To worry that there may be no hope for a good future thanks to AI

[Designless]

I use it, it up skills me a lot, I am at the top of my game but.... I think I'll be lucky to reach retirement age still in work and I despair for young people trying to get entry level jobs. Everything that I did to get on the ladder is done by AI now.

I know the nebulous cope response is "that's what the luddites said - NEW jobs will arise" but I think this is different. AI can think. AI allows a handful of unbelievably wealthy people to control everything.

Someone please post something hopeful before I pop from despair thanks :(

It will change everything.

Yes, it hallucinates, makes mistakes, etc. As do humans.

Its capabilities are accelerating at a remarkable pace though with little sign of plateauing.

GenAI models fail the Turing Test now not because they are less capable than humans but because they are more capable at so many things, thus easy to distinguish. That list of things they are better at us at will grow and grow.

We have been the most intelligent type of thing for all of history. That is about to change forever.

Yes it’s a problem. I suspect capitalism may need to be significantly rethought, with no guarantee that there is a desirable stable equilibrium available.

AI will generate massive consumer benefit. But the monetised bit is going to be hugely concentrated.

(or put differently, AI is likely to increase the returns on capital far more than those on labour)

The optimistic scenario is that the world will get so much richer than even a smaller slice is still good money.

The pessimistic one is that as AI outcompetes us on more and more skills, almost no one can earn more than a subsistence wage.

Then we need to do redistribution. But that only works: I) for countries that have significant AI wealth to tax, II) with the consent of the wealthy.

Maybe the wealthy will become the default government. And you cross your fingers that their idea of a landscape they wish to live in (peacefully) doesn’t require gulags.

Unless/until the ai decides it doesn’t need its human wealthy overlords anymore. And then it could be anything from terminator to a lovely world to live in cos ai like emma.love got networked overlord ai all interested in abstract, human “feeling” concepts.

I really do need to go and pootle in the garden. I can feel the pull of a 48 hour rabbit-hole nose dive coming on.

Nobel Prizewinner Demis Hassabis said the impact will be 10 times bigger than the Industrial Revolution and 10 times faster.

People calling AI a fancy auto complete that can't really think are in denial or haven't used LLMs at any serious level.

If you're not sure, look at the share price falls for software, data and analytics company's since Claude released an updated Cowork agent a few weeks ago.

Gartner down 33% in a month, Thomson Reuters down 30%, Factset down 34%.

Agentic AI can perform tasks and complex processes and is already transforming many white collar jobs. Law, finance, tech, sales, media, data analysis, for example.

A lot of my job involves writing and I can already see what's coming for my role.

Interesting to read here how many people actually think that companies won't change too much in a world of AI and that their role is safe.

What AI do you use in your work place? The frustrating thing is people think AI is ChatGPT and CoPilot when it’s not. It’s so much more now and so much AI being developed to tackle so many problems. I agree with you.

Yes I think that's a real risk.

Here's some rabbit hole reading for you: intelligence-curse.ai/

The idea is: many countries that have natural resource wealth are badly governed, because the elites don't need the people for anything, they just need to be able to co-opt the resource that pays for everything. AI may be like that.

@OttimoMassimo

Hello writer, from teacher. Also know my days are numbered working the way I do now. If nothing else considerable numbers og my clients (white collar professionals in the main) might not have enough spare money to consider my help a necessity anymore.

I love listening to Mo Gawdat, because his voice is lovely and his humanity shines like the sun.

But he convinced me some time ago that this is not something to be unconcerned about even in the shorter term.

I gave you applause, but < curses >

Today will just have to be sacrificed to my inner Alice.

😅

The idea is: many countries that have natural resource wealth are badly governed, because the elites don't need the people for anything, they just need to be able to co-opt the resource that pays for everything. AI may be like that.

The problem with our world is that the vast majority of the planet is already governed by corrupt and authoritarian governments (both go hand in hand). The real issue behind authoritarians is opacity of governance - you just don't know who is up to what because there is no accountability. In a world dominated by AI, accountability totally goes out the window. Unfortunately, AI dominance can only undermine human freedom. That explains techno-fascism, by the way. It's a feature, not a bug.

A response debunking the Matt Shumer "Something big is happening" claim. https://press.airstreet.com/p/reflections-on-reflection?hide_intro_popup=true

This is moderately serious, iteratively trained AI. I am not minimising it. However….

See my post of 18.50 yesterday for a description of perhaps the first, certainly one of the first, objective experiments in setting AI loose on very small open research problems. Actually the solutions had been posted, in encrypted form, for the sake of academic integrity - but not accessibly to AI.

The authors set state of the art AI loose but offered no help. Full results will be revealed tomorrow morning as I described earlier but it is already clear that AI’s performance is very disappointing, to say the least.

I think ‘disappointing in open research problems’ is a pretty narrow definition of AI's impact.

It may not be solving theorems yet, but it is being applied at scale across corporations right now - coding, analysis, customer service, communications etc etc

Companies will usually go with something that is good enough, fast and cheap, rather than perfect.

I can see that there are already plenty of office jobs that agentic AI could replace today - even if it's no good at doing open research.

@GeneralPeter

Still reading but this jumped out at me.

“believe in straight lines on graphs”

I do. But the speed at which entire nations of people either chose, or were obliged, to prioritise Not True, but Kind above True, straight lines on graphs might be a hard sell on a societal level.

I haven’t read far enough yet (there has been some gibbering about becoming a citizen of a future Congo being my fate) to know if enough people believing in straight lines on graphs is fundamental to steering us away from the worst iceberg.

But it occurred to me, if belief in straight lines on graphs IS fundamental, was it pure bad luck and ignorance that suddenly and rapidly took us into politely ignoring reality.

Or was is all part of a cunning plan ?

NB My ADHD does make me prone to seeing a pattern and leaping into conspiracy theory mode. At least temporarily.

I’ll keep reading. After a walk in nature because the meat sack that ambulates my brain is being told by brain that she needs birds, breezes in trees and the odd deer to feel better. I can believe in straight lines on graphs AND feed my human need to feel at one with the universe. So I will.

Thank you for posting the link, my instincts do better if “people who know stuff” are able to articulate my surface level wonderings and let me make more sense of them.

I didn't see this yesterday, so have read it today. Interestingly, Times Radio was discussing it this morning between 8:00 and 10:00.

Because I am retired, I am likely to encounter AI as a consumer rather than as a competitor, but I have been especially interested to read the views of writers and teachers here. I was a writer, but as my specialism was converting the spoken words from private interviews into prose reports for publication, I'd be interested in understanding how AI might inveigle itself into this process, which relied on privacy and confidentiality until the material went into print.

AI is also quite crap at customer service, in my opinion. I can’t think of a time it’s been able to handle any query except the most generic.

That would be fine if it would realise its limitations and pass you to a human. But no, it ties itself in knots answering questions you didn’t ask.

I agree it is decent at basic coding and can be trained to excel at anything based solely on pattern recognition. I think there is incredible potential in certain areas of the medical field, for example. It can already read CT scans better than humans and this is obvious when you think about it: pattern recognition, unparalleled attention to detail and no fatigue factor. What could be better?

Any time you can rigidly apply a sorting algorithm and get a well defined answer, or opt to refer back to a human, AI will excel Knowing when to do the last is vital. But there are still too many times AI cannot account for nuance. That’s when trouble arises.

Agree. I think it’s hugely minimising the applications of recent AI tools to say they are useless at doing open research or solving problems. Just look at DeepMind’s AlphaFold and how that was absolutely instrumental in solving the 50-year old protein folding problem (and for which the Nobel chemistry prize was won).

Protein folding is a profound example of pattern searching. That’s what AI does best, much better than humans.

BTW these were tiny research problems, that’s important.

Sure it’s one example, but the AI tool was able to rapidly solve a problem that had hitherto been unable to be solved by humans. The same will the case for diseases, materials, new forms of energy. These are not “small” problems and they are not the only problems being solved. These are the ones that are, at the moment, published. It’s an exciting time for science.

In today’s Guardian there is a front page article ‘Fears over AI errors in social workers’ records’.

It describes an analysis of AI transcription services, performed by the independent Ada Lovelace Institute, of notes of meetings between social workers and clients across 17 English and Scottish councils.

The AI tools are Microsoft Copilot and the proprietary Magic Notes, which costs councils between £1.50 and £5 per hour of transcription.

Many serious errors were turned up, including fictitious reports of suicidal ideation and the inclusion of lots of absolute gibberish. The AI tools could not cope with accents.

The report highlights the potential for serious harm to clients from errors of both omission and commission by the AI.

These transcription services are used across
dozens of councils, having been championed by HMG last year.

This is not exactly a cutting edge application, or shouldn’t be. Whilst the potential for harm to clients is the primary consideration, it is galling that councils on the edge of bankruptcy or worse are paying for this garbage.

In the absence of objective, third party analysis, we simply don’t know how much of this kind of thing is happening. I’ll put my £££ on more rather than less.

None of my criticisms detract from what AI excels at - anything that can be constructed as a well defined problem in pattern recognition. Protein folding - thank you, @BridgetJonesDaiquiri - and various aspects of medical analysis are two outstanding successes.

The ‘small problems’ were the ones from the list in the paper ‘First Proof’ that I referenced upthread.

Apologies for the confusion.

I haven’t finished reading all of @GeneralPeter ‘s link yet.

But I’m taking a moment to collect myself after “freedom tags” and “manufactured pandemic”.

Currently on on the side of whoever is doing their best to train AI to believe “Human’s are adorable Panda’s, not annoying mosquitoes”.

In the hope we’ll be provided with all the fun, stimulation and care cos Bot-they are hoping we’ll breed (outside of captivity and not with a freedom tag thank you very much) so they can delight in our antics. And we can devote ourselves to our (not hurting other people) passions and purposes.

On another note, I hadn’t actually consider the Matrix version of the future and had never really understood the “humans as energy source” part of the plot before. But suddenly do. Not sure part of which essay made it pop up, but popped it did.

I do not want to be a battery ! Mind you due to the menopause I struggle to produce energy as it is. So I’ll probably get junked in the battery choosing phase.

Does anybody know a pleasant/optimistic future based film (not Back To The Future, still waiting for my hoverboard) I can watch to cheer me up ?

There is another relevant research paper posted yesterday at arxiv.org, in Computer Science/Machine Learning. It is called
‘Towards Autonomous Mathematics Research’ by Tony Feng et al.

The authors have built a custom version of Deep Mind. With human help it solved one moderate maths question. Apparently it devised a method one of the authors had not thought of to then do a research level computation.

The authors then turned their AI tool loose on a well known catalogue of 700 open problems, many relatively small. It claimed a fairly high success rate, nearly 10% IIRC. But upon inspection, most of those ‘solutions’ were revealed to be either vacuous or incorrect.

This leaves the authors claiming that their AI tool solved 13 problems from the catalogue. However, there are issues. Some of the problems are likely to have solutions or partial solutions available to AI and some of the 13 solutions are partial. In the fine print, the AI tool gave complete and, it is thought, original answers to only 2/700 problems for a hit rate of .285%.

As a point of context, it is known that a few of the problems in this catalogue are likely to be difficult whilst some are probably quite approachable. (these are the problems left by Paul Erdos, for those to whom his name means something).

I have seen it suggested that these results are likely to be a slight improvement on the results in ‘First Proof’ (to be revealed tomorrow), probably reflecting the better AI. But nothing revolutionary.

I’ve read most of the thread but not every post so apologies if this has already been referred to but this article posted on X, from someone who works in AI, is very interesting. I’ve hardly used AI but it’s obvious to me how quickly it is developing and improving. I do think anyone who thinks it’s not a threat to their job, and probably the way of life as we know it, is either fairly ignorant or delusional. We need to be realistic and prepared, but I’m not sure how to go about that.
https://shumer.dev/something-big-is-happening

You are right OP, the current youngsters looking for jobs are already starting to struggle, I’m not hopeful, I’m really scared

Already posted and debunked upthread.