The phrase “AI Slop” is already commonplace, and it inspires me to invent more neologisms like “Cry me a data center” or “Grok willing and the creek don’t pop-up”. I have written all 670 blogs over the last two decades without AI, or have I? I am a late-adopter with Post Technology Stress Disorder and you can see that in my blogs from the aughty-aughts. Heck, you can see it from a couple of months ago when I ranted about how real retail service companies cannot ever, ever change an old email or recognize a different phone number, no matter who is dead. And they are willing to give up months of revenue just to carry that rotting corpse further.

So what do I think of AI? My first reaction to AI writing was that it was like a 14-year-old who has been asked to turn in a three-page essay but only has one page of material. Repetitive. You see it in tons of social media posts which look like proper long-form discussions but hammer the emotional hinge of the story like six times in 400 words. That is the written version of the AI image that has the wrong number of fingers or arms. Of course, both of those examples are a year old so I suppose by now AI is like 15 and knows how to draw hands better.

This is a real life photo from 12 years ago no AI.

I have attended one workshop and one training in AI in the last 24 hours so here is what I learned, mostly from Huy Pham and Paul Kim, whom I first saw present on this issue almost a year ago in Arkansas.

AI is a marketing term for the latest iterations of pattern recognition technology. Basically it looks at large amounts of data and identifies patterns and recurring themes. So, like you learned in computer science 40 years ago – GIGO. Garbage in, garbage out. Huy and Paul work for APIAHiP (Asian and Pacific Islander Americans in Historic Preservation) and if they ask AI about Asian sites in the National Register of Historic Places and if AI looks at all 100,000 nominations and sees less than 1% it will tell you basically NO. There aren’t any. Not significant. Not important.

Since AI is basically pattern recognition, everything it produces will tend toward the mean, the median, the average and ordinary. It is not a place to look for exceptionalism or new ideas. Culturally, it has a flattening effect. What is new about it in the last few years – as opposed to the pattern recognition software of 20 and 30 years ago – is the Large Language Model. Basically you give it tonnes more examples. So it gets even meaner. It cannot provide conclusions, only predictions. It reflects what is common, not what is true. Garbage in, garbage out.

I guess I’m not a big fan

If you want the National Register of Historic Places to identify underrepresented stories, you need to do it yourself, because it requires not lots and lots of old data, but a new approach. You need to reconceptualize the process, which is not something in AI’s wheelhouse. Garbage in, garbage out.

“Hallucinations,” however, are in AI’s wheelhouse. They will confidently answer a nonsense question like “Who won the Nobel Prize in Economics in 1956?” and even provide fake citations from nonexistent journals, as we have seen recently. It seems the AI doesn’t know how to say “I don’t know” which is a prerequisite of real intelligence. It eagerly answers queries because – like the 14 year old – it can’t not. Even if it is hallucinating.

If you get the “Nobel Prize in Economics,” you will get it here, in Stockholm, but it will actually say Sveriges Riksbank Prize in Economic Sciences in Memory of Alfred Nobel, and it only began in 1969. Four of the five original Nobel prizes are also delivered in Stockholm under the terms of Alfred Nobel’s 1895 will, with the Peace Prize awarded in Oslo.

Presents predictions as conclusions. Repeats what is common rather than what is true. Hallucinates.

No wonder the predator class and the oligarchy are so enamored of it.

More microdosing and unlocked tranches than Sand Hill Road.

So, does AI work with historic preservation? I have always said that the beauty of historic preservation is that it treats every resource as an individual, with individual characteristics and individual significance that cannot be commodified. Zoning treats properties as a category, like grain being graded and loaded into a silo. It becomes a commodity sans terroir. Pattern recognition technology might be suited for zoning or other commodifying tasks, but telling the story of a place is specific and nonfungible.

Terroir.

So, to answer the question: AI can help if you need to find out how many 1926-29 Tudor Revival houses in the Monte Vista district have casement windows or how many National Historic Landmarks are shaped like elephants.

But if you are nominating a site with a story that has not been heard before – AI doesn’t know. Because it hasn’t been told. And it doesn’t know what it doesn’t know.

Why doesn’t someone get AI to do the dishes or wash the floor? Something useful, you know?