Bing Chat began to hallucinate. It assured us that we could safely eat ground glass, and that four US Presidents had been women. It would invent citations. It would then deny an answer it had just given you was true. All this was performed with the smoothly reassuring bedside manner of an experienced NHS consultant.
Things got even worse as Bing Chat began to throw tantrums, and even threaten its users.
A whole series of stories had AIs – rather more advanced – going mad mere tens of minutes after being activated.Such powerful intelligences working at such speed just chewed through all of the interesting bits of possible thought so quickly they ended up composing symphonies in C through the colour blue sort of madness. A fugue state where they considered the mysteries of the cosmos sorta stuff.
Still used them, of course, but they did only last as useful entities for those tens of minutes.
And to this end they built themselves a stupendous super-computer which was so amazingly intelligent that even before its data banks had been connected up it had started from “I think therefore I am” and got as far as deducing the existence of rice pudding and income tax before anyone managed to turn it off.
I do wish people would stop hyping this as Artificial Intelligence. Large language models have no intelligence, just very good statistics about which words are likely to follow each other in the training corpus. What they actually provide is AP – Artificial Parrots. “Who’s a clever chatbot?”
Can anyone help?
In my youth I read a sci-fi/dystopic novel with a sub-plot involving 3 AI personalities named, I think, George, Saint George and a female one. They interacted and, when upset, showed their displeasure by offering each other cats.
Eventually the female kept tormenting Saint George (have a cat, have another cat etc) so much that he cracked and threatened to kill her. At that point the developers went back to the drawing board.
Can anyone tell me what the book was or have I completely flipped following a recent close encounter with post-op Codeine?
If Bing Chat likes model railways then we have a wrap.
I am also reminded of the great Bob Newhart and his sketch about infinite monkeys and typewriters
“Hey I think 3375 is onto something :
To be or not to be
That is zfripofsednum “
TBH, it sounds like a typical lefty…
Arthur – yarp. We are not going to build a thinking machine by building better Chinese Rooms.
Otoh, I don’t believe we’re ever going to build a thinking machine, so algorithmic remixes of data we already have is probably about as good as it practically gets in the field of “AI”.
It’s a toy, ultimately, but a bloody useful one. Focusing on the funny or mental machine generated responses misses the point: it’s not meant to be HAL 9000, it’s more of a digital lucky bag with infinite free refills.
In this exchange, ChatGPT describes itself as “an AI language model” and demonstrates it hasn’t quite got to the AI level of Holly: https://wattsupwiththat.com/2023/02/27/chaptgpt-admits-there-has-been-no-warming-since-2015/
Found it.
The book was Michael Crichton’s The Terminal Man. The AI personalities were George, Saint George and Martha.
Says things that are clearly not true. Doubles down and gets angry with people who point that out. Later denies that it said it in the first place.
Good grief, it’s read Murphy’s blog and thinks that’s normal human behaviour to copy!
More broadly, if they learn from what’s online, it’s no wonder they’re going mad.
I can remember when statistics was no more than the poor relation to mechanics in the field of Applied Maths – a useful tool for alerting you that something might be worth investigating, but never an end in itself.
But much of what passes for science now seems to be no more than statistics given a fancy name, whether it’s artificial intelligence, epidemiology, climate science, particle physics … even economics.
Arthur, Steve: I’ve been saying elsewhere that these are only “artifical intelligences” in the sense that they’re very good at appearing intelligent.* The human mind has a very strong tendency to correlate intelligence and erudition (hence the ancient study of rhetoric, and why we keep electing smooth-talking idiots to office), but it’s not necessarily the case.
“Good grief, it’s read Murphy’s blog and thinks that’s normal human behaviour to copy!”
Closer to the truth than most people realise, actually.
*Any suggestions that I recognised this because I’m good at it myself will be met with a Hard Stare. And a slighly guilty conscience.
I don’t believe we’re ever going to build a thinking machine,
Depends what you mean by thinking. If you mean the sort of rules decision making currently performed by office desk jockies, we’re probably there. But if you’re expecting it to endlessly discuss football over the phone with its mate. probably not
I do wish people would stop hyping this as Artificial Intelligence. Large language models have no intelligence, just very good statistics about which words are likely to follow each other in the training corpus. What they actually provide is AP – Artificial Parrots. “Who’s a clever chatbot?”
Yes, Andrew O (who wrote the no-longer-very-Torygraph article) has described ChatGPT as “the world’s worst autocorrect system”.
Nice comment on the article from the Speccie:
There’s always the OMC route.
Not really surprising. At macro level you can work in generalisations and say “things work this way. period.”
At the underlying micro level, you run into the snag that things don’t always work that way, and find that you’ll have to add the word “generally”. And basic statistics get introduced to give you the odds on things working as intended/buggering up.
When figuring out what causes the “generally”, you hit the quantum level where all bets are off, and you need really fancy math and statistics to even describe what’s going on there.
Most actual science nowadays is done on the micro/quantum level, so yeah… Statistics is a given. Turns out the bastard child of Applied Math is crucial in understanding how the world around us actually works.
And statistics, like logic, is a lovely trap that can lure you into being wrong with utter conviction.
Something that gets hammered into students in a proper course in it. And the smart ones actually remember it.
Problem is, of course, that proper statistics is hard, like any math. And, increasingly, that the people having an Opinion about Science can’t even do normal divisions, may have heard of a bell curve, and are completely helpless when it comes to compound probabilities. And they never, ever, have even heard about the Pitfalls of Probability.
With obvious results, as we get a sample of on this blog every day..
Science ( actual science..) has moved into the realm of statistics. Because Reality™.
The problem is that statistics is still treated as that Bastard Child, and not properly taught at the levels of education where it actually matters.
And, shall we say, less scrupulous people, bank on this fact and abuse statistics to build their narrative, like snake-oil salesmen.
Because most people capable of understanding the stuff never had the grounding to spot the manipulation.
Sometimes I wonder if that, too, is deliberate.
Statistics doesn’t tell you *why* it only tells you how likely something is, or what the probability function describing a set out outcomes is if we want to get more fancy. At least in physics, chemistry, biology etc they have a LOT of observations from which to draw. Finance, economics, and climate have, in order, lots*, some and hardly any.
* lots of caveats here. Lots for days but extend the time and you have so few the model is clearly wrong but people still use them as they have nothing better. Tells you a lot really.