Law prof predicts generative AI will die at the hands of watchdogs
- Reference: 1713957308
- News link: https://www.theregister.co.uk/2024/04/24/generative_ai_death/
- Source link:
For Amazon, Google, Meta, Microsoft, and other tech titans that have bet heavily on machine-generated content, this is a dire forecast, though perhaps not so bad as it is for smaller companies eyeing chatbots and automated content creation.
It's a flood of regulations and red tape the makers of generative AI models are trying to redirect through initiatives like the recently formed [1]industry consortium focused on AI safety. Participating firms aim to discourage or prevent generative AI from being used to create child sexual abuse images – because failure to do so guarantees legislative intervention and costs.
[2]
Goldman outlined the coming regulatory wave last week in a presentation you can see below at Marquette University School of Law in Milwaukie, Wisconsin, and in an accompanying [3]paper titled, "Generative AI is Doomed."
[4]
[5]
[6]Youtube Video
Generative AI refers to machine learning models trained on text, audio, and images that produce text, audio, or images in response to a descriptive prompt, like GPT-4, Gemini 1.5, and Claude 3, Midjourney, DALL-E, LLaMA 3, and so on. These models are trained on massive amounts of other people's content, often without consent or authorization.
[7]
There have been numerous copyright lawsuits over alleged infringement by the makers of generative AI models, many of which are still pending and could limit the viability of generative AI.
"I didn't fully address the indexing copyright litigation in my talk, but copyright law remains a major potential barrier to the success of Generative AI," Goldman told The Register .
"If copyright owners have a viable claim against Generative AI indexing, that would create an unmanageable rights thicket with millions of rights owners. Licensing schemes and statutorily-created rights clearinghouses could partially mitigate the issue, but only by dramatically increasing industry costs (exacerbating the [8]Sport of Kings problem ). Further, to avoid those costs, Generative AI model-makers might attempt countermoves that reduce their models' functionality."
[9]
We should note the Sport of Kings is a term that has been applied not just to polo but also to another notoriously costly pastime, [10]patent litigation , which is an equally apt reference in this context.
Speaking of LLM disasters
Microsoft last week introduced "WizardLM-2, our next generation state-of-the-art large language models, which have improved performance on complex chat, multilingual, reasoning and agent."
The open source model family was [11]touted for its performance, but evidently was released without adequate safety testing. So the Windows giant [12]withdrew the model , or tried anyway. The model had already been downloaded many times and thus can [13]still be found in the wild. Enjoy or not, while supplies last.
But infringement risk is not the primary focus of Goldman's concerns. He worries that the current animus against Big Tech and the accompanying regulatory environment has become too hostile for generative AI to thrive. In his paper, he hearkens back to the 1990s, when the internet reached a mainstream audience and the word "tsunami" was used in a more benign sense to evoke the social impact of emerging digital technology.
"It might be impossible to imagine today, but 1990s regulators often took a deferential and generally hands-off approach to the new technology," wrote Goldman. "This stance was fueled by prevailing concerns that overly aggressive regulatory responses could distort or harm the emergence of this important innovation."
What Goldman would like to see are laws like Section 230 of the Digital Millennium Copyright Act and the Internet Tax Freedom Act that have allowed the internet and businesses to grow and prosper, while providing a flexible, balanced structure.
That's not how today's lawmakers have addressed concerns about generative AI, as he sees it.
"The regulation will come as a tsunami," he wrote, citing figures from the [14]Business Software Association to the effect that state legislatures saw more than 400 AI-related bills introduced during the first 38 days of 2024, a sixfold increase.
"Not all of those bills will pass, but some already have and more are coming," he observed. "Regulators are 'flooding the zone' of AI regulation now, and each new bill threatens generative AI’s innovation arc."
[15]UK unions publish AI bill to protect workers from 'risks and harms' of tech
[16]What's up with AI lately? Let's start with soaring costs, public anger, regulations...
[17]Ex-Amazon exec claims she was asked to ignore copyright law in race to AI
[18]AI PCs are here but a killer application for biz users? Nope
There are various possible reasons why the optimism of the 1990s has faded, Goldman posits. First, there's the lack of public awareness of the internet when it first emerged. Relatively few depictions existed in science fiction at the time, particularly not dystopian ones. That's not the case for AI, which for decades has been featured in books and films as a malevolent force.
Then there's the general tenor of the times. In the 1990s, techno utopianism and cheerleading accompanied the rise of the internet and the spread of communications technology. Today, there's a lot more skepticism, what Goldman calls the "techlash."
Given images of drones dropping grenades on the battlefield, robocar collisions, warehouse robots stealing jobs, mobile device-based tracking, algorithmic labor oversight, and the extreme wealth of tech billionaires who insist on dominating public discussion, that's perhaps not surprising.
It is an incumbent’s effort to hinder its competitors. Many regulators will happily support these requests, even when they are being played
Third, Goldman cites the political polarization in the world today, and warns that partisan use of generative AI represents an existential threat to the technology.
Fourth, he points to the difference between incumbents then and now. In the 1990s, he opines, the telcos were the dominant players and the mood was strongly anti-regulation. Today, Big Tech is pouring money into generative AI, creating financial barriers to entry, and is trying to shape the regulatory landscape for a competitive advantage.
"OpenAI has openly called for the increased regulation of generative AI," wrote Goldman. "This move doesn’t prove that such regulations are wise or in the public interest. More likely, it is an incumbent’s effort to hinder its competitors. Many regulators will happily support these requests, even when they are being played."
He adds that these large tech firms look likely to embrace licensing fees as a way to mitigate legal risks, which in turn will raise costs and thus limit competition.
Goldman predicts regulators will make their presence known in all aspects of generative AI, with few limitations imposed by existing US laws like Section 230 or the First Amendment.
"The regulatory frenzy will have a shocking impact that most of us have rarely seen, especially when it comes to content production: a flood of regulation that will dramatically reshape the generative AI industry – if the industry survives at all," he concludes. ®
Get our [19]Tech Resources
[1] https://blogs.microsoft.com/on-the-issues/2024/04/23/microsoft-thorn-all-tech-is-human-child-safety-generative-ai/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2ZiksnjvRc22qF@a8RBiIUQAAAEw&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4802313
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZiksnjvRc22qF@a8RBiIUQAAAEw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZiksnjvRc22qF@a8RBiIUQAAAEw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.youtube.com/watch?v=o3tzZ_nH-AI
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44ZiksnjvRc22qF@a8RBiIUQAAAEw&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.cbsnews.com/news/the-sport-of-kings-polo-08-04-2012/
[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33ZiksnjvRc22qF@a8RBiIUQAAAEw&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[10] https://www.technologyreview.com/2004/04/28/232981/patent-litigation-the-sport-of-kings/
[11] https://web.archive.org/web/20240415163253/https://wizardlm.github.io/WizardLM2/
[12] https://www.linkedin.com/posts/igorbarshteyn_so-microsoft-released-their-wizardlm-2-llm-activity-7185833374035378176-s5o7/
[13] https://huggingface.co/MaziyarPanahi/WizardLM-2-7B-GGUF
[14] https://www.bsa.org/news-events/news/bsa-analysis-states-intensify-work-on-ai-legislation
[15] https://www.theregister.com/2024/04/19/uk_unions_publish_ai_bill/
[16] https://www.theregister.com/2024/04/15/stanford_report_ai/
[17] https://www.theregister.com/2024/04/22/ghaderi_v_amazon/
[18] https://www.theregister.com/2024/04/18/analysts_2024_ai_pcs/
[19] https://whitepapers.theregister.com/
I, too, haven't watched the video - but I skimmed [1]the paper .
The problem is that it starts with the presumption that Generative AI is for the most part benign and beneficial and that in order to flourish it needs the same light-touch regulation that greeted the commercialisation of the Internet. The alternative point of view is that having seen the consequences of light-touch regulation, lawmakers are keen not to make the same mistake twice.
He also suggests that well-funded incumbents in the AI space may be encouraging regulation as a means of excluding potential rivals, which is possible, but perhaps simply an argument for better rules. He also seems to be arguing that the speed at which the technology moves makes regulation (worse than) useless. But that's just a counsel of despair reframed as lawyerly disdain.
And he likens the ingestion of training materials to "indexing" by a search engine, which at the very least might be considered slightly disingenuous.
However, he acknowledges that his talk takes "a contrarian stance", which might be a roundabout way of describing it as "clickbait"...
[1] https://papers.ssrn.com/sol3/Delivery.cfm/SSRN_ID4802313_code332758.pdf?abstractid=4802313&mirid=1&type=2
Thanks. Those arguments are by no means uncommon. Stating to be taking "a contrarian stance" sounds to be similar to prefixing someone's social media rant with "unpopular opinion". It usually isn't.
I've followed a similar path, read the article (very nicely written), watched the video at 1.25x speed skipping familiar parts, listening to Q&A at 54:24 (questions about the EU GPDR, AI folks leaving to other countries, insurance companies), and skimming over the paper. Goldman's angle is that regulation can stifle AI innovation, which seems fair. IMHO, part of the push to regulate comes from the over-extra-ultra-hyper-hype that this tech has been given by its promoters, which should freak-out just about every reasonable human being on planet earth. At the end of his paper, Goldman suggests that " It would help to rebrand Generative AI to distance it from “AI.” ". Maybe ... and just for laughs I would suggest: Computer Algorithms for Creative Assistance (CACA) (eh-eh-eh!).
Another big issue though is the degree to which genAI is truly creative, rather than a "plagiarist", or a "liar". I did not find the images produced by DALL-E in Goldman's talk to be particularly creative (they seem cut from one same, relatively boring, cloth, to me). Also, plenty of content creators and artists have found genAI's more convinving outputs to be essentially copies of their own work, sometimes verbatim (hence the many copyright lawsuits). And, if Chief Justice Roberts believes AI can assist the poor in their legal defenses, and it is so great, even as OpenAI (and Microsoft) trust that they are developing superhuman-intelligence, that can pass entry tests to med school, law school, and what have you, then why don't they just put their money where their mouth is and use ChatGPT as their one and unique lawyer in the copyright court case against them brought by the New York Times?
...just my two-pence on this.
I doubt a lot any sort of law and regulation will be able to close Pandora's box again, but if it succeeded then it would immensly increase my respect for law makers, politiciens and laywers.
Eric Goldman has never legislated, or been involved in a legal case, involving the use of lamps and escaped genies it seems.
Man who deals entirely inside the law cannot see outside it
Sure legal ruling will prohibit the use and building of generalAI but when has that stopped literally anything it was intended to.
Usually such a ruling is a high flag that what they are banning is worth a LOT and even more the the right people.
Headline should be more accurately what above OR: Lawyer Speeds Up Race For Illegal General AI.
"Generative AI is destined to drown in a tsunami of regulation"
Good.
I fail to see what benefit the making of entirely fabricated content can bring. If I want to read something that has a chance of being interesting, I want it to have been produced by a human, under human oversight.
Sure, throw in an image made by Leonardo. That's not a problem because it will be the human that has decided that the image was relevant. If said human had asked another human to create the image, the result would have been the same. Leonardo is just destroying the livelyhood of actual human content creators. If everyone is OK with that, then let's do that.
But I refuse to read a book that has been generated by pseudo-AI. I refuse to read entire articles generated by a machine. I do not see how they can possibly be engaging and interesting.
GenAI can drown. It's no loss to me.
Outlawed tech is used by outlaws
Pandora's box has been opened and we threw away the box in the process.
You can ban the general public from accessing these AI models, but it surely will not prevent any state or private home user from using it in a frenzy of whatever they want to do with it.
You cannot stop this any more. The ghost has left the bottle, we threw away the bottle and the ghost is now haunting us.
When you outlaw the tech only outlaws will be using the tech.
generative AI will die at the hands of watchdogs
Seems to me that it's much more likely to die because it's very expensive to do at scale and not really as good as it's made out to be.
If history tells us anything its that legal obstacles have rarely stopped or got in the way of the Hyperscalars.
The Prof is delusional, and the regulators will be fighting a rear guard action - just like Ad-tech, Privacy/GDPR etc etc etc
I admit I haven't watched the video, but from reading the article I get the impression that Goldman is describing the incoming wave of regulations as a decidedly bad thing?
There are other points of view, you know. Sometimes regulation is strangling useful things, but sometimes it's strangling things that need strangling, even if they appear to be interesting.