Can blockchain help solve AI copyright theft?

It’s always nice when a solution looking for a problem actually finds a problem to solve. Blockchain’s reputation may be redeemed if it can fix the looming copyright issues for AI.

For the past few years, many CEOs were convinced that blockchain would be a great tool and they promptly went hunting for ways to implement it. Pretty soon those same CEOs realised that the blockchain didn’t solve many problems at all and they quietly moved on to the next thing.

What was the next thing? Artificial intelligence.

At this point, AI has proven so successful at fixing certain problems that entirely new ones are now popping up. Since training for AI models requires them to ingest oceans of data, there’s a risk these models will soak up copyrighted material for that training before spitting them out the other end.

Understandably, artists are furious about this, and dozens of lawsuits have been filed to compel AI companies to stop their thieving.

The core issue here is that artists lack any solid way to 1) know if/when AI companies are using their material and 2) block the illicit use of their works by anyone without a license. But scream as loudly as they want, these aggrieved artists simply don’t have any practical ways to enforce their copyright.

And that’s why blockchain could be a lifesaver. To see why, let’s do a quick refresher on what blockchain is.

At its simplest, blockchain is just a complicated ledger. The entire point of the software was to fix a flaw in accounting called “double-spend” whereby the same dollar is accidentally spent in two places. If this issue balloons out of control, there won’t be enough cash to pay receivables. Blockchain fixes this by using an immutable ledger to check if a dollar has already been allocated.

It is ingenious software that many companies have rightfully embraced, but its specific application meant it was never going to take the world by storm. However, with a little imagination, blockchain could become a useful solution for protecting copyright while increasing the transparency of how AI companies are using creative works.

It might play out a bit like this.

A centralised body, like the United States Patent and Trademark Office, could set up an encrypted ledger that stores records of every creative or patented work in existence. Each of these items would be allotted a unique “key” to prove not only who owns it but which entity may be using it under license, and for how long that license lasts.

This unique key would have software triggers that flag when an unauthorised person or company may be using any part of a protected work. The USPTO could also run its own AI program specially trained to recognise copyrighted material anywhere in the world. This program’s sole job would be to scan the internet like a soft-talking sentinel that carries a big legal stick.

If an individual AI company wants to use copyrighted material, it can apply for a license directly to the USPTO blockchain and it would be automatically processed – no humans needed. An appropriate licensing fee would be charged that compensates the rights holder with royalties while including a nominal maintenance fee to keep the USPTO blockchain alive.

Since the USPTO blockchain would show who currently has a license, any protected material that appears in the wild without the unique key would be considered a breach of copyright and escalated perhaps to the ISP level. The USPTO could then issue fines to the infringing party and force them to delete the unlicensed material.

To reinforce this USPTO blockchain, regulators can compel AI companies to open their training databases for inspection by the regime at any time. Companies that refuse to open their databases to inspection will instantly be judged as in breach of copyright and subject to maximum penalties – potentially even sanctions. That’s the big legal stick.

Of course, we understand that full transparency of private databases would be unreasonable. So, if transparency is out of the question, what about translucency? Rather than a glass window, a frosted window might be better – the public could see broad shapes of what’s in a database, but not the details. That seems like a good compromise.

To achieve this translucency, the USPTO should be the only entity with full access to AI databases. All databases will be registered with the USPTO the moment they are created. The USPTO would also be responsible for verifying if copyrighted material has been deleted from a database if a license expires.

The cool thing about a USPTO blockchain is that while anyone can view who holds the licenses, the exact identities of the licensees would be masked. All the public would see on the ledger is that “Company #12345 has the license for 1 year,” rather than “Microsoft has the license for 1 year,” for example.

There’s no reason why every country couldn’t have its own copyright blockchain.

Will blockchain solve the problem of AI companies illicitly using copyrighted material? Not entirely. Cyber-pirates will always find a workaround.

However, the point isn’t to stop copyright infringement. The point is to add more friction to the process of infringement which will lower the chances of that crime occurring. It’s human nature that when something is hard to do, fewer people will do it. That number won’t be zero, but it will drop closer to zero and that is surely a win for the good guys.

A bit of friction coupled with an easy-to-use licensing system run by a trusted central source will do a lot to convince AI companies that perhaps the most efficient path is to respect copyright rules – and stop making it harder for creatives to earn a living.