Whose Lie is it Anyway? – Guest Article on AI from Chris Lynch, IT Director at Source Insurance

In a recent judgment, a court in Germany has decided that Google should be responsible for mistakes in the information it generates for its AI search result summaries. It’s a long-overdue return to common sense.

But common sense hasn’t been terribly fashionable in tech for the past few years; we’ve much preferred mystical prognostications about AI, AGI, super-intelligence, hyper-scaling, and the odd apocalypse thrown in for good measure. Engineering made way for “vibes” and computer code that did as it was told was sidelined in favour of a machine that probably did it right, most of the time. 

Now the law is catching up and noticing that what comes out of the AI goose isn’t always a golden egg. If other countries follow the German example, the impact could be enormous not only for AI vendors themselves but also for any business using AI.

What’s the Current Situation with AI?

We’ve all become used to seeing disclaimers telling us that AI can make mistakes and that we should, as human beings, check all of its output ourselves. These disclaimers are in some way an extension of the special status that search engines and social networks have also long benefited from, and that protected them from liability for the things that users posted on their platforms or on the wider internet. 

(Legal nerds will no doubt enjoy the rollercoaster thrill ride that is Section 230 of the 1996 Communications Decency Act. Everyone else should take my word for it.)

What this rule boils down to is – the search engine (in this case, Google) doesn’t write all of the websites in the world; it just indexes them and tells you where you might find the answer to your question. Once you leave Google, you’re looking at someone else’s website, someone else’s content, and it’s up to you as the reader to determine if what you are reading is sensible, reliable, and actionable.

Bit by bit, the line between search engines, social networks, and traditional publishers like newspapers has blurred over the years. The more personalisation we see in search results and social feeds, the closer the activity of the tech companies gets to that of a publisher, actively selecting what content to feature and what not to feature. The important difference was always that the tech companies didn’t create any content – that was everyone else.

What Changed?

With the launch of AI tools, tech companies started generating content. 

There are many, many legal cases currently in progress where individuals and organisations are suing AI companies for using their copyright-protected materials without consent to train AI models. The most commonly cited line of defence from the AI companies is that the training of the models is “transformative” and therefore “fair use”. AI companies have worked very hard (but somewhat unsuccessfully) to prevent their models from outputting copies of copyright-protected content verbatim, claiming that the output from their models is original work.

Well, here’s the catch… If the output of the AI model is original work unconnected to the original input… just who is responsible for it? That “fair use” defence may be about to become a serious liability for the AI companies relying on it.

What Should You Be Doing Now?

A disclaimer alongside your AI-generated content may be sufficient today, but it won’t necessarily protect you in the future. If you’re using AI to create something, remember that you are just as responsible for the output as if you wrote it yourself. “Human in the loop” is no longer a value-add but an essential safety feature in any AI process.

If you already have a lot of AI-generated content that you haven’t fully edited and fact-checked yourself, Google’s troubles in German should be the “Ghost of Trouble Yet to Come” haunting your otherwise pleasant evenings. Ensure you’ve checked existing work and content for potential errors before they come back to haunt you as well. You should always treat AI output as a draft, not a finished product. We are not yet, nor may we ever be, at a stage where you can simply assume that the output from the AI system is correct. 

Risk management also has a role to play. Organisations should have robust AI usage policies in place to ensure that all staff are fully apprised of the risks, as well as the benefits, of AI. Before you unleash AI on a task, consider the level of risk involved and the consequences of the AI making a mistake. A good rule of thumb is the “work experience test”. If you would comfortably give the task to, or take the advice of, someone who’s just started their work experience with you and who may or may not be qualified to do what you are asking them to do… then that’s a job for AI. 

Keep up to date with changing regulations. If a butterfly flapping its wings in Japan can change the weather on the other side of the world, then the ripple effect of legal changes and cases won and lost around the world will almost certainly reach us. AI is a global phenomenon. 

(And if you think that’s a Fable… just ask the White House) 

For more guest pieces from Chris, visit the rest of our blog.