Well the the artist’s work is protected intellectual property, I don’t think it is easy for UK government to ignore copyright and just use the property anyway.
The point that the AI developers and AI fans don’t want to understand is that permission must be obtained before anything is done.
It shouldn’t be too difficult for AI developers to have a sit down with copyright lawyers and strike a fair deal where the developers pay something for the right to use protected works. Just like hair dressers, bars, and any public place where the TV or a radio station or music is playing, a licence is needed.
For example, would it be okay for a £billion company to use your likeness to model clothes and earn millions without your permission? You would expect to be contacted so you can review a proposal that would benefit you finacially before you are used to market products.
AI Developers do not have the right to do as they please, we have to move away from this mentality.
This is the law of the land I’m afraid, If I had created works of art I would want them protected as would you.
The actor/estate will still get paid if their likeness is used. if permission is not sought then there will be a court claim.
I think anything produced will just be for personal use or presenting to a small audience like kids for fun. It would allow families to cast themselves as the stars of a popular movie which I think lots of familes will have fun with.
Training data isn’t just appended to a dataset, it’s used for training. What the generative AI is doing is fundamentally no different than a human artist studying a piece of art and learning from it. The AI doesn’t directly copy anything, and in the handful of exceptions where a generative model has more or less accurately recreated an existing item it isn’t because those things are preserved intact, it’s that they recur so often that it’s learned them verbatim, for example the Mona Lisa example.
Which musical Generative AI program do you think would be more useful and generate more money, one trained with all copyrighted songs or one trained with no copyrighted songs?
I think this might be what it boils down to in the end, the former would be more lucrative so for that reason AI developers would need to make an agreement.
It’s very different. the generative AI will have the ability to understand the song on a level which humans cannot without the use of technology. Lets say there is a trumpet player who can play the trumpet in a unique way where nobody else can copy it, A generative AI would be able to immediately have a mathematical understanding of how the trumpet is played in this unique way and be able to reproduce it for others to use.
If this is not the case then AI developers should fine with paying a not so well known trumpet player to create examples for the generative AI right?
It does seem as if there will be hurdles to creating this generative AI, to me it definitely seems they like are moving too fast. If what I have typed here is true and factual then the copyright on whatever generative AI is to be trained on may actually turn out to be one of the biggest hurdles. I know of music copyright cases that have gone on for a good 20 years until they were resolved.
Money has been invested in hardware and buildings and if they do infact need to seek copyright permissions the goals of AI might take considerably longer.
I just want our models to be the very best they can be. They’re going to (already are in several cases) save us a lot of money, massively increase our productivity, and democratise creativity. I’m a decent piano player, but I can’t be a whole band (I’m no Tom Cardy). With music models, I can put together entire albums quickly, as long as I come up with good lyrics first.
A generative model doesn’t understand anything, least of all how to play the trumpet. During training you give it a brief text description, the lyrics, a pure audio file, and a deliberately corrupted audio file, and teach it how to filter away the corruption. As it gets better at this you gradually add more corruption until you’re training it on pure noise. Once it’s fully trained you take away the pure audio file and only give it random noise, lyrics, and a description. It will filter away noise until what remains matches what it’s been taught those lyrics and that description most likely should sound like, and the output will be more or less good music. Very similar methods are used to train image generators, and are also coming into vogue for LLMs.
The artist’s album and lyrics are used during training, but once the model has been trained, it doesn’t actually contain any music, just a lot of algorithms that remove noise from files of pure noise such that what remains matches the input descriptions. The model has no clue what a trumpet is, what it knows is that the word “trumpet” means it should be removing noise that doesn’t match a general idea of what a song with a trumpet in it sounds like.
Thanks for your explanation here, this is very interesting how an AI learns to identify a trumpet/Instrument.
I think the concern from the artists perspective is that the AI will still be storing information on the song which relate to rhythm, pitch, timbre, the AI will store a mathematical image of all attributes relating to the unique song.
I am reading about a company - Suno AI who have created an AI music program which has generated a lot of money. If you have a look on google you can see that literally every record label is rigorously pursuing legal action against them. I think their only way forward is to pay the music industry or face bankruptcy from trying to fight lots of unrelenting legal cases.
It does look like they have reached an agreement with Warner Music, artists associated with Warner can opt in or out to have their likeness, voices and compositions featured in AI generated music where they will be paid.
From what I can see AI is definitely new to the legal scene and it is clear that many are unsure about how to move forward legally with regard to AI. This is definitely a clear sign the industry is moving too fast because with any business venture everything is considered before any money is spent. The AI industry seems to have paid for and delivered massive data centers with racks of GPU’s without checking if it’s legal to use the very data the AI is to be trained upon.
Some people’s expectation of AI is unrealistic, real world experience that humans have will always be better than an AI. This guy didn’t understand that a district judge would always have the final say over such a case and once the facts are understood the goal would simply be to honour the contract.
It would also seem there is copyright issue with regard to programming languages, it would seem the libraries and framworks for many programming languages are copyrightable.
Github has brought a claim against AI developers where AI’s have been trained on code stored within Github without authorisation. A Judge has dismissed some of Github’s claims because they couldn’t prove the code generated by the AI was similar to the code database found on Github.
Hmm, I would have thought the AI developers would be found liable in this case, if it is Github’s policiy that code cannot be used for AI and then someone used it for AI then this is a clear breach of policy where they wouldn’t need to prove the code generated is similar as the focus should be the breach of policy.
I no longer tolerare AI to waste my time. Either working with it, or talking about it. I’m considering to make me some cheeseburgers this evening. Anyone else make them at home? I can’t believe people still go for McDonalds, given the little time it takes to build a good burger yourself.
Hmmm, factually 6 months ago Oracle’s share value was $345 each, today it’s around $140 each as it seems they have gone all in on AI which is also a factual.
There is obviously a huge problem if over half of their share value is gone.
I hope this thread gives people more insight into AI and the pursuit of it. I actually think AI is amazing and I use it extensively, recently it’s been helping me sell my phone chargers on Ebay where i have given google AI extensive queries regarding the voltage/amperage of usb-c chargers, their cables, and other extensive info on all charging requirements of laptops and phones and tablet. As I’ve said I want to see it advance but it must be a unified effort where we are carefully providing new jobs instead of taking them away.
This is really happening, I don’t think AI is the direct problem I think it’s the people in charge of the companies that are crucial to AI advancement. Lets keep an eye on this, I really like this thread.
In october 2025 Sam Altman signed a deal to buy 900k DRAM wafers per month with Samsung and SK Hynix where both weren’t aware of the deal with the other. The deal amounted to 40% of the global supply. It’s reported the deals would have looked different if the suppliers knew about the deals with the other.
This is annoying to be honest, it’s surprising that all three parties involved didn’t check the practicality of such a deal and how it would affect the industry. My speculation is that the memory makers actually manufactured the wafers to the spec needed for AI then Sam altman and co were unable to make payment on it as it does seem odd that all 3 big memory manufacturers lost value on their shares at the same time.
I’m unsure if the wafers can be re-manufactured for PC use or if it can’t be changed and I’m not sure if DRAM prices will remain high to cover the loss, hopefully we will get a more clearer picture in the comming days.
Google has worked out a way to use an old mathematical technique to scale down a dataset so it will use less memory and accorting to independent testers it does reduce the size of data with minimal data loss and no extra processing. so effectively they are able to compress the data before it is stored in memory.
This is actually a massive announcement by google which is less than a week old,
I’m thinking if this process could work in both directions where they are able to “reconstitute” the compressed data…I asked Google AI this and it gave me a great answer, generally lets say you have a 3d model of an Eve ship like a Raven, you would shine a light on it and record the Shadow of the raven and then delete the actual 3d Raven leaving the shadow alone, it is then impossible to rebuild the 3d model of the raven using the shadow because some of the data is lost. An AI using this google technique has this same problem.
Perhaps then the next step might be to find a way to additionally compress and store the the data that is normally lost so that it can work in both directions.
So depending on who uses this method (Johnson-Lindenstrauss transform) it could well result in a much smaller memory requirement for AI datacenters as well as smaller AI setups which will help on a global scale due to the AI effort needing a lot of memory, hopefully all parties involved in AI will try to research a similar method. I am impressed with Google’s engineers for looking at this with a different perspective and they seem to be leading the way.
This is what happened to Amazon when they introduced automated AI systems with no human oversight.
It would seem they saved money buy firing 30k engineers and then lost money when their systems crashed. An AI system used to design code will simply never have the experience of a collective of human engineers maintaining a high traffic global web service in my opinion, I believe the AI can gain experience on these matters but that endeavour would be cheaper if humans were paid to do it.
In general it appears the overall cost of creating large scale AI where the full cost of data centers and everything else needed to train/operate AI far outweighs the value of the revenue it can generate. Any business has to be able to pay their liabilities and once their revenue is lower than their liabilities they will eventually fold, I think a few big tech firms need to abandon or scale back their AI ambitions due to this issue as there is only so many billions of dollars they can borrow at a high interest rate.
