AI hypocrites can’t be allowed to destroy property rights

Promoters of artificial intelligence (AI) are fond of religious metaphors – Microsoft even calls one project Prometheus, after the Greek God who stole fire and gave it to humanity. But the Gods appear to be having a laugh at their expense – and one story suggests that they have a wicked sense of humour.

The giants of the tech industry are engaged in a frantic arms race to build the largest and most capable AI models but it may all be in vain.

Last week, Eliezer Yudkowsky, co-founder of an AI lab, made a startling discovery: AI models can copy each other far more easily than anyone previously thought, and the larceny is almost impossible to stop.

“If you allow any sufficiently wide-ranging access to your AI model,” he explained, “you're giving away your business crown jewels to competitors that can then nearly clone your model without all the hard work you did.”

The result? “You no longer have a competitive moat,” he says, referencing industry jargon for a competitive advantage that’s hard to beat.

There have been calls for Britain to establish its own sovereign AI to help us compete on the world stage. Yudkowsky’s findings suggest that, in practice, a homegrown “BritGPT” could get up and running fairly quickly.

But just as quickly, it would be ingested and cloned by a ChinaGPT, or for that matter, any offshore pirate GPT. Fancy that.

For decades, the tech industry has gradually eroded the power of industries that rely on copyright to underpin their markets. The Hargreaves Review of Intellectual Property in 2011, instigated at Google’s behest, urged copyright rules be tweaked to grow the tech sector.

Yet now tech companies find themselves in the same boat as irked rights holders. No value can be created from AI when you can’t stop copying and rivals can take what they please.

It’s a shame the Government is still in the swooning phase of its love affair with AI, and hasn’t figured this out yet. On the same day Yudkowsky announced his discovery, our Government announced it wanted to change copyright to benefit AI companies. Ministers rejected the idea last November after a consultation, but it’s returned on a wave of hype.

On the advice of Sir Patrick Vallance, Jeremy Hunt wants to force rights holders to the table for negotiations and has threatened to legislate if they don’t capitulate.

The tech giants view copyright not as a negotiation, or as a cost of doing business, but as an impediment. In the name of progress, they feel they should be able to take what they want.

Underpinning today’s AIs are large language models, or LLMs. These need feeding, and they have a voracious appetite. The models hungrily ingest everything they can find on the internet and, given enough material, can burp up an extraordinarily faithful replica of it.

Newspapers, image libraries, and commercial databases are all services that successfully operate in a market that depends on a strong and clear property right. The world has a thirst for English language content and our nimble, market-minded entrepreneurs are clever at making it work.

But in his review, Sir Patrick clearly sees copyright-based industries as subservient to feeding the AI models. He displayed all the tone deaf arrogance typical of the AI advocates when he declared that: “The content produced by the UK’s world-leading creative industries, including for generative AI, is fundamental to the success of the tech sector.”

This is a strange assertion to make. Tech giants may find it inconvenient to pay for a licence, just as I find it inconvenient to pay for sweets at my newsagents, rather than steal them. But it’s no more than a minor inconvenience and I obey.

Artists and illustrators are also livid at AI designers, and Getty Images is launching a lawsuit against one operation which couldn’t even be bothered to remove the Getty watermark.

Where’s the problem that needs fixing? In recent years, the creative sector here has grown at a faster rate than the UK economy - and yet we have nurtured a flourishing AI sector, too, one reckoned to be third in the world behind the United States and China. That growth came alongside robust copyright laws.

Worryingly, IP industries no longer have an independent counterweight to the tech lobby, after the recent Whitehall reorganisation saw smart Department of Culture, Media and Sport ministers subsumed into a new Department of Science, Innovation and Technology. Guess who they’re batting for.

While Britain rushes to water down intellectual property rights in search of the AI Holy Grail, a new mood of sobriety can be detected amongst the more thoughtful AI leaders.

Last week Yann LeCun, one of the fathers of machine learning, warned that the generative AI being hyped today is really a dead end on the road to intelligent software.

Truly intelligent AI that can “reason and plan, will have a very different architecture from the current crop of Auto-Regressive LLMs,” he wrote.

One of the most utopian of all AI advocates, Max Tegmark, a physics Professor and best-selling author, has also urged caution.

“An unregulated race to the bottom will end badly for the human race,” he said.

The irony is exquisite. Artificial Intelligence designers now need to erect barriers to stop copying, while demanding that their models are allowed to copy what they like from everyone else.

Now, having discovered that intellectual property is actually quite a useful tool to grow a business, perhaps they might then think twice before destroying other people’s.