Legal and regulatory regimes often struggle to keep pace with emerging technology, but the challenge of AI is uniquely complex.
Intellectual Property (IP) is a particularly confusing component, for who can claim ownership over content created by a machine?
Tom Lingard, a partner at law firm Stevens & Bolton who specialises in intellectual property and technology law, says there's still no clear answer to the question.
"While there has been a lot of artificial intelligence being used and developed, it hasn't been producing outputs which have been publicly available or which are necessarily intellectual creations in their own right," he tells Techworld.
"It's only now that it's being used to create things like music and news reports and things like that that suddenly the question of protecting those outputs is coming into focus a bit more."
Existing legislation was passed before the potential implications of AI were understood. It includes provisions that could be relevant to content created by intelligent machines, but it remains unclear exactly how they would apply.
International differences on IP law
These legal regimes also vary across borders. The USA and Germany have previously made rulings that the only things that can be created and protected by copyright are those that have been created by humans. In the UK the law is rather different.
"In the Copyright Act in the UK there is a provision which dates from 1988 which says that if there is an invention or a literary, dramatic or musical work which is computer-generated then the author is taken to be the person who made the arrangements necessary for the work to be created," explains Lingard.
This provision was drafted at a time where there was always a direct line between the input to a program and the output it produced.
Now that AI systems are generating unpredictable outputs drawing on surprising sources of information, the creator of the content has become more ambiguous.
In the long run Lingard says the law might not offer protection for AI-produced content because it wouldn't be created entirely by a human. It's more likely that ownership rights would go to the person who was most responsible for the arrangements that led to the output – think a composer having copyright to a piece of music.
"The slight complication with that is you could end up with a number of different people who could fill that category," he adds. "You could have someone who created the AI itself and has created a program which does that, who then gives that to someone [who] can then set some parameters or put in different inputs."
Lingard points to Microsoft as a current example of the issue. The tech giant owns the IP rights to Microsoft Word, but doesn't profess to own the IP to all the documents written with the program.
In the case of AI, the law would have to decide whether the individual who chose the inputs has contributed enough to be named the author, and whether a contract should govern the terms of the relationship. The line will likely be drawn in different places depending on the individual circumstances of each case.
Who owns AI outputs with multiple authors?
There could be further issues if a third party uses someone's information in its AI platform. That individual may have an ownership claim to the output.
"If that's your information and that person is using that without permission, then arguably they're reproducing it by putting it into the AI in the first place is an infringement of your rights," says Lingard. "Then the question is who owns the output."
The output could have overlapping rights, covering both the creator of the new material and the individual whose original work provided its foundation. Determining ownership would be a tricky task. It would likely depend on whether the original information survived in the final output.
Lingard recommends that anyone using AI systems that produce valuable content should think from the start about whether they want to claim ownership of the output.
If they are producing something that business partners could use in further applications, they may want to agree at the outset who will own any new creations.
These discussions often only take place once valuable content has been created. It makes for a much harder conversation once there's tangible value and risk at stake.
Ultimately, there will be little clarity on IP rights for AI outputs until precedents are set in the courts.
"Fundamentally the building blocks are there, and potentially in the UK at least the legislation is there," says Lingard.
"It's just a question of the courts or the government making a decision as to whether that does in fact apply in all circumstances to artificial intelligence, or whether it's either an amendment or a separate clause, or if some sort of clarification is needed."