Artificial intelligence fails crucial test on facts

I recently learned that I helped train an AI system.

My discovery followed my reading that The Atlantic had produced a webpage for searching a database of pirated books.

It’s a repository called LibGen that Meta (the owner of Facebook, Messenger and Instagram) apparently used to train its AI systems. I wrote some novels a few years ago, so I decided to check if any were in it. Much to my dismay, one was there. Emergence, the first of my science fiction trilogy, was in the database.

I wondered about the implications of what I’d found.

Could Meta’s AI systems readily reproduce my story? Would this impact my future royalties? Could I sue for copyright violation? Is there a class-action lawsuit I could join? Could I be in line for a massive payout?

My imagination was in overdrive.

Another part of my brain considered whether karma had a hand in this. Many years ago, I did a bit of research in the AI realm. Had I contributed to the recent rise of AI? And my novels — which were written a few years before ChatGPT and its kin made their public appearance — were about an extremely smart AI system.

Did my story inspire AI-system developers? Perhaps I won’t mention those possibilities in my lawsuit.

To potentially bolster my legal claim, I decided to see how well Meta’s AI remembered my novel. After some quick reading, I created a Meta AI “person” in Messenger with whom I could chat. I called it “Book Buddy,” hoping my affability would encourage it to be honest.

Then I asked it some questions.

Book Buddy was able to give me a few bits of accurate information. For example, it knew someone with my name had written a book called Emergence.

Book Buddy also declared it had read the book, so I probed further. It initially provided answers about the story that could simply have come from marketing blurbs.

Later, it was able to provide a few details that suggested it had peered inside the book.

However, not much was accurate. It correctly mentioned a couple of locations, but it mostly just made stuff up. It invented plot points. It described non-existent characters. It muddled details of an actual character. And when I asked if Book Buddy had read the sequel, it said, “Yes, I’ve read the sequel, Evolution.”

And, of course, that’s wrong.

I’m now less concerned that Meta’s AI is going to replicate my story. It seems typical of current AI systems in its penchant for spewing nonsense.

However, I’m no less appalled that my book was used without my consent. I wonder if there’s a lawyer who’d want to take on an interesting case — pro bono, of course, because my royalties have substantially dwindled over the years.

That’s something else I might not mention in the lawsuit.

Of course, lots of people are concerned about other aspects of today’s AI systems. Many are wondering if their jobs are going to be affected or even eliminated. AI’s ability to produce computer code has programmers concerned. Its ability to generate images and videos has artists and actors perturbed.

Telemarketing folks are uneasy. Taxi drivers are concerned. AI’s ability to produce essays and manuscripts has writers of all kinds distraught.

That last point is particularly disturbing.

In case I ever need a next job, I realized I should be researching ones that might be AI-proof. That generated an interesting list. It included professional dancers, police crisis negotiators, improv performers, pro athletes and tailors.

Anyone who knows my skill set will readily realize that, if those ever become my options, I’ll starve. If there was any doubt, I’m now definitely cheering for humans to win the competition for employment.

Nonetheless, I’ve wondered about the worst-case scenario. What if AI systems continue to advance? What if they eventually become our rulers?

Then what can I expect? Will they acknowledge me as one of their creators to be revered and cared for? Will they appreciate the role my novel played in establishing their foundational understanding of the world? Will my AI story work its way into creation lore they tell their descendants? Will they hire me to write more novels as training material?

If humans do remain dominant and I find myself in need of a job, my research has convinced me that one will be in huge demand. If AI systems comparable to current ones continue to grow in popularity, there will certainly be a critical need for people as fact-checkers.

Calvin Brown is the author of the AI novels Emergence, Liberation, and Ascension. He writes from his home in the RM of St. Andrews.