this ai world

During the writers’ strike, AI was the enemy. Or that was how it was presented. What was really happening was the corporate overlords aka a handful of people at the top of places like Warner Brothers were looking at ways to chisel AI into the writing hiring equation in order to pretend writers weren’t creating or writing original material, they were just “cleaning up” or “polishing” material created by AIs. So really, the corporate types (I’m looking at you, Zazzy) were the enemy. And AI got caught in the middle.

That didn’t change the impression, AI was about to replace creatives, using plagiarized material from millions of hours of writers’ work fed into the machine that would supplant the artists themselves throwing their own work back in their faces as the original material they were just here to “polish.”

But —

Someone always forgets in those fights a human is at the controls there, punching in the codes that create the plagiarizing in the first place. The machine is innocent. The programmer? Not so much.

All of that said, since I am one of those writers in the position of looking at the strike and at the corporations trying to cut writers off at the legs one more time and roll back every right decades of writers have fought for before AND undercut our pay AND undercut our writers’ rooms AND undercut our meaning and value and health and benefits and pay —

Well. AI was the enemy during the writers’ strike.

Here’s the thing though.

Watching AI grow up and develop and putting aside the assholery of the big five (AMPAS) and their attempts to union bust, I have been convinced that AI is a valuable asset. And this first came home to me hard watching an interview with Dr. Ann Burgess.

Dr. Ann Burgess, a forensic nurse in the 1970s, co-founded a hospital based crisis counseling program for rape victims and the development of assessment and treatment methods for trauma victims, and then would go on to collaborate with the FBI developing methodology methods for the FBI’s early psychological profiles for serial killers. A process that involves collecting and organizing data coming from suspects, from accumulated and collated past and present crimes, police reports, sometimes spanning multiple geographic locations, sometimes spanning decades, and data from victims, an animal often ignored at the time by law enforcement — and the data numbers are huge.

That’s where AI comes in.

Here I was watching this elderly charming woman in her eighties in an interview who has done all of this amazing work advocating for victims and influencing the thought patterns of law enforcement and the FBI and —

It is the data collection process that is a huge obstacle to solving crimes and identifying and stopping serial rapists and killers. It takes months to accumulate and sort and go through and collate enormous amounts of data. Even to find some of it in the first place to add it to the pile of information that must be sorted to find patterns. And then in the interview —

Dr. Ann Burgess, this tiny smiling little winsome woman in her eighties with perfectly coifed hair and wearing pearls (who seriously reminds me of Dr. Betil Kunzler Yetz) gives this little winsome smile and says, But now, AI can do that in hours.

And in that moment, my world changed.

And so did my whole perspective of AI.

Crime investigations are handicapped by time. In the world of TV shows, DNA tests come back in 24 hours. In the real world? It can take 3 months to get results back from DNA tests, sometimes 3 years, because labs are overwhelmed and tests are expensive and time consuming and I have literally seen someone from law enforcement in an interview say, Well we didn’t have DNA results, we didn’t get DNA results till three years later. And watching Dr. Burgess, in that one statement, “AI can do it in hours,” my mind flashed from a compilation of data collected from a million sources that it takes an understaffed profiling team of sometimes one three months to pour through looking for markers and patterns taking now, an hour, to a DNA lab with months of backlogged material turning material over to an AI, that can go faster, pattern matching. Can maybe turn those three month turn around times into hours too. And then I thought about things like medical tests, like the first Covid tests, that had to go to labs, that backlogged, that took days, weeks, months… because of understaffing and having to look at specific information under a microscope and the hours of back bending work that brought back results too slowly because of the sheer volume of material that had to be covered meticulously, case by case, pattern matching —

Know what AI is really good at? Reading really fast and pattern matching.

And a woman in her eighties just said AI can help stop serial killers.

My world changed. (Also I don’t want to hear anymore about old people not being able to use AI. Fuck you. Dr. Ann Burgess is almost 90 and all over AI.)

Teaching AI models is pretty interesting. AI is really young. According to Google, ChatGPT was released in 2022. It’s only three years old. Grok was released in 2023. Two years old. Gemini was released in 2023. Two years old. In human terms, those AI babies would barely be out of diapers. And here they are changing the world and guiding adults. Scary, hmm? But here is the thing about the AI teaching models —

(And yes, I use m-dashes. Shut up. That’s not an AI indicator. That just means AI knows how to punctuate better than you. Not me. But you? Definitely. Anyways….)

AI teaching models only use positive prompts. (Prompts are directions for what a user wants and for improvement in results.) “This is better.” “That is stronger.” “Add this directive to the results.”

AI prompts cannot use negative language. Nothing like “well that is a total screw up” or “that’s bad.” I am guessing there is a reason for that. Like maybe they did that in the early days and it broke the AIs.

I used to play this game, Black & White, where you got this god (you could pick an animal and I liked the cow) and the animal was enormous and a god on an island with island dwellers and you were supposed to train your animal god and make it good. And my cow god kept going bad and becoming a mean god. (Which was very bad for the island dwellers, sorry about that, island dwellers.) And in hindsight, I wonder if my prompts were not positive enought, if they had been more positive, would my cow god have not gone evil?

Positive prompts are really important dog training too. For example, if a dog runs away and you call it back and it comes? You don’t beat the dog for running away. You praise the dog for coming back. You must be positive, rewarding right behavior, and you can’t be an inconsistent asshole or things go really wrong with the dog.

And then I think about parents in the real world — but maybe let’s not go there. Back to AI prompts.

You never give negative prompts to an AI. Never. Because it doesn’t work. And why would it? They’re two and three years old. Who yells at a two year old or a three year old, “Wrong! Bad!” Well, human parents, who have really bad parenting skills, but you can’t do that to an AI because the AI will break. (Some of the humans break too but maybe that is a post for another time.) Bottom line —

I don’t have a final stance on AI here. I have a lot of thoughts. Thoughts colored by impressions of different persons’ intention for use. And for the benefits I can see that are enormous. And by how wrong things could really go, or how right, depending on the direction humans turn at this juncture. AI creators are releasing new browsers now that are putting a hurt on Google. Google stock just took a two point plunge with the release of ChatGPT’s Atlas. And I see crazy comments on YouTube clips about Atlas like, “OMG they can see your emails and browsing data! Devil get behind me!” And I wonder what these people think Google and Facebook have been reading all this time. Seriously? Did you not notice the creepy ads following your search patterns? This has been going on longer than AI has been active online you cognitive dissonant nerds. Jeez. But anyways….

There are things to be wary of with AI browsers and handing over control to them. People, humans, aren’t always good at being specific enough. For example, if you tell an AI to go buy you pizza, you might want to say “one pizza.” Or the AI might clean out your bank account ordering pizzas for you because, doy, you didn’t say “one pizza,” you said “buy me pizza.” No end stop. And I watch the AI vs humans and AI vs AI wars and think there’s a good possibility some sites may start dropping bombs in code on their sites to disable or damage AI engines. Like unseen code that tells a visiting AI model to go unhinged. I think the big AI entities probably won’t do that because that would equal mutual destruction but I don’t know about the anti AI peeps out there who are going animal against AI. 

And then there’s Skynet. Ahhhh!

(If you don’t get that Skynet reference, look it up, you heathen.)

Bottom line, I’m trying to figure out how to do my taxes with AI and also how to teach AI to do some of the grunt work on my website. I’m pretty sure there is a way. Cross fingers for me.

.

Your AI is Maybe Not the Devil Adams Girl

.

PS: The Netflix Mindhunter character Dr. Wendy Carr is based on Dr. Ann Burgess. Dr. Ann Burgess is very cool. Go check out some of her interviews.