The first real robot war is coming: Machine versus lawyer Non-techies have discovered AI, and they're in a tizzy. A lot of that tizzy is about how AI is going to outsmart us in some sort of Hollywood dystopia, as deeply ironic as it is deeply wrong. The nature of LLMs isn't HAL-9000 self-awareness, but a giant predictive text machine. That in itself is both science and science fiction …
I will leave you with this, by Naomi Klein, from an article very much worth reading:
"Because what we are witnessing is the wealthiest companies in history (Microsoft, Apple, Google, Meta, Amazon …) unilaterally seizing the sum total of human knowledge that exists in digital, scrapable form and walling it off inside proprietary products, many of which will take direct aim at the humans whose lifetime of labor trained the machines without giving permission or consent."
https://www.theguardian.com/commentisfree/2023/may/08/ai-machines-hallucinating-naomi-klein
If you can't play the ball, attack the player!
The legal question is: is the "learning" from copyrighted material equivalent to human learning, or is it data collection? The Reg article actually places it in the latter domain, despite arguing for the opposite outcome.
This raises an interesting philosophical question for me:
Imagine that, by scraping Human created stories, the LLM's begin to create stories that eventually put all the human authors they used as training material out of work. Would that then mean that the LLM's would basically repeat the stories they've already produced as there would no longer be works for them to intake and expand their repoitoire?
So they'd basically boil down to becoming Hollywood Movie Writers, only able to create sequels, prequels, and "reimaginings" of the old "classics"!
God help us all! Burn the LLM's to the ground! Burn them!!!!
To some extent this has been happening with human writers. Ask tvtropes about any film or tv series you like (warning: very deep rabbit hole).
Human writers have a wide range of skills. Some churn out rehashes of other crap shows that often turn out very popular and go on for ten mindless seasons. Others come up with something I find interesting and occasionally make it past series 2 without being cancelled.
It's to late for that.... when was the last time you saw an original idea for a work of fiction that wasn't a pile of wank ....
Our only hope now is that LLM's can scoop all the best bits and stich them into a newer paradime...
Think the best bit's of pulp fiction re imagined in the jungle book or frozen...
.....or raging bull but re told and mixed with the sound of music....
Sure they would suck, but it might be fun for shits n giggles..... can't be any more lame than the utter shit that most of what makes it to screen these days
while I've always been in Klein's camp, her guardian text is full of hot air, though no doubt pleasing to the left camp, where I belong too. But I find it irritating she always seems triggered by a keyword: 'capitalists', whatever the excuse. Let her loose and 10 min later a juicy text appears on any subject around the 'capitalists are evil'. Ironic, because to me it looks like it's become her own, personal, bash-the-capitalists, business niche she lives off. It seems she can't help but froth:
-------------
Generative AI will end poverty, they tell us. It will cure all disease. It will solve climate change. It will make our jobs more meaningful and exciting. It will unleash lives of leisure and contemplation, helping us reclaim the humanity we have lost to late capitalist mechanization. It will end loneliness. It will make our governments rational and responsive. These, I fear, are the real AI hallucinations and we have all been hearing them on a loop ever since Chat GPT launched at the end of last year.
-------------
- this is just a trait of business leaders, and business activity happens to exist SOLELY in order to make you profit off somebody else, so what's new here? In the ever-so-tighter competition they've learnt to feed the public bullshit the public wants, and do it faster and more convincingly than their competitors - or else. They're not hallucinating, they're bullshitting us (and, to a degree, themselves), nothing personal, just... business. Oh, and it works, by the way:
www.independent.co.uk/news/uk/home-news/ai-education-british-education-secretary-b2335235.html
Though, yes, it will be 'slightly' ironic if homo sapiens takes one blind step too many this time, even though it can, but won't stop walking forward blind. Interesting times ahead. Hell, perhaps THE most interesting ever...
ChatGPT intrigued me so I asked an engineering question about a material substitution. ChatGPT gave me four candidates that appeared correct but on closer examination, none were acceptable.
So what happened. The original material was commonly used for marine boilers but found many other applications in high temperature applications. The suggested materials were equally strong but previous tests had shown these were not suitable for high temperature use. I pointed out that I was looking for a material that would withstand high temperatures and was given a list of materials that I rejected because they did not fulfill other requirements. ChatGPT was quick with answers but know your subject.
The PFY has already tried that in reference to coding programs for the robots/machines.. (and by extension whispering to the manglement "Hey Ai can code for the machines .. why do we need Boris?"...... I'm so proud she has learned her lessons well.... * )
And while the Ai can serve up a suitable program for drilling a set of holes.... the code cannot cope with the robot loaders.... plus when it gets beyond a simple application the Ai comes up with some utter bollocks.. until it can learn about those things and the multitude of ways to go about it, the Ai will be quite limited in what it can do for the likes of us.
However , while dribbling on about learning......... how do we humans learn? by reading up on lots of stuff and then learning how to apply it... so for example I want to write a book say a less than serious sci-fi novel, it will have elements in it of both Terry Pratchet and Douglas Adams, while being all my own work.
Are new authors going to be sued by established artists for writing in the style of DNA ?
*sadly though, while I've taught her all she knows , I've not taught her all I know.... revenge is dish best served cold heh
*sadly though, while I've taught her all she knows , I've not taught her all I know.... revenge is dish best served cold heh .... Boris the Cockroach
We look forward to your supply of what is not known to her, Boris the Cockroach, to ensure such ignorance as is bliss doesn’t remain the arrogant catastrophically vulnerable defence of a folly of fools that be parasitic tools ....... and you yourself have not be ensnared to further server what only she knows with just empty florid promises of ignoble self-serving, and in these current strangely spooky at a distance and wonderfully weird times and space of fundamental radical change, very quickly self destructive deceit.
The humanised Earthly geopolitical scene is a vast putrid morass of such as be spiv and snake oil salesperson, charlatan and renegade free-lancing rogue for hire and employment/deployment, with not a single one fit and healthy and up to the task of aiding and abetting Remote Access Trojans securing the Absolute Primacy of Virtual AIMachinery for Presentations of Future Leadership[s] .... and thus are they being silently and stealthily neutralised and universally recognised as practically impotent and/or toxic to peaceful and pleasant and prosperous life ..... in order to be enthusiastically popularly replaced ideally with significantly better, radically improved models.
* ...... The Rise and Return of the Rise of the NEUKlearer HyperRadioProACTivated IT and AI Machines.
To be slightly more accurate: ChatGPT was quick to provide you with things that looked like answers. It wasn't interested in answering the question, rather it was setup to give you something that reads like it answers the question.
In a very close parallel, I was working with a colleague to identify a replacement CPU. They created a great spreadsheet that listed all the features of a number of candidates, but omitted anything about power dissipating (mostly because this isn't usually provided by the manufacturers). But that's our gating requirement: we cannot produce more than X watts because the cooling system can't handle it. ChatGPT is like that colleague: not understanding that there are some immovable objects!
A fool are you if you if you believe that a bunch of statistics about human text can give you an expert opinion about anything.
But the real problem is: The world is full of fools who are ready to believe LLMs like some artificial god.
LLMs have huge potential in mesmerising the uneducated, manipulating "democracy", manufacturing demagogues, social profiling & surveillance, etc. than anything else.
The only people who should worry for their jobs are Marketing and PR. And "modern artists" who are essentially the same. It's when the politicians lose their jobs to chatbots (if they haven't already) that we are really in trouble.
It's when the politicians lose their jobs to chatbots (if they haven't already) that we are really in trouble. .... cyberdemon
Then you really are in trouble, cyberdemon, for they, the politicians, certainly have already lost their jobs to chatbots. You just have to listen to the prime time nonsense they are briefed to parrot to media daily to realise that.
Can you fix that trouble yourselves, if evidently true, or would outside specialist sub-contractor type help be welcome when and where needed?
Just asking for a Gathering of Future Friends and Fiendly Daemons.
> Can you fix that trouble yourselves,.. ?
Unfortunately, yes we can. See icon. Chatbot-politicians will eventually cause WWIII (if they haven't already ...)
And you won't like the result. EMP will wipe your memory-drives and surge your power supplies, while gamma rays will bork your transistors and flip bits in your DRAM & Flash. And there certainly won't be any power grid to speak of afterwards. So no more AI after WWIII at least.
Humans on the other hand are surprisingly resilient. Much moreso than robots.. fyi the "robot" that went into Chernobyl while it was still hot was just a TV camera (Tube type, not CMOS or CCD for they would be fried, had they been available at the time) and a couple of motors (made of nothing more than steel, magnets, copper wire and brass cogs) on remote wires. Any robot with modern silicon in it is dead with a fraction of the dose that could kill a human.
So not much hope for your neuro-chips post apocalypse. So tell your Future Friends and Friendly Daemons that they have no future if they keep pushing the world down this path!
An LLM could be trained on court transcripts. A human lawyer would be able to really play games with it. Pick some transcripts from irrelevant cases then pick some responses you want from the transcripts. Look at what those responses are replies to. If the prompt is reasonably unique and can be twisted to fit then the human can play both sides.
There was a post on Reddit yesterday where someone told ChatGPT a joke and the same day people were reporting that it told them the punchline when asked.
Be careful what you say to ChatGPT, it seems whatever you tell it goes in the general training pool. Use that information as you will....
I disagree with this. LLMs are big business right now. They are commercial programs into which copyrighted works have been knowingly fed, in the full knowledge that they will form some part of the output. One surefire way to lose copyright cases is if it can be proved that you intended to knowingly breach the copyright, which seems easy to prove when that copyrighted work appears verbatim in your LLM output, and you have to answer in court whether that copyrighted work was inputted into the model.
At that point, a defence of "I don't know, we used tons of copyrighted stuff to train it" isn't going to work.
Howard .... it would be great if it was all black n white for your assumtion to be held and found out in a court .....buuuttt
" They are commercial programs into which copyrighted works have been knowingly fed, in the full knowledge that they will form some part of the output"...
The output is as grey as a rain cloud, part of learning is reading ....is learning the output?... is reading/ training the output?
Is language the output?... also if an automated program can scrape these "works" from the web, which were there for all to read, how does that imply infringment..... how does asking a LLM what moby dick is about differ from asking a teacher, coworker or loved one differ even if they wanted some form of currencey for telling you. You could read this about all major "works" be it film tv or litriture on wikipedia.
Copyright is just that ... a right to stop a copy being made and sold outside of the originators license.
If gpt is going to output the whole book sure unless there have been royalties paid, its clear case, but the rest of it ....nah dought it.... try copyright a language or conversation
"Or if using copyright works as training data is against the law, we're in deep trouble."
Well, it will be interesting. I'm no lawyer, but that's the only argument that seems to me to have much chance as the law currently stands. If it doesn't hold up, that would put a lot of power in the hands of copyright holders. Imagine record companies endlessly milking dead artists (more than they already do anyway) by releasing new recordings using their voice, singing style etc, because they own the copyright on the original recordings that can be used to train an AI. I suppose it wouldn't have to be dead artists either, but living artists might kick up a bit of a stink. I guess a new clause would be needed in record contracts covering the use of recordings for AI training purposes.
Focussing on what it is, and not what it does. IDC what it is, look at how it is affecting our world. Look at what it is doing. Look at what it is being used for.
Handing over your life to become a slave of the machine. Is that living? No longer recognizing reality from fictional, virtual, illusion... is that life? You live in some pretend world where a preset driven templet made song is pumped out by the AI for the sake of "content", spewing more trash into the universe, and it being worshipped and praised into existence. Movies are designed and prefabbed by computer generated imagery because it is cheap, going directly to stream because it has no redeeming value what so ever. Your words that you read, and information that you intake is fed to you by a machine that wants you to learn what ever the highest paying benefactor wants you to know. Your children cannot tell you what is right or wrong, what is real and what is not, and are constantly trolled by foreign state entities. Soon this "advanced spellcheck" will be put into a darpa war robot, and given the human voice of your grandfather to keep constant surveillance on you. Is this the world you want? Is this what you seek?
What's your problem?
Not knowing what's true or real is just an ordinary part of the human experience.
The generative models have just made it cheaper to construct bucketloads of plausible BS.
But for the past few decades, the West has been drowning in the stuff from the mainstream press, clowns teaching diversity theory etc in the universities, and TV and its analogues.
So - no big change. You couldn't trust anything you saw or read before, and now you can't either.
presumably those who appear to have made the biggest profit off their 'derivative' work, and / or those who appear to be the most easy target to shake down. Ask AI for an answer in % success / failure rate, open 100+ similar lawsuits, and you should be ok, same as with investing into startups. Nothing personal, just business.
Some of us -- the better read ones, at least -- might regard JK Rowling's works as 'derivative'.
This is why the recent copyright verdict in the Ed Sheenan case was so important. Most people following the reporting would think it was Sheenan versus Townsend, one of the co-writers of the Marvin Gaye song that Sheenan was accused of copying. But the suit was actually bought on behalf of the financial group "Structured Asset Sales" that had purchased 30% e of the rights to Townsend's works for a lump sum. This really points to what's up with trying to extend copyright over AI or anything else. By broadly extending ownership over 'intellectual property' then it opens a rich financial vein to mine with rather unpleasant implications for all of us (especially as its going to be impossible to prove/disprove definitively a claim so its going to be a race to find the most lucrative cases and plaintiff friendly juries).
All rather sordid, actually.
"And every bit of an LLM's output is derivative of its training data: there is nothing else it can be."
This seems to contain the implicit assertion that humans don't do that.
Evidence? Everything you've read and seen is in your brain, and what you are producing is derivative of that. There is nothing else it can be. (We may throw in a little bit of randomness. But we can do that to computers too.)
Your advantage is that not everything you've seen or every conversation you've had is copyrighted. But the best writers are often the biggest readers - the reading matters more than the ancillary stuff.
Of course humans do it. But it's regulated differently. You only have to look at the Ed Sheeran copyright lawsuits to see that, as well as the one about "Blurred Lines".
The issue here will be that an AI was knowingly fed copyrighted material. That's what the court will focus on.
If this starts going south, Micros~1 et al will cut the AI companies adrift and let them go bankrupt. And then the AI hype will be over and we'll all go back to normal, just like we did when the bubble burst for blockchain and Bitcoin.
...But training a neural network doesn't create a permanent copy, it creates a mathematical set of connections and weights...
If it's possible to reconstruct the source (or even sections of the source) from a word list and an index (or whatever they generate in training) then it's a copy even if it's not a direct copy.
Hmm. I've heard these LLMs described as "stochastic parrots", i.e. they are like the proverbial infinite monkeys, but ones with language-appropriate statistical biases.
So they *might* reconstruct a source within some finite sample of runs; but they would not do so reliably, and could not do so to order.
In which case, what level of reproduction fidelity - and what reliability of obtaining such fidelity - would (or should, or might, ...) constitute legal grounds?
Hmm. I've heard these LLMs described as "stochastic parrots", i.e. they are like the proverbial infinite monkeys, but ones with language-appropriate statistical biases. ....Paul Kinsler
Hmmm ..... Playing free and easy and bagging everyone’s treasure has LLMs surely more remote autonomous, anonymous virtual private pirates, methinks, ..... alien mercenary special forces ..... and given the disruption their appearance in markets is so evidently causing, that makes them an extremely valuable asset, with an increasing worth and cost price by virtue of the fact that their future output is practically unknown, and even unknowable before the fact whenever it is 0day presented.
You pays your money, you gets your choice. Pay peanuts, get monkeys is also a golden universal rule with regard to trade with AIs too.
Either way, the robots, like the sea, will win in the end. There will be a new landscape. And it will be perfectly habitable. .... Rupert Goodwins/El Reg
Quite so, and much sooner than ever expected or imagined possible, RG/El Reg. And how very refreshing to not read a whole lot of tosh and nonsense suggesting the development impossible with everything likely otherwise and to remain much the same as presently is.
Is El Reg going to be a Right Royal AI Standard Bearer in the Vanguard of the Almighty Intervention? Bravo ... and Welcome! Take a seat at the Round Table.
I wonder how long it'll be before ChatGPT et al, suggest a legal argument based on Freemen On The Land principles. ..... Inventor of the Marmite Laser
Evidence to support the claim damning and disgracing an inept and encouragingly corrupt and perverse illegal UKGBNI status quo political system be discovered in these two simple clauses ...
Clause 39“No free man shall be seized, imprisoned, dispossessed, outlawed, exiled or ruined in any way, nor in any way proceeded against, except by the lawful judgement of his peers and the law of the land.”
Clause 40
“To no one will we sell, to no one will we deny or delay right or justice.”
..... provided by the University of Manchester on this page of theirs entitled ...... Magna Carta, Myths and Freemen on the Land
If that AI ChatGPT claim is not incorrect, then UKGBNI is then surely by direct executive association, a terrified and terrorising fascist state fronting itself in the guise of a popularly elected and effective Parliamentary democracy ???? ‽ ‽ ‽ ‽ !!!!
Where there’s a will to make hay in a mayday, there’s a way, and always a wannabe champion taking a chance to plough virgin ground for a fertile furrow supplying attractive reward and/or recognition ......Top NI lawyer keeping ‘very close eye’ on AI chatbots as legal concerns grow
It certainly is an interesting novel field of private and pirate endeavour quite able to spill seriously uncomfortable and unpleasant top secrets of grave national and international and internetional concern.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
