AI: What's it good for?

[QDRPHNC]

I'm a designer and am of course worried about the impact of AI on my job. I'm not as worried as a lot of people because frankly I'm a good designer, and while AI will eventually come for my job, it has to work its way through the "I'll get my nephew to design the logo, he has Photoshop" level folks first.

And I've resisted demonizing it, because I think it's genuinely pretty amazing, and it can be used by artists in really interesting ways that I don't think have been fully explored yet. I also think that, in much the same way the internet itself removed the barriers to entry for getting your creative work in front of an audience, AI is taking this a step further and allowing people like myself, who want to create things on zero budget, opportunities to make the sort of work that would previously have been out of reach.

And of course, it's not going anywhere. I understand the copyright issues and the fears of writers, illustrators, designers, but it all just seems like such a waste of breath to me. It's here and it's not going anywhere. So you can give up, ignore it, or work with it.

ANYWAY. What made me make this thread was a document I received this morning. Yesterday I had a zoom call with two of my clients and we reviewed a bunch of work and they gave me their feedback, all very usual. They had their AI companion on the call, it sent me a summary of the call. Now, this call lasted about 2 hours and I thought it was going to be a good, clean transcript of the whole thing.

But instead, it had summarized the whole thing one page. There was a paragraph that gave a very accurate overview of the whole call, then broke it down into about 5 different topics, and had bullet points for each topic, capturing who said what, what we looked at, and made note of agreed-upon dates for when certain things would be done. It's brilliant.

And it got me thinking about one of things I hate most about working at a design agency that I don't have to deal with as an freelancer, and that is middle-management's love of regurgitation. You have a client call, the designer wants to roll up their sleeves and get on with it. The project manager, or some other warm body standing between the client and work, says no no no. We have to summarize our learnings. So instead of getting to work on what the client wants, first we have to create a Word document that captures everything they said. Now we need to create a branded Powerpoint that makes it pretty. Now we send it, then we wait for them to say "Yep, that's what I said," then we can actually start the work. And we get to charge the client 7 hours for that, so that's good.

So to wrap up this lengthy and meandering first post: AI is good for many reasons, but one of the best is that it will automate the tasks of otherwise fairly useless people who have made careers out of positioning themselves as the middlemen between the people who do the work and the people who want the work. I think artists should be concerned, but I think middle managers everywhere should be shitting themselves immediately.

Edit: I realize that last part comes off as bitter about agencies, but I am bitter about agencies, so take it for what it's worth.

[Alberon]

It's always hard to predict where new technologies like Machine Learning might go as there could always be an unexpected brick wall just around the corner.

But generally, yeah, people are looking at graphic designers and copywriters but many admin jobs are equally under threat. It's hard to not see them going the way of the typing pool in the fairly near future.  There are AI chat support bots right now and they're utter shit, but that won't be the case for long. Many (most?) call centre jobs will go.

Technology will be making inroads into middle management in a way not seen before.

[Sebastian Cobb]

It's good for detecting and finding patterns at scale.

Some positive stuff is it's supposed to be quite good at reliably finding small cancerous cells that humans might miss.

I'm guessing the LLMs are better at translation than traditional probabilistic translation, which admittedly is a lot better than it used to be. Of course this can be a touchy subject, I wouldn't trust it over a human translator for anything important ever and I think that's a bad application of the technology, but as a layman, it can be very handy to translate the odd webpage or even run foreign subtitles for untranslated tv shows/films, for small-scale stuff like that it's not like I'd be paying a human translator to do the job in the technology's absence.

[TrenterPercenter]

Yes, you experienced the actual practical applications of machine learning, something that gets lost in the hysteria that generally surrounds AI.  These things have been around for some time it's only recently they've breached the public consciousness because of LLMs like ChatGPT with the reasons being now a lay person has access computing power that would only be available to coders plus this ability to convert language into relational data allow for massive training environment i.e. the internet.  Learning how to operate these systems is important and the positive applications of them are huge, for example automated diagnostics in a world were not everyone can see a doctor for an assessment, and these ML systems can more accurate than actual human doctors.

They won't end the working world, just the same way calculators didn't end accountancy, they'll just change it in various ways.

[Shaxberd]

My workplace are currently getting very excited over AI tools but I'm not sold on the generative text stuff. Me and a couple of colleagues tried out the meeting summary stuff and it was ok, but missed some key points.

I work for a local council, so a lot of the serious meetings are still going to need a human to make sure the meeting minutes stand up - although I'm certainly in favour of using transcripts to help with that, it's easy to miss things if your mind wanders during a long session.

I can see value in science and tech research in using machine learning to detect patterns better than humans can, though. And seeing as Google has got worse in the past few years, I've heard ChatGPT et al's knowledge banks make them quite handy for looking up things like Excel formulas and bits of code without wading through old Reddit and substack posts where someone posts "don't worry I solved it myself" without ever saying how they fixed the same problem you're having.

[touchingcloth]

There are AI chat support bots right now and they're utter shit, but that won't be the case for long.

Who's to say that's so, given

there could always be an unexpected brick wall just around the corner.

?

It could be less a brick wall, more a wall of the limitations in reproducing human skills in Turing machines, or doing so in ways that use economically viable amounts of fossils, minerals, and water.

[TrenterPercenter]

I've heard ChatGPT et Al's knowledge banks make them quite handy for looking up things like Excel formulas and bits of code without wading through old Reddit and substack posts where someone posts "don't worry I solved it myself" without ever saying how they fixed the same problem you're having.

ChatGPT will even write the code for you.  I would say the best thing about them currently is as learning aids.  They are excellent teachers in that you can ask it to present information in a variety of ways to aid your understanding - like having an infinite set of examples of specific problems that you can tailor to your own preferred learning style.  This is the way to use them as a supplementary aid to self improvement - you see how they will eventually be incorporated into wearables you will be able use them on the fly.  Augmentation to human learning will be a big thing in the future I expect.

[QDRPHNC]

What I found really impressive about the transcript was how intelligently it summarized things. It didn't just repeat the words that were used during the call, but broadly captured minutes of back and forth conversation (including interruptions, filler words, offhand comments, etc.) really well, ie "Client A and Client B expressed satisfaction with the direction and were positive about the refinements to XYZ which were suggested by QDRPHNC and which will be discussed in more detail on Monday."

But I take your point, @Shaxbeard, it may well miss some of the finer or more nuanced points of a conversation, and obviously it's not very good (yet) at understanding tone of voice or irony / sarcasm.

Was reading about that Rabbit R1 - likely that something like that will replace phones as the default personal device. Something like Google, or even your phone app, will be relegated to a background process that the user no longer has to directly interact with. In the same way that Gen Z are becoming less and less familiar with the mouse and keyboard type of interaction, eventually it will seem amusingly antiquated that you had to "go to" a "web site" to "look" for something.

[Alberon]

It could be less a brick wall, more a wall of the limitations in reproducing human skills in Turing machines, or doing so in ways that use economically viable amounts of fossils, minerals, and water.

I do feel the bar is low enough for chat bots that they will make the leap we've seen in the production of graphics over the last couple of years.

At the moment I think what we'll see, say in the production of media for instance, is fewer staff reviewing and editing AI produced content.

And while it could all change no one has won betting against the continued increase of computing power available over the last fifty years.

[Sebastian Cobb]

What I found really impressive about the transcript was how intelligently it summarized things. It didn't just repeat the words that were used during the call, but broadly captured minutes of back and forth conversation (including interruptions, filler words, offhand comments, etc.) really well, ie "Client A and Client B expressed satisfaction with the direction and were positive about the refinements to XYZ which were suggested by QDRPHNC and which will be discussed in more detail on Monday."

But I take your point, @Shaxbeard, it may well miss some of the finer or more nuanced points of a conversation, and obviously it's not very good (yet) at understanding tone of voice or irony / sarcasm.

Was reading about that Rabbit R1 - likely that something like that will replace phones as the default personal device. Something like Google, or even your phone app, will be relegated to a background process that the user no longer has to directly interact with. In the same way that Gen Z are becoming less and less familiar with the mouse and keyboard type of interaction, eventually it will seem amusingly antiquated that you had to "go to" a "web site" to "look" for something.

SUMMRY is pretty good at summarising large paragraphs of text, it's been around longer than these LLM's and as far as I can tell uses a statistical algorithmic approach rather than ML.

SMMRY (pronounced suh·muh·ree) was created to summarize text.

SMMRY's mission is to provide an efficient manner of understanding text, which is done primarily by reducing the text to only the most important sentences. SMMRY accomplishes its mission by:

Ranking sentences by importance using its core algorithm.
Reorganizing the summary to focus on a topic; by selection of a keyword.
Removing transition phrases.
Removing unnecessary clauses.
Removing excessive examples.

[BlodwynPig]

I'll be glad to be dead thanks to AI being so pernicious. We are fast becoming reliant. Human exceptionalism driving nature to collapse and hoping AI will smooth the descent.

[touchingcloth]

And while it could all change no one has won betting against the continued increase of computing power available over the last fifty years.

There's a school of thought that says Moore's Law has slowed down, and the school of physics says it will come to a final stop eventually. The hard problem of consciousness isn't resolved, and it's possible that consciousness won't be reproducible in machines, or won't be economically producible given the limitations like Moorer's Law.

Of course, a lot of this depends on what specifically is meant by "artificial intelligence", and on whether consciousness is a necessary component of intelligence.

[QDRPHNC]

SUMMRY is pretty good at summarising large paragraphs of text, it's been around longer than these LLM's and as far as I can tell uses a statistical algorithmic approach rather than ML.

Cheers, never heard of SUMMRY. Maybe this just seems all new to me, because in many ways I'm quite a tech-phobic person. Not that I'm anti-tech, just that working in an industry for almost 30 years that's so obsessed with "innovation" and "transformation", I've seen so little of it which is truly useful — mostly answers in search of problems.

[Milo]

Most of my job is producing microscope slides for biological research. I can see AI potentially increasing the demand for my work as it might make it more practical to analyse slides in bulk, making different types of experiment viable.

[Alberon]

There's a school of thought that says Moore's Law has slowed down, and the school of physics says it will come to a final stop eventually. The hard problem of consciousness isn't resolved, and it's possible that consciousness won't be reproducible in machines, or won't be economically producible given the limitations like Moorer's Law.

Of course, a lot of this depends on what specifically is meant by "artificial intelligence", and on whether consciousness is a necessary component of intelligence.

A lot of people have been surprised of what machines are capable of with language and images without a single shred of consciousness.

Computers as they currently are probably are not the future. I've seen nothing to suggest we can't ultimately reproduce neural nets on a level of complexity, or eventually exceeding, of the human brain. There could be more efficient designs that step around the baggage of hundreds of millions of years of blind evolution.

[touchingcloth]

My workplace are currently getting very excited over AI tools but I'm not sold on the generative text stuff. Me and a couple of colleagues tried out the meeting summary stuff and it was ok, but missed some key points.

Some people got very excited at the idea that AI tools could create reports from the data in our CRM system and other data sets, but it produced garbage. The trouble with doing analysis on our data typically isn't that it's difficult analysis per se, but that the information architecture is dogshit, which is a very human problem for the foreseeable. Computers can take it over when they get to the point of being able to interview humans and ask "this field here that says 'net profit' - is it actually storing net profit, or is it just some numbers you spammed in there once?"

[Shaxberd]

Some people got very excited at the idea that AI tools could create reports from the data in our CRM system and other data sets, but it produced garbage. The trouble with doing analysis on our data typically isn't that it's difficult analysis per se, but that the information architecture is dogshit, which is a very human problem for the foreseeable. Computers can take it over when they get to the point of being able to interview humans and ask "this field here that says 'net profit' - is it actually storing net profit, or is it just some numbers you spammed in there once?"

Yeah, AI's inability to "know what it's doing" is an issue right now.

I was on a demo earlier this week where IT were showing off Microsoft's AI tool in Edge browser. One feature I liked was that it provided citations for statements it made in text so you could verify what it was saying.

One less good feature was its attempt at reverse image search. The demonstrator was trying to get it to identify a statue from a photo and provide more information. The picture she used was the Robin Hood statue outside Nottingham Castle, but the AI kept thinking it was a different statue that's in Ireland, and then claiming this Irish statue is in London. Nice idea but more work needed there.

[Sebastian Cobb]

There's a school of thought that says Moore's Law has slowed down, and the school of physics says it will come to a final stop eventually. The hard problem of consciousness isn't resolved, and it's possible that consciousness won't be reproducible in machines, or won't be economically producible given the limitations like Moorer's Law.

Of course, a lot of this depends on what specifically is meant by "artificial intelligence", and on whether consciousness is a necessary component of intelligence.

It has apparently slowed due to chip manufacturers struggling to go below certain nanometre size-limits quick enough. Although I think Moores law kind of still tracks for parallelism instead, I've also seen people claim it tracks for overall cpu/memory you may have spread across multiple devices although that sounds like moving the goalposts a little too much for me.

However it looks like there's some groundbreaking processes coming along anyway that might make making ever-smaller transistors less challenging.

https://www.nature.com/articles/s41586-023-06811-0

[Video Game Fan 2000]

i like the idea that in three or four years, AI could optimise energy grids and distribution to make both batteries more feasible and allow a crucial extra few years to decarbonise

for my own work - id like to tell an AI to dig for similar phrases or the origins of ideas across different texts and compare them, including in languages i dont read so well. i really dislike how 'big data' and numericisation is currently used in the study of literary and historical texts, its borderline pseudo-science to me the way people dig for word-usage and create diagrams and charts, and i think AI could realistically allow us to actually do what some academics have been only pretending to do for the last two decades

[Video Game Fan 2000]

it would be cool if AI killed the entire discipline of sociology too but lets not hope for too much

[touchingcloth]

A lot of people have been surprised of what machines are capable of with language and images without a single shred of consciousness.

I'm not sure if anyone has good empirical reasons for extrapolating this into new capabilities. Large Language Models can do some interesting and surprising things, but does throwing more training data and compute at them mean that how good they are at those things or how many novel kinds of things they get good at scale linearly without hitting hard barriers? Back in the 60s people extrapolated the decreasing costs it took to launch a kilogram into low earth orbit and predicted that we'd be launching for pennies now, but in hindsight we hit some limits and the steep early decreases flattened off. As someone else said in another thread where AI was discussed

AI optimists vs AI doomers. They think they're radically opposed. In reality they're just having a family dispute, and haven't stopped to consider that both their respective "beliefs" are built on an incredibly shonky paradigm. Mention S-curves to either faction and they'll both get very angry at you.

Computers as they currently are probably are not the future. I've seen nothing to suggest we can't ultimately reproduce neural nets on a level of complexity, or eventually exceeding, of the human brain. There could be more efficient designs that step around the baggage of hundreds of millions of years of blind evolution.

I have similar thoughts on this, really. What are the limits of complexity and efficiency? Is scaling both of those things sufficient to make an AI that is the equivalent of a human? If it's not the equivalent of a human that we're after, then to what end are we pursuing AI given that computers have long exceeded human capabilities in certain domains, e.g. pocket calculators beating mental arithmetic.

[Milo]

Is there something in the idea of AI being enslaved? If you get general artificial intelligence then you'd either need it to want to do work for you or you'd have to force it?

[Alberon]

In principal you could have specialised neural nets for specific tasks. If a true General AI is made equal, or above, a human intellect than that opens a whole can of worms about sentience and rights.

My own personal feeling is that if you have an AI neural net modelled on a human brain, then it is human as we see the universe in a specific way that it would follow. But the whole issue will get very messy as many will claim AI doesn't have a soul or is only mimicking sentience (though if it is doing it perfectly then surely it is as sentient as we are, assuming we are also in the first place).

[greenman]

I'm a designer and am of course worried about the impact of AI on my job. I'm not as worried as a lot of people because frankly I'm a good designer, and while AI will eventually come for my job, it has to work its way through the "I'll get my nephew to design the logo, he has Photoshop" level folks first.

And I've resisted demonizing it, because I think it's genuinely pretty amazing, and it can be used by artists in really interesting ways that I don't think have been fully explored yet. I also think that, in much the same way the internet itself removed the barriers to entry for getting your creative work in front of an audience, AI is taking this a step further and allowing people like myself, who want to create things on zero budget, opportunities to make the sort of work that would previously have been out of reach.

And of course, it's not going anywhere. I understand the copyright issues and the fears of writers, illustrators, designers, but it all just seems like such a waste of breath to me. It's here and it's not going anywhere. So you can give up, ignore it, or work with it.

ANYWAY. What made me make this thread was a document I received this morning. Yesterday I had a zoom call with two of my clients and we reviewed a bunch of work and they gave me their feedback, all very usual. They had their AI companion on the call, it sent me a summary of the call. Now, this call lasted about 2 hours and I thought it was going to be a good, clean transcript of the whole thing.

But instead, it had summarized the whole thing one page. There was a paragraph that gave a very accurate overview of the whole call, then broke it down into about 5 different topics, and had bullet points for each topic, capturing who said what, what we looked at, and made note of agreed-upon dates for when certain things would be done. It's brilliant.

And it got me thinking about one of things I hate most about working at a design agency that I don't have to deal with as an freelancer, and that is middle-management's love of regurgitation. You have a client call, the designer wants to roll up their sleeves and get on with it. The project manager, or some other warm body standing between the client and work, says no no no. We have to summarize our learnings. So instead of getting to work on what the client wants, first we have to create a Word document that captures everything they said. Now we need to create a branded Powerpoint that makes it pretty. Now we send it, then we wait for them to say "Yep, that's what I said," then we can actually start the work. And we get to charge the client 7 hours for that, so that's good.

So to wrap up this lengthy and meandering first post: AI is good for many reasons, but one of the best is that it will automate the tasks of otherwise fairly useless people who have made careers out of positioning themselves as the middlemen between the people who do the work and the people who want the work. I think artists should be concerned, but I think middle managers everywhere should be shitting themselves immediately.

Edit: I realize that last part comes off as bitter about agencies, but I am bitter about agencies, so take it for what it's worth.

Honestly it seems to me QDRPHNC that the focus on "AI Art" is somewhat about marketing, the idea AI can potentially replace a human in work viewed as very creativity focused I think naturally tends to seem more impressive.

I spose it has direct applications in terms of various net tools as well but I do kind of suspect that the impact of AI on the workplace will be more along the lines you point out. Basically taking over office work tasks that previously had just about been able to justify retaining a human because automation may have been a bit too expensive.

That seems like the potentially most disruptive aspect of AI to me, potentially taking away or devaluing very large numbers of office/middle management jobs. Seems like that could be a very large shift in society/politics since I think you could argue up until now a good percentage of the population with such careers being just comfortable enough to be resistant to change was vital to support the status quo.

If you suddenly have millions of people on the scape heap as happened in industry is that going to be sustainable?

[Dex Sawash]

You can call me Al

[jamiefairlie]

In principal you could have specialised neural nets for specific tasks. If a true General AI is made equal, or above, a human intellect than that opens a whole can of worms about sentience and rights.

My own personal feeling is that if you have an AI neural net modelled on a human brain, then it is human as we see the universe in a specific way that it would follow. But the whole issue will get very messy as many will claim AI doesn't have a soul or is only mimicking sentience (though if it is doing it perfectly then surely it is as sentient as we are, assuming we are also in the first place).

Humans don't have souls either though so we can put that to the side.

I don't think it can have sentience in the human sense as I believe that is a biological body phenomenon. However I'm fairly sure we'll be able to grow programmable humans from scratch at some point, which will be interesting to say the least.

[Head Gardener]

[touchingcloth]

I do kind of suspect that the impact of AI on the workplace will be more along the lines you point out. Basically taking over office work tasks that previously had just about been able to justify retaining a human because automation may have been a bit too expensive.

That seems like the potentially most disruptive aspect of AI to me, potentially taking away or devaluing very large numbers of office/middle management jobs.

Given how companies already weaponise algorithms against their workers (e.g. back in 2020 - Uber sued by drivers over 'automated robo-firing'), I could see "AI" being used to crack the whip some more. TAIlorism.

Why simply force you warehouse workers to piss into plastic bottles rather than risk punishment for going to the bathroom when you can add some cameras and around the workfloor to monitor people's eyeballs and dock their pay if the AI notices that their gaze has wandered, or make your staff wear headbands with alpha wave detectors that can be analysed for patterns associated with thoughts of unionising?

[Alberon]

Humans don't have souls either though so we can put that to the side.

Oh yeah, but you can bet many people will make that argument.

I don't think it can have sentience in the human sense as I believe that is a biological body phenomenon. However I'm fairly sure we'll be able to grow programmable humans from scratch at some point, which will be interesting to say the least.

I'm not convinced being biological has anything to do with it. My own thought on the subject is that it doesn't matter what the neurons are made of it's the connections between them that do.

[jamiefairlie]

Oh yeah, but you can bet many people will make that argument.

I'm not convinced being biological has anything to do with it. My own thought on the subject is that it doesn't matter what the neurons are made of it's the connections between them that do.

Yeah but the more we understand of the human ecosystem the more it seems that other systems outside the brain are also involved e.g. the gut microbiome effect on emotions

That's not to say it's impossible to map but extremely complex