6 ms·
Whenever I come across some new AI hype thing in the wild with the usual buzzwords, a sentence Dijkstra wrote way back sort of flashes through my head. It alway
by ryloric 4y ago
Whenever I come across some new AI hype thing in the wild with the usual buzzwords, a sentence Dijkstra wrote way back sort of flashes through my head. It always reminds me to be careful and precise with words.
"...the question of whether Machines Can Think, a question of which we now know that it is about as relevant as the question of whether Submarines Can Swim."
In the same "post" he also talks a lot about hype and how it hurts the field. The last few years of computing have been exceptionally hype driven, even more than usual. We could've been better with a few more critics like him in that period.
[0] https://www.cs.utexas.edu/~EWD/ewd08xx/EWD898.PDF https://www.cs.utexas.edu/~EWD/ewd08xx/EWD898.PDF
- idealmedtech 4y agoThere are certainly critics, but their opinions, if consistently opposed to the hype cycle, get either sidelined or lambasted. Just sort any post on some new ML application by controversial comments and you'll find the critics aplenty.
- tomxor 4y ago> get either sidelined or lambasted "No people"'s comments only seem to be worth enough in attention currency once things explode and it turns into an "I told you so". At least within the world of facehype.com and hypper.com et al. Behind closed doors people are probably more critical.
- qsort 4y agoIt's also a sentence constantly on my mind. It's interesting to note it's a very different type of criticism than the ones that usually balloon when something new or supposedly new becomes public. It's the kind of criticism that leaves room for genuine acceptance of the technical excellence and usefulness of those tools, while not buying the hype in the slightest. EWDs are a gold mine. He's an exceptional thinker and his prose is beautiful.
- baxtr 4y agoCan you recommend any book by him? Or essays?
- 082349872349872 4y agoessays collected at: https://www.cs.utexas.edu/users/EWD/ https://www.cs.utexas.edu/users/EWD/ books listed at: https://en.wikipedia.org/wiki/Edsger_W._Dijkstra#Selected_publications https://en.wikipedia.org/wiki/Edsger_W._Dijkstra#Selected_pu...
- baxtr 4y agoThanks for sharing! I realize I should have been clearer in my request. I was looking for 1-2 specific recommendations. I will look at the lists nonetheless.
- rramadass 4y agoTwo books that i can recommend; 1) A Method of Programming. 2) A Discipline of Programming. Browse the EWDs and read those whose titles interest you. My favourites are those dealing with Program Correctness and Languages. Note that understanding Dijkstra can be hard; so don't be discouraged if you don't "get" something at the first read.
- ryloric 4y agoEWD-898 I liked above is pretty great imo. I'v re-read it over half a dozen times over the years and still find it valuable.
- qsort 4y agoMy personal favorites: On the cruelty of really teaching computer science: https://www.cs.utexas.edu/users/EWD/ewd10xx/EWD1036.PDF https://www.cs.utexas.edu/users/EWD/ewd10xx/EWD1036.PDF How do we tell truths that might hurt?: https://www.cs.virginia.edu/~evans/cs655/readings/ewd498.html https://www.cs.virginia.edu/~evans/cs655/readings/ewd498.htm... The humble programmer: https://www.cs.utexas.edu/~EWD/transcriptions/EWD03xx/EWD340.html https://www.cs.utexas.edu/~EWD/transcriptions/EWD03xx/EWD340... The notational conventions I adopted and why: https://www.cs.utexas.edu/users/EWD/ewd13xx/EWD1300.PDF https://www.cs.utexas.edu/users/EWD/ewd13xx/EWD1300.PDF The Mathematical Divide: https://www.cs.utexas.edu/users/EWD/ewd12xx/EWD1268.PDF https://www.cs.utexas.edu/users/EWD/ewd12xx/EWD1268.PDF Why American Computing Science seems incurable: https://www.cs.utexas.edu/users/EWD/ewd12xx/EWD1209.PDF https://www.cs.utexas.edu/users/EWD/ewd12xx/EWD1209.PDF
- DarmokJalad1701 4y agoWhen I first saw the title of this post, my first thought was that that this was a letter written by ChatGPT pretending to be Djikstra. So I guess there is something to be said about hype ... not sure exactly what though.
- rramadass 4y ago>Whenever I come across some new AI hype thing in the wild with the usual buzzwords Next time use this: "Is it your belief that we are all idiots?" From: https://www.cs.utexas.edu/users/misra/Speeches.dir/Banquet-WG2.3-March2010 https://www.cs.utexas.edu/users/misra/Speeches.dir/Banquet-W...
- quietbritishjim 4y agoUntil I read your link (which was very interesting BTW) I interpreted that quote the other way: Is it your [the parent commenter] belief that we [AI researchers] are all idiots?" I mean, sure there's a lot of marketing junk, but it's almost as naive to assume that there's no substance as it is to believe everything the marketers say. Having said that, I don't think any serious AI researcher would claim that their creations are close to actually thinking.
- qsort 4y ago> I don't think any serious AI researcher would claim that their creations are close to actually thinking. I don't think anybody seriously believes that -- but setting reasonable public expectations from emerging technologies is at least in part the responsibility of the "technically literate class" broadly defined, research scientists, practitioners, even developers in ML/AI-adjacent fields such as myself. Stuff like "GPT3 is coming for your job" is already out to the press. Nobody thinks that, nobody has even said that, but nobody has done anything to set expectations straight either. We ought to be more responsible in this respect.
- CamperBob2 4y agoEh, they're not wrong. Given another 20-30 years of progress at this rate, it's coming for almost every knowledge worker's job. Which is pretty freaking cool.
- qsort 4y agoIt's been 20-30 years in the future for several decades now.
- Kiro 4y agoI think AI lives up to the hype. The stuff we take for granted now was pure science-fiction just a couple of years ago. If this pace of development holds up it deserves to be hyped.
- deelly 4y agoSorry, english is not my native language, could you explain "...whether Submarines Can Swim." meaning? Swim as fish (answer - no, because submarines swim using propellers and analogs), swim as underwater ship (answer - yes, because submarines build to swim)?
- Cornelius267 4y agoI think that you actually understood the point perfectly! I'm a native english speaker, and to me, it looks like he is trying to raise the exact ambiguities that you are asking about. I don't think that his statement is meant to be an answer, rather, I think it's meant to be a statement to jump off of. It's bringing up how definitions matter, and that in different contexts the answer to the same questions could be yes or no.
- herodotus 4y agoLike many words in natural language, "Swim" does not have a precise meaning. A dictionary will probably say something like "Propel through water using limbs", although the word "swimming" can sometimes be used metaphorically, for example "I have bits of bacon swimming in my potato soup". Cleary, bacon does not have limbs! Now even though Submarines (usually) have propellors, one can easily imagine a design that used limb-like oars to propel it. Does this mean that such a strange submarine is swimming? The point Dijkstra was making is that the comparison is pointless: there is a fundamental difference between an organism swimming and a machine moving through water. Trying to equate these two actions is pointless, since it adds no new depth of understanding. He therefore dismisses the question of "can machines think" as having no usefulness.
- ryloric 4y agoYou're correct. My intent in quoting was to point out how careless usage of words in AI can fool people, especially the layperson and create wrong expectations. Most can see the difference in the usage of word 'swim' w.r.t fish and submarines, but public knowledge about AI is not at the point where they can make a similar distinction when using 'think' or 'chat' between human and machine. He uses it differently though, he's using that sentence to criticize Alan Turing's early hypothesizing about AI. If you're interested, that document linked has more context.