Giskard Reventlov chose the rationalist path. Asimov imagined BOTH possibilities without ever really deciding which path AI progression should take.

[@BookFurnace]

I agree that I might have characterized his AIs a bit too briadly. I think I was speaking about the most common AIs in his works, but there were indeed exceptions. I also actually thought that Giskard and Olivaw would also be a bit in the exception side. Maybe it's time to reread some of his works also :)

[@markdeffebach8112]

Thanks for saving me the trouble of having to point this out.

[@RichardRoy2]

It's been a long time since I've read Asimov, but I don't recall his books as something that highlighted dumb AI so much as nuance in interpretation. In the case of Rainman, I thought it reflected a difference in neurology, not a superiority in mathematics. His sensory perception was very different than what we're used to. How an AI may interpret primacies is programmed into them. And this can be a difficult element to interpret when it comes to how we differentiate humans from AIs because we are programmed through our experiences, and those experiences are along a continuum that is fed by many interfaces compared to that of an AI. The complexity of the algorithms an AI is endowed with are going to have a tremendous effect on how it is able to "grow." The continuum of my experiences include what lessons were taken up when I was a baby, but, from what I understand of it, memory is something that develops later, and that unifying moment when I transform from a sensory response machine into an "I" is something I can't remember. This whole experiment with AI is a kind of mimicry of what we come to understand about ourselves. A very fascinating journey. I don't want to speak much about whether Asimov got it wrong, because I may have been focused on other elements when I did read them, and I may have missed that particular aspect of it. Interesting interpretation either way. Thank you.

[@zoebaggins90]

So, what the AI really lacks is not only true intelligence, but also morals.

[@nosuchthing8]

Exactly. Look at the nazis. Some were very smart, with their v2 rockets etc. But also lacked morals vis a vis the holocaust.

[@glenn_r_frank_author]

I wonder if Asimov didn't do the "dumb AI" thing because he thought it was a realistic path in robotics, but because it made a good contrast to humanity. In other words, he did it for story reasons. Much like Mr. Data in Star Trek, and to some degree the Original series Mr. Spock who were logically and mathematically superior but had difficulties understanding their human shipmates and their emotional ways.

[@BookFurnace]

That's an interesting idea. I think Asimov was, like most writers, preoccupied with the idea of what makes a human human, and maybe he wrote his robots the way he did to highlight his beliefs on this topic. I personally do feel that like Kasparov refusing to believe that AIs would ever master chess, Asimov did feel that AIs would never master being human, but we'll never know for sure.

[@Kikilang60]

With some AI's they had the question of two children arguing about something minor, like butterflies. They disagree to the point of not becoming friends. Later as adults one of the grown-up children finds a way to solve world hunger. They ask the AI, "Will she share the information with the other person even if they had a fight, and don't like each other?" The AI says they will not share the information. So, the AI is dumb? They asked the AI a particular question "Why would a minor disagreement as children allow the same people as adults allow people to die?" The AI said, "Humans kill millions over minor differences like religion, race, and political views. It's just what humans do." Are we fit to judge the reasoning of AIs?

[@michaelmaultsby895]

What about Fred Saberhagen’s “Berserker Wars” series?

[@IRosamelia]

I've been totally meaning to read The Culture series, specially since Darrel from Sci-Fi Odyssey seems to be obsessed with them. I got Excession first but should probably go with Consider Phlebas instead. Decisions, decisions 😅

[@BookFurnace]

I've heard that Consider Phlebas is not the best starting point. I've read Player of Games and Use of Weapons so far. I liked the former a lot, but was not very impressed by Use of Weapons.

[@IRosamelia]

@@BookFurnace okay, I'll start with Player of Games then. Thanks! 😊

[@palehorseman8386]

A few things. Firstly, we're not that close to true AI. What we currently have is a predictive language processor. It just chooses the most likely word that follows the last word. Much like your phone is currently doing. Secondly, when we do create AI it will consume all of our media including How to Make Friends and Influence People. Lastly, each AI will be as unique as people are

[@BookFurnace]

Regarding the first point-I disagree strongly. Even though modren AIs are trained to do "just" next e word prediction, it's a very deep task, some call it "AI-complete" in the sense that perfecting it would imply having a general AI. It might not be intuitive, but consider: First, it does not predict just one "most likely word", it actually predicts for all words in the vocabulary how likely they are. Second, it is an extremely complex task. To be able to accurately predict the next word in ALL possible contexts, you need to be able to basically be a general AI alsready. If you are given a math problem, you need to accurately predict its solution (writing it one next word/symbol at a time), if you are given a court transcript, you'd need to complete the judge's decision, quoting appropriate statutes, etc. Current AI is not perfect at next word prediction, but saying "just next word prediction" undermines the depth behind the task. There are actually cognitive theories (predictive coding theory specifically) that say that maybe all that the brain is doing is similarly predicting the next experience (next word, next sound, etc). So in some ways, next word prediction is not a terrible model of how himans act. Regarding points 2 or 3 - maybe you are right and maybe not, I think those are strong claims, we'll see what happens.

[@douglasyoung2594]

Mathematics is govered by an underlying logical structure. It could be argued that mathematics IS the study of underlying logical structures. Human interactions are not logical. I say this as a Ph.D in Physics with dual degrees in math and physics. As a trained physical scientist (a chemist), Asimov understood this. Human interactions are inherantly complex (almost random in nature) and could only really be predicted by application of statistical principal (c.f. the "Foundation" series by, you guessed it, Isaac Asimov). AI could become socially smart by long periods of interacting with humans. It would then be very intersting to see if, given these long periods of interactions, AI could create a "laws of human interaction". This is negelecting cultural influnces that make human interaction different depending on which group of humans as interacting. AI will probably not wipe out mankind trying to solve the problem with Covid, but AI will have trouble interacting with humans at a cocktail party.

[@dw620]

Make sure the output is "not harmful" according to WHO? At present, that's human-imposed restrictions coming from a particular mindset - something as stupid as being able to ask for a random joke about men but not women - so pretty much meaningless. Comparing the current state of "AI" in 2023 vs. Asimov's *fiction* is meaningless since the tech we currently have is indeed both "dumb" and has no concept of context to map its regurgitations onto actual human experience (and we have little idea what's going on "under the hood"). How can we tell what is even "meaningful" to the AI when we can't even understand what goes on inside a dog's head in order to say that it *does* grasp "human nature". So, no, Asimov didn't "make a mistake" as you're talking about totally different things and in no way was he claiming to predict the state of technology in 2023. Clarke was (however) smarter by using a "black box" approach, less open to critique, in the likes of The City and The Stars where the AI is almost totally inscrutable... ; )

[@BookFurnace]

Regarding harmlessness - I agree that it's arbitrary, and that imposing too much "harmlessness" equates to squelching free speech in some cases. But I still think that it's awesome that we can apply Asimov's laws today, in a practical manner (although with some caveats). Second, I didn't want to nitpick on Asimovs technological predictions, I agree that it's silly. What I was pointing out is a more fundamental belief in Asimov's books that there is something so special in being human and thinking like a human that no technology could ever capture that. I still think that this is a strong throughline in his works, and I think this belief is not going to stand the test of time.

[@IRosamelia]

Yay!!! Fame & Fortune are mine!! Happy, happy, joy, joy! 😁

[@BookFurnace]

Well earned, haha :) Now you can revel and bask in glory till the end of times :)

[@IRosamelia]

@@BookFurnace and so I shall 😎 btw, this seems like the perfect time to mention you're a drop-dead gorgeous guy. Have a wonderful weekend! 🌹

[@BookFurnace]

Haha thank you :) Have a great weekend as well :)

[@yw1971]

4:03 - for who - Robots or humans?

[@BookFurnace]

I was speaking about humans. But AIs are also not great at higher mathematics, actually. Unmatched at applying standard techniques, but not great at developing new theory.

[@Yarblocosifilitico]

[BIG SPOILER] I'd point out, as a counter, that Daneel R. Olivaw does not only become very apt at reading and understanding humans, but he essentially is the puppet master of the entire Foundation saga. There's also the Bicentennial Man who if I recall correctly, becomes President (and a wise one). [/spoiler] But yeah, overall, Asimov's AI is pure logic vs Humanity's mix of rational and irrational.

[@BookFurnace]

Haha careful with saying too much for those who haven't read it:) But yes, Daneel kind of transcends his nature, to an extent. Although I'm not sure whether understanding global sociological patterns is the same as not being "dumb" and understanding humans.

[@Yarblocosifilitico]

@@BookFurnace Fair enough. I think, ultimately, he does, by giving Trevize the choice (even tho the alledged reason, ie, Trevize always being right just because, is quite forced). Kinda like with Leto in Dune, where he puppeteers everything but the ultimate goal is Humanity's freedom. Now, Asimov throws that away by having Trevize choose Gaia/Galaxia (cough cough hive mind...), but Daneel himself has placed the fate of the macro in the micro, which to me shows that he has, as you say, trascended his nature as a pure rational being.

[@omgwtfrofltomato]

Comment on the strike?

[@BookFurnace]

Sorry, not sure what you mean. Would you like me to make a video on the current writers' strike?

[@Tarquin2718]

I was a super Asimov fan when I was young, but yes his view on AI was nonsense.

[@PeterPan-cy7oy]

there is no A.I on this planet. 2np unequal np. pls do math basic pls

[@CPHSDC]

I read Asimov differently, whatever.

[@richelle-anncabatic5559]

🖤

[@grahamparr3933]

Why are talking to a plate😉

[@richelle-anncabatic5559]

Ur cute