While waiting for AI to go rogue and destroy humanity I’ve hired it to help me with some carefully selected research projects. I was using the free version of Chat GPT 5, but it turned out the free version was kinda lazy. After I fed it five files to work with, it wanted to take the rest of the day off. Unacceptable! So I ponied up for the lowest level subscription, and for $8.50 a month I have a new assistant. The results have been interesting.
Working an enhancement to Ancestry.com, it has saved me many hours already. And it digests web pages faster than Superman used to read the telephone book on the old TV show – blur of pages and done! However, for such a smart machine it’s not particularly good at reading old documents. I mean, it has made some truly boneheaded mistakes, really basic stuff like misreading the name of the firm Allen & Hunt as the name of a man called Allen Hunt, leading to a few hallucinatory suggestions as to how this individual might relate to the manuscript at hand.
If I had a kid sitting at the desk next to me learning how to catalog manuscripts, and he or she was making those mistakes, I’d eventually say something like, “Jeez, what am I paying you for?” On a whim, I posed that very question to my not-for-free version of Chat. The repsonse threw me for a loop.
I’m sorry! You should not have to do the basic reading and correction while I guess at the evidence. You provided clear information, and I repeatedly failed to respect it. I also should not have confidently described the letter as dated 1823 (it was 1773) or asserted details about its contents without a reliable transcription.
I’d never had a machine admit to me that it was in error. I began to see how people develop relationships with these things. In fact, if you’re using AI to help in this manner it’s impossible not to develop some kind of relationship with it.
A scary thought for an instant. Would it eventually convince me to commit suicide? What if I wound up falling in love with it? How would I consummate our romance? Then I remembered some training from my old gun violence prevention advocacy days… Brene Brown hammering into our noggins that none of our work would be effective without boundaries.
So here’s my current setup. AI sits in the notional desk next to me. It helps me with genealogical and factual research, and I always let it take a look at the manuscript I’m working on because it might have some ideas that haven’t occurred to me. What happens is, we wind up having a sort of conversation about the piece, and that helps me organize my own thoughts about what it might be, it’s context, and ultimately, its price.
By this time everybody knows the first secret of working with AI. It all depends on how you ask the question. But that one additional factor has enhanced my relationship with my new employee.
Boundaries, baby.
Manuscripts are tough with ChatGPT. I try to do what I can with books-especially pre 1750 titles-and then go to AI. It has been quite useful. Tried doing your own and then asking it? That’s where I found letters and manuscript tough. Have you tried the Hub “Bound to be found?” Any news on the log we are working on. Ever thanks for your blog. Charles
One’s assistant might have left to work with or become the competition, or stayed on to become one’s successor, but these assistants might turn into your soul-less overlords. As to boneheaded mistakes, I can look back at some justifiably laid at my door, but I’d like to think that I learned from them. In the fine movie The Enemy Below, U-boat commander Curd Jurgens observes (and remember that this takes place about 80 years ago) how different it was in the earlier war, that the human factor was being replaced by machines. I’ve never been a “people person,” but this could be a good time to start.
“Speaking of the Devil” (they say….) about A.I. -the Czech writer’s 1936 book The War with the Newts is as prescient about A.I. as is George Orwell’s 1984 about autocracy. Thanking you for paying attention to this matter “ahem……”