With Great Functionality Comes Great Legal Responsibility

It's unclear to me how many of you were actually old enough to use the internet back in the late 90s, but if you were, you might remember this charming little search engine called Ask Jeeves. It had a friendly cartoon butler, Jeeves, ask you what you wanted, and delivered you links, with a witty quip or two. It wasn't extremely powerful, but boy did it have personality!

Good old Jeeves.

Good old Jeeves has been retired for the last couple decades, but Google saw fit to try something in the same vein.

Google's version

It used to be that when you searched for something, Google would act like a normal search engine and give you a series of links to what you searched for. At most, it might suggest that you had mistyped your search, or aggressively promote content that people had paid them money to promote.

Relatively harmless run-of-the-mill stuff, as it were.

Then they started giving us an AI Overview, where an LLM would summarize the results of the search without you needing to visit those sites personally and draw your own conclusions. Which the site owners, understandably, weren't too thrilled at due to the decreased user traffic.

Google summary
on a platter.

At some point, the LLM-generated summary would even appear on top of all other searches (sort of like Jeeves serving you information on his silver platter) giving users very little incentive to do anything other than read that convenient summary. And that's where the trouble began.

In Munich, a regional German court took Google to task over two cases where the AI overview had gone wrong, with disastrous results. Apparently some publishing companies had been erroneously tied to scams and shady business practices due to a hallucination in the LLM. Not something completely unheard of, to be fair.

Google argued that they had explicitly stated that AI overviews could be wrong, they had included the original links, and users were able to click on those links for verification of the AI overview. The court tore that apart with the assertion that very few people ever did (kind of like reading a EULA and then clicking the "I have read the Terms and conditions" checkbox at the end) and that AI Overviews constituted Google's own content which it was legally liable for.

I'll concede that Google's AI Overviews are largely directionally correct, and the odd hallucinations are the exceptions that prove the rule. Unfortunately, due to Google's massive scale and reach, even a 0.1% chance of something going wrong represents a really bad day at the office for an unacceptable number of people.

A case I was reminded of...

Back in the day, Social Media platforms like Facebook and TikTok were being raked over the coals legally for misinformation of users. It's not the exact same case, but there are interesting similarities.

These platforms argued that they were mere content-displaying platforms and could not be held to the same standards as publications. However, the unfortunate fact was that they were implementing algorithms that largely dictated what users saw first. As long as they were doing this, the courts argued, they were effectively publications in their own right.

TikTok as a publication?

That does make sense. After all, don't publications also put their most compelling stories on the front page?

How are these cases similar, though? Well, in the original form of the services, all these companies did was, as claimed, display content provided by other parties. It was the value-added service by the companies themselves, which put them at legal risk. Also worth noting is, in both these cases, the offending mechanism were the default states. Meaning, they weren't configurations that the users could opt into. Therefore, the argument can't be made that "the users chose this", because they quite literally didn't.

Conclusion

User consent factors quite heavily into these lawsuits. It's often not enough to serve users something and then belatedly seek permission to do so. I could be wrong, but I suspect that none of this legal action would have happened if Google had gone with the classic display and only served the AI-enhanced version upon request.

Talk about courting trouble!
T___T