AI and science and art

Discussions about serious topics, for serious people
Post Reply
User avatar
discovolante
Stummy Beige
Posts: 4084
Joined: Fri Oct 11, 2019 5:10 pm

AI and science and art

Post by discovolante » Wed Mar 01, 2023 10:11 am

Sorry in advance, I had a massive cup of coffee this morning and forgot that I'm pretty sensitive to it now, so I've got the jitters.

Anyway I'm sorry for this thread too because even though I started it, I'm getting slightly tired of the obsessing over Chat GPT and DALL-E 2 and all that. But I've listened to a few discussions about it anyway, from tech and arty and philosophical perspectives, with people who specialise in each of those areas. And I feel like each time, they find themselves bumping up against their ignorance of the other disciplines. Artists don't fully understand what the AI does or what it might or might not be capable of, scientists/tech people don't really understand what art is, philosophers are good at asking questions of both but can't really answer any of them. One thing all of them seem to grapple with unsuccessfully is the question of what consciousness/sentience actually is. I may be a bit behind here but I feel like until now there's been a kind of underlying assumption that it's ultimately an ability to be aware of your own existence, or think about yourself. But listening to these discussions it does seem to be that creativity is at the root of it, because we seem no more capable of being able to properly identify what creativity is than what consciousness is. I think the artists probably have a better grasp of it, as they understand the process more, but they also have more of a vested interest in identifying it as something uniquely erm, not human but biological I suppose.

Possibly I seem more stoned than high on caffeine but there you are.
To defy the laws of tradition is a crusade only of the brave.

User avatar
Brightonian
Dorkwood
Posts: 1429
Joined: Mon Nov 11, 2019 3:16 pm
Location: Usually UK, often France and Ireland

Re: AI and science and art

Post by Brightonian » Wed Mar 01, 2023 2:31 pm

Trying out ChatGPT etc. is on my to-do list, especially after this effusive video from Tom Scott.

Or maybe it's all a con, and these things are really a massive number of real people in an office block on the outskirts of Manila.

User avatar
nekomatic
Dorkwood
Posts: 1377
Joined: Mon Nov 11, 2019 3:04 pm

Re: AI and science and art

Post by nekomatic » Wed Mar 01, 2023 3:14 pm

discovolante wrote:
Wed Mar 01, 2023 10:11 am
Artists don't fully understand what the AI does or what it might or might not be capable of, scientists/tech people don't really understand what art is, philosophers are good at asking questions of both but can't really answer any of them.
A dilettante in all three fields writes…

The obvious comment on these AI models is that they aren’t intelligent, conscious or creative because all they do is produce mashups of pre-existing human-created content in response to their inputs, thus they are simply tools or techniques akin to sampling or collaging, just in a more superficially sophisticated way.

The obvious riposte to that is to ask whether human artists actually do anything fundamentally different, give or take the occasional step-change disruption of a Picasso or a Miles Davis.

I expect that philosophers can give us multiple possible definitions of consciousness, and for each one we could answer fairly well whether the AI meets it or not, but not a verdict on which definition is actually correct.

I guess my take on it - which I’d welcome philosophical comment on - is that if we accept that consciousness is ultimately a property of our brains which are physical systems, and not some spiritual or metaphysical woo thing, then there’s no fundamental reason that a sufficiently complex and powerful computer model that’s capable of modifying itself over time can’t achieve what we would define as consciousness. That doesn’t mean ChatGPT is anything near there though.
Move-a… side, and let the mango through… let the mango through

User avatar
dyqik
Princess POW
Posts: 7527
Joined: Wed Sep 25, 2019 4:19 pm
Location: Masshole
Contact:

Re: AI and science and art

Post by dyqik » Wed Mar 01, 2023 4:25 pm

Brightonian wrote:
Wed Mar 01, 2023 2:31 pm
Trying out ChatGPT etc. is on my to-do list, especially after this effusive video from Tom Scott.

Or maybe it's all a con, and these things are really a massive number of real people in an office block on the outskirts of Manila.
If those people have limited English, then you're in the Searle's SE Asia Chinese room scenario.

User avatar
discovolante
Stummy Beige
Posts: 4084
Joined: Fri Oct 11, 2019 5:10 pm

Re: AI and science and art

Post by discovolante » Wed Mar 01, 2023 7:19 pm

Brightonian wrote:
Wed Mar 01, 2023 2:31 pm
Trying out ChatGPT etc. is on my to-do list, especially after this effusive video from Tom Scott.

Or maybe it's all a con, and these things are really a massive number of real people in an office block on the outskirts of Manila.
I've had a go at it and asked its views on stuff and its responses could probably be best described as 'undergraduate level' - reasonably considered and seeming to 'understand' the subject matter but no original thought or ability to form a conclusion of its own.
To defy the laws of tradition is a crusade only of the brave.

User avatar
discovolante
Stummy Beige
Posts: 4084
Joined: Fri Oct 11, 2019 5:10 pm

Re: AI and science and art

Post by discovolante » Wed Mar 01, 2023 7:36 pm

nekomatic wrote:
Wed Mar 01, 2023 3:14 pm
discovolante wrote:
Wed Mar 01, 2023 10:11 am
Artists don't fully understand what the AI does or what it might or might not be capable of, scientists/tech people don't really understand what art is, philosophers are good at asking questions of both but can't really answer any of them.
A dilettante in all three fields writes…

The obvious comment on these AI models is that they aren’t intelligent, conscious or creative because all they do is produce mashups of pre-existing human-created content in response to their inputs, thus they are simply tools or techniques akin to sampling or collaging, just in a more superficially sophisticated way.

The obvious riposte to that is to ask whether human artists actually do anything fundamentally different, give or take the occasional step-change disruption of a Picasso or a Miles Davis.
I think they/we do. In fact this came up in the interview I was listening to that prompted this thread (an interview with Adrian Tchaikovsky). They discussed it generally but he also gave an example of the Transformers TV series and how repetitive it is, with basically the same plot each time but with a different villain. I thought that was a pretty poor example because 1) it's for children who tend to like repetition and 2) it's an almost entirely commercial venture with no real interest in actual creativity.

I think one person (on the arty side of things, very much not an AI expert) described the difference quite well. They said that you could ask a person to draw a picture of something but what they end up drawing could be something only loosely connected with what you've asked them to do because they have the ability to let their imagination take them off in different directions.

Of course lots of stuff can be derivative of or influenced by stuff that came before, but that doesn't mean it's nothing more than a kind of collage.The same facial features, or even groups of facial features, can appear on different people, but that doesn't mean that any two faces will be the same.

This sort of reminds me of the Borges short story where someone tries to write Don Quixote word for word, but not actually copy it, heh.
I expect that philosophers can give us multiple possible definitions of consciousness, and for each one we could answer fairly well whether the AI meets it or not, but not a verdict on which definition is actually correct.

I guess my take on it - which I’d welcome philosophical comment on - is that if we accept that consciousness is ultimately a property of our brains which are physical systems, and not some spiritual or metaphysical woo thing, then there’s no fundamental reason that a sufficiently complex and powerful computer model that’s capable of modifying itself over time can’t achieve what we would define as consciousness. That doesn’t mean ChatGPT is anything near there though.
Yeah it's a difficult one to escape isn't it, although I suppose one fairly happy conclusion from it is that we don't necessarily need to see other animals as fundamentally different from us - obviously not something to get carried away by but I think it makes life a bit more interesting.
To defy the laws of tradition is a crusade only of the brave.

User avatar
nekomatic
Dorkwood
Posts: 1377
Joined: Mon Nov 11, 2019 3:04 pm

Re: AI and science and art

Post by nekomatic » Wed Mar 01, 2023 8:55 pm

discovolante wrote:
Wed Mar 01, 2023 7:36 pm
you could ask a person to draw a picture of something but what they end up drawing could be something only loosely connected with what you've asked them to do because they have the ability to let their imagination take them off in different directions.
Well, I don’t think we can exclude that a big computer model could do that, if its parameters are adjusted to let that happen. Is imagination anything more than the ability to draw conclusions that are further than usual away from our inputs?
Move-a… side, and let the mango through… let the mango through

User avatar
Woodchopper
Princess POW
Posts: 7057
Joined: Sat Oct 12, 2019 9:05 am

Re: AI and science and art

Post by Woodchopper » Wed Mar 01, 2023 10:10 pm

nekomatic wrote:
Wed Mar 01, 2023 8:55 pm
discovolante wrote:
Wed Mar 01, 2023 7:36 pm
you could ask a person to draw a picture of something but what they end up drawing could be something only loosely connected with what you've asked them to do because they have the ability to let their imagination take them off in different directions.
Well, I don’t think we can exclude that a big computer model could do that, if its parameters are adjusted to let that happen. Is imagination anything more than the ability to draw conclusions that are further than usual away from our inputs?
I spent a bit of time in December playing around with one of the AI image generators. Just enter some keywords and it produces an image. It was doing that, coming up with takes on the keywords that I hadn't thought of.

I'm not suggesting that this is the same as an artist being creative. I think that Nick Cave has the last word there.

But what is happening is that the realm of things that humans can do that computers can't keeps getting smaller. Its incremental steps and with each the humans have to redefine ways that they are unique. I have no idea where the process will end. Despite its effective mimicry Chat GPT is a very long way from replicating human intelligence.

User avatar
Woodchopper
Princess POW
Posts: 7057
Joined: Sat Oct 12, 2019 9:05 am

Re: AI and science and art

Post by Woodchopper » Wed Mar 01, 2023 10:13 pm

nekomatic wrote:
Wed Mar 01, 2023 3:14 pm
I guess my take on it - which I’d welcome philosophical comment on - is that if we accept that consciousness is ultimately a property of our brains which are physical systems, and not some spiritual or metaphysical woo thing, then there’s no fundamental reason that a sufficiently complex and powerful computer model that’s capable of modifying itself over time can’t achieve what we would define as consciousness. That doesn’t mean ChatGPT is anything near there though.
Maybe, or at least it'll become harder and harder to tell the difference.

But our brains are built very differently from any machine yet devised. So what we think of as consciousness could be a property of the specific way that our meat computers are made.

User avatar
Brightonian
Dorkwood
Posts: 1429
Joined: Mon Nov 11, 2019 3:16 pm
Location: Usually UK, often France and Ireland

Re: AI and science and art

Post by Brightonian » Fri Sep 08, 2023 5:34 pm

Brightonian wrote:
Wed Mar 01, 2023 2:31 pm
Trying out ChatGPT etc. is on my to-do list, especially after this effusive video from Tom Scott.

Or maybe it's all a con, and these things are really a massive number of real people in an office block on the outskirts of Manila.
This AI startup that "generated" 3D models from 2D art was actually using human workers

User avatar
Brightonian
Dorkwood
Posts: 1429
Joined: Mon Nov 11, 2019 3:16 pm
Location: Usually UK, often France and Ireland

Re: AI and science and art

Post by Brightonian » Sat Sep 09, 2023 6:26 pm

Has anyone used any AI tools to produce programming code? I'm wondering whether I should even bother. GPT-4 is starting to be downright annoying.

I've just spent about three hours trying to get it to generate code that won't fall over. I wrote the requirement in just 15-20 minutes. Without even having to clarify my requirement, I went through 12 rounds of:
GPT-4 wrote:Here's your code <etc.>
Brightonian wrote:No, got error message <etc.>
On the 13th round it finally worked.

Sometimes it's just me, but usually I can express my requirements clearly, and it's GPT-4 that's away with the fairies. This time it really took the biscuit. But that's fine - I really didn't want to go sitting outside the pub with a beer. :x

User avatar
bjn
Stummy Beige
Posts: 2916
Joined: Wed Sep 25, 2019 4:58 pm
Location: London

Re: AI and science and art

Post by bjn » Sat Sep 09, 2023 7:13 pm

Not going there. Partly on principle, because if they want to use my data to train their LLM, whether my requirements or my code, they can pay me. Partly because what I tend to code isn’t sufficiently with in their training set to generate anything useful (I am special me). Partly because the code generated isn’t particularly good anyway.

User avatar
dyqik
Princess POW
Posts: 7527
Joined: Wed Sep 25, 2019 4:19 pm
Location: Masshole
Contact:

Re: AI and science and art

Post by dyqik » Sat Sep 09, 2023 7:14 pm

Brightonian wrote:
Sat Sep 09, 2023 6:26 pm
Has anyone used any AI tools to produce programming code? I'm wondering whether I should even bother. GPT-4 is starting to be downright annoying.
It's a known route for introducing security exploits into code, particularly via libraries that GPT makes up, and which hackers then create to bundle exploits onto the host machine.

Imrael
Snowbonk
Posts: 505
Joined: Tue Nov 12, 2019 5:59 am

Re: AI and science and art

Post by Imrael » Sun Sep 10, 2023 7:56 am

Brightonian wrote:
Sat Sep 09, 2023 6:26 pm
Has anyone used any AI tools to produce programming code? I'm wondering whether I should even bother. GPT-4 is starting to be downright annoying.

I messed around with getting solutions to a database admin technical issue. The result was a process of legal correctly described steps that might have solved the problem. But failed to take account of some normal factors in a production environment and could have been actively destructive.

The issue here I think is that an inexperienced person could try to follow advice like this and cause havoc.

User avatar
snoozeofreason
Snowbonk
Posts: 485
Joined: Fri Nov 15, 2019 1:22 pm

Re: AI and science and art

Post by snoozeofreason » Sun Sep 10, 2023 5:31 pm

I've tried getting ChatGPT to solve cryptic crossword clues, and found the results quite revealing. For example given the clue

Police document conflict in NHS, rat race ghastly (6,7)
ChatGPT wrote: The answer to this cryptic crossword clue is "MINUTE BOOK"

Explanation:

"Police document" refers to a record kept by the police,
"Conflict" could be a homophone for "clash"
"NHS" is National Health Service, a government-funded healthcare system in the UK
"Rat race" could be a metaphor for a busy and hectic lifestyle
"Ghastly" could mean something is unpleasant, or could also be a homophone for "lastly"

So "Record kept by the police of a clash in the National Health Service, Busy and hectic lifestyle lastly" is "MINUTE BOOK" (6,7) means the answer is two words and the length of the first word is 6 letters and the second word is 7 letters.
The interesting thing about it is that the explanation, on a superficial level, seems to display an understanding of how cryptic crosswords work. It makes it look as if ChatGPT knows that homophones and metaphors are relevant and that the numbers in brackets refer to the lengths of the words in the clues. But then it provides an answer that has no relation to the explanation it has just given (not even getting the lengths of words correct).

(For non crosswordy types, the real answer is "Search Warrant" - anagram of "NHS, rat race" placed around the word "war").
In six days the Lord made the heavens and the earth, the sea, and all that is in them. The human body was knocked up pretty late on the Friday afternoon, with a deadline looming. How well do you expect it to work?

User avatar
Rich Scopie
Snowbonk
Posts: 536
Joined: Mon Nov 11, 2019 1:21 pm

Re: AI and science and art

Post by Rich Scopie » Mon Sep 11, 2023 10:31 am

discovolante wrote:
Wed Mar 01, 2023 7:36 pm
I think one person (on the arty side of things, very much not an AI expert) described the difference quite well. They said that you could ask a person to draw a picture of something but what they end up drawing could be something only loosely connected with what you've asked them to do because they have the ability to let their imagination take them off in different directions.
We did this in one if those pointless meetings that HR likes to schedule you in for. We were all asked to draw a flying horse.

Lots of crap depictions of Pegasus, a few horses held up by balloons etc, but I was the only one who drew a pub.

I suspect that says more about me than HR really needed to know.
It first was a rumour dismissed as a lie, but then came the evidence none could deny:
a double page spread in the Sunday Express — the Russians are running the DHSS!

User avatar
shpalman
Princess POW
Posts: 8244
Joined: Mon Nov 11, 2019 12:53 pm
Location: One step beyond
Contact:

Re: AI and science and art

Post by shpalman » Mon Sep 11, 2023 11:10 am

Rich Scopie wrote:
Mon Sep 11, 2023 10:31 am
discovolante wrote:
Wed Mar 01, 2023 7:36 pm
I think one person (on the arty side of things, very much not an AI expert) described the difference quite well. They said that you could ask a person to draw a picture of something but what they end up drawing could be something only loosely connected with what you've asked them to do because they have the ability to let their imagination take them off in different directions.
We did this in one if those pointless meetings that HR likes to schedule you in for. We were all asked to draw a flying horse.

Lots of crap depictions of Pegasus, a few horses held up by balloons etc, but I was the only one who drew a pub.

I suspect that says more about me than HR really needed to know.
flying horses
having that swing is a necessary but not sufficient condition for it meaning a thing
@shpalman@mastodon.me.uk

Post Reply