Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
AI and the Flimflam Man
Courtesy of NightCafe. An AI-drawn mentalist, though more fantastical than I had in mind.
To the amazement of the crowd nearly one third of them stand up. “I knew it!” shouts the mentalist, “but I almost didn’t pick 37; I thought of 39 at first. How many of you picked 39?” Over half the audience is now standing, to wild applause.
I think I pulled that example from Michael Crichton’s State of Fear, but it’s been so long since I’ve read the book that I’m no longer sure. Let’s figure out the trick. If I asked you to guess a number between one and 50, fair odds say you’d get it right 2% of the time. The mentalist does a great deal better than that. How? He isn’t asking for a number between one and fifty.
Since the number has to be two digits we lose [1,2,3,4,5,6,7,8,9] right away. Since both of the digits must be odd we lose everything in the 20s and 40s too. Not quite; there’s going to be some guy in the audience fervently thinking “29” who’s forgotten that requirement, but leave him aside. Since both digits are odd you only get half of the decades that are left. Here’s all remaining possibilities:
[11, 13, 15, 17, 19, 31, 33, 35, 37, 39]
That’s only ten numbers; the odds are way better already. From there you have to go into psychology. People are less likely to choose a number with a repeated digit [11, 33], and perhaps the kind of folks who believe already will shy away from an unlucky number [13]. After that (and I’m afraid I’ve forgotten the specifics) people pick some numbers preferentially. Survey enough people and you get the idea that 37 stands out. Between that and how the mentalist took credit for two numbers you can see how he’d get half the room.
The same picture as the first, only a craggy mountain valley.
What does that all have to do with AI? Whether it’s a stage mentalist or a phishing email or a pretty lady in a bar who just happens to come from your home town, the way to suss out the scheme is to watch inputs and outputs. You’ll hear a lot of talk about the middle — how the machine learning algorithms rely on matrices full of numbers that not even their programmers can decode, or the mystical wonder of telepathy, or how the quants in the basement have the stock market beat — but keep a very sharp look at those inputs. A flimflam man will pick up on a signal you didn’t realize you sent and amplify it right back to you as a successful prediction.
That’s why I’m not worried about AI. It isn’t a scam, but think about it in that model for a minute. When I went to make that picture up top, I typed in a description of what I wanted to see; an excited mentalist on stage. That’s the input. The AI didn’t have the genius to come up with the idea itself; it didn’t pluck a particle of inspiration out of the ether. Want proof? Take a look at the second drawing. It’s the same one as the first, only I changed the prompt to “Show me something you want to draw.” That’s nonsense to an AI. It can’t understand what I asked, so it gives back essentially a nonsense answer. It drew the same scene, only as a mountain valley.
Now outputs. The input is you tell it what picture to draw, the output is you tell it what picture you’ll accept. The AI took my first prompt and decided that I wanted a fantastical scene with actual magic going on. The first picture was so interested in the crowd that it crowded out the mentalist. I amended the prompt to “Excited mentalist on stage, with focus on the mentalist.” I was hoping for more of a stage magician approach, but the picture is cool enough to share.
AI isn’t a scam. There’s magic going on inside that black box even if you have to provide some yourself. Whether or not those pictures are what exactly I had in mind, they’re better than I could create myself, at least without intensive training. To the extent that this post could be written by ChatGPT it’s because the machine has successfully duplicated all the rules of composition and grammar that I only imperfectly understand. Machine Learning is a powerful tool, and I look forward with hope for it to automate away the drudgery in a great many jobs.
But remember those inputs and outputs. By prompting the AI with your idea, and by selecting an acceptable execution of that idea, you’re providing the creativity that the machine can’t automate. The machines will make it easier to replace art. They will not replace art as a profession. The robots can now churn out pulp novels. They won’t make the next great work of fiction. It might be that your secure job is now threatened. Your job was never secure from the future. It’s only con men who told you otherwise.
Published in Science & Technology
His point was more rigorous. It’s not about emotional responses, it is about flashes of insight, breakthroughs, etc.
Certain things are not computable like what happens when the quantum wave form collapses. He suggested that there might be a process in the neurons that would cause a quantum wave form collapse, but I think he had to back off of that one.
But we did find an entirely different way to achieve Psycho History.
I have to think about this. :) But I thought creativity and insight, and flashes of inspiration were what he was talking about, this is as emotional as logical, for example, dreaming an answer.
Yes, I agree and perhaps I misunderstood your comment about emotional responses, and I think we actually agree on this. These are the things that happen to a wet consciousness that don’t happen to a Turing machine. For whatever reason.
It’s like the brain’s background processes. So why, when you are facing a difficult problem and are thinking about it a lot, does the solution pop into consciousness when one is in the shower, asleep, or driving, and most importantly, not thinking about the problem. Your consciousness keeps working on things even when they are not at the conscious level. Which seems like a contradiction but it nonetheless my experience.
Oh my, I had to look that up. Asimov is such a bore…
He wasn’t even all that great back then, unless you were like 12 years old or something. Although to be fair a lot of the competition wasn’t much better. Basically, Asimov was something to read if nobody better had written anything new.
He had some good stuff, and he had some mediocre stuff.
He had none of the insight that his contemporaries Arthur C. Clarke or Robert Heinlein had. Some interesting ideas, but his prose is tedious and the stories are ultimately unsatisfying.
I’m not disagreeing here but redirecting. Producing an answer from unconscious thought running in the background sounds to me very much like what a computer does. But creative representations of beauty and elegance that is truly original and whole in its composition that strikes the viewer or reader or listener as profound, seems harder to create, perhaps impossible, for a machine.
And even if and when it can, I don’t think any machine can appreciate it. It can critique it, but I doubt it will ever be moved by it or perceive it as beautiful.
I dunno, my recollection is that his “good” stuff wasn’t all that good.
I agree.
But I’ll be a little pedantic and say that my underlying point is that in general, not just with creative representations, what consciousness does cannot be fully replicated on a Turing machine.
I think he was a better short story writer than a novelist. I remember really liking Waterclap and several of the robot short stories.
I suppose. But I’m not all that sure what a Truing machine is? I thought it was, in relation to what we’re talking about now, a relatively primitive machine that can imitate human conversation so well that it can fool people into thinking it was a live person.
Even though I’ve heard that ChatGPT writes pointless odd stories with poor plots, I think we’re on the cusp it creating acceptable pulp fiction — all derivative, I’d think. So basically, I’m agreeing with you.
But to follow my point about a synthetic consciousness like human consciousness, I wonder if any machine will ever be able to be made speechless in appreciation at the wordless wonder of a trumpet solo or a mostly white-on-white Monet of cherry blossoms. That I’m pretty sure is impossible.
That’s a pretty good summary.
You’re conflating the Turing machine with the Turing test. The machine is an abstract machine equipped with a tape of infinite length. The tape is divided up into cells. The machine can read and write a finite set of symbols, advance the tape one cell in either direction, and keep track of which one of a finite number of states it is currently in. Turing asked questions about the programs that could be run on such a machine, and whether or not the machine would be able to determine if it was in an infinite loop or not.
The Turing test is meant as a measure of artificial intelligence. If you are watching a “conversation” between a human and a machine, would you be able to tell which was which?
Thanks.
My “advanced project” for the head of the Math and Computer Departments at Oregon State University in 1978 was a “Normal Algorithm” processor which is somewhat similar to the Turing machine. It was written in FORTRAN, over a weekend, first in longhand on yellow paper pads, and the following week I keypunched it myself and it worked the first time.
The coding was tedious but not difficult, at least not for me. The only “problem” I had was that it was going to run on a CDC Cyber timesharing mainframe system, which had a 60-bit word length and 192k words of system memory. Most text processing on those systems packed 10 6-bit characters (upper-case only, etc) into each 60-bit word. But since I was dealing with string manipulation on a character-by-character basis, if I had done that the system would have spent 99.9+% of its time just unpacking and repacking those words. So I put ONE character in each 60-bit word, which meant much of the system’s timesharing capability was unavailable when even ONE person was running that program. Its use had to be rather strictly limited in times of day.
I agree with your last paragraph wholeheartedly.
A Turing machine is a mathematical model of a general purpose computer. So in today’s context, any computer. https://en.wikipedia.org/wiki/Turing_machine
I think that Google, Facebook, et al are getting there.
To be clear, I was reading Beyond All Weapons, in Astounding’s November 1941 issue.
It’s not the greatest story; this revelation is treated as if it’s a surprise. I’m not sure whether or not that’s because it’s been done so many times in the years since. Maybe Nat Schachner is the innovator who came up with the idea, in which case he deserves more credit than I’m giving him. Asimov’s psychohistory viewed individuals as fundamentally unpredictable, but groups as statistically reliable. This guy’s psycho-history goes further,treating psychology as more of an exact science.
While I’m at it, every month readers would submit their top five ranked stories from that issue, and two months on they’d publish the results. Here’s the listing from September 1941:
Caleb Saunders is Heinlein writing under a pen name. I think Methuselah’s Children got ranked higher than it would because it was the conclusion of a serial. Heinlein is a better writer than Asimov, but Nightfall was better than Methuselah’s Children. And Elsewhere, but that was always one of Heinlein’s weaker stories. Asimov is a better writer than y’all are giving him credit for, for all that his stories never have villains. Clarke I tend to rate lower than most people do.
For the record, I read The Emperor’s New Mind two years back. You can find my Ricochet post on it here. How do I know it was exactly two years ago? I included the Wellerman sea shanty in the post.
To summarize the post, Penrose was of the opinion (which I share) that
Having spent some time thinking about AI, I’d elaborate that argument and this post as follows:
It can and probably will replace the pretenders and hacks. A stroll through an art gallery or two will provide clarity.
Makes me think of the Auditors of Reality from Discworld, dissecting a painting into its component atoms to try and figure out what made it aesthetically pleasing.