Attempt at an automated column fails

I was sick this past weekend, and didn’t feel like writing my column. So I decided to turn it over to an artificial intelligence — or AI.

You’ve probably read stories about ChatGPT, a chatbot created by OpenAI with some funding from Microsoft. Recently, Google introduced a similar chatbot called Bard. A lot has been written about how these products might enable students avoid homework, the same way I wanted to avoid writing my column.

I’d read the stories about chatbots, but didn’t think they applied to me. But when my daughter and two granddaughters visited from Seattle, I watched them playing around with ChatGPT. So I asked it to write a short biography of Bruce McKinney.

It immediately wrote a very nice biography of another Bruce McKinney, which may have been accurate or not.
Twenty-five years ago, when I was the author of a semi-famous programming book, I always came up first in a Google search for Bruce McKinney, but now the first reference to me is entry No. 20, following a bunch of obituaries of Bruce McKinneys. So I narrowed it down: “Write a biography of Bruce McKinney of Silver City, N.M.”

ChatGPT took this as an invitation to make stuff up. It identified me as the author of this column, and of my ancient programming book, but it made up my birthdate and the name of my column. It claimed I had a computer science degree from the University of Minnesota, which I’ve never visited. (The author biography of my book clarifies that my computer science degree is from the University of Hard Knocks.)
When I put the same question to Bard, it gave a sort-of correct answer: “I do not have enough information about that person to help with your request.”

I imagined that an AI chatbot would be something a like a superhero who could read thousands of Google entries in seconds, and give a not-very-smart synthesis of all the relevant information. When I read more about the way chatbots actually work, it seems that they are significantly more complex, but ultimately more stupid.

The thing that surprised me the most was that a chatbot would make stuff up while failing to find facts in the public record. There is a long online record of me, going back to a 1960 newspaper article about my brother and I winning a Denver reading contest. I have songs on YouTube, service on boards, appearances before the city council, this column and much more. A superhero Google searcher should be able to find more without having to make up anything.

Still, there are things that chatbots can do. I read a chatbot interview with Jonathan Swift, who died in 1745. Swift is most famous for “Gulliver’s Travels,” and all his work is in the public domain and on the internet. A chatbot could find all the information it needed to answer interview questions.

The interview was a good summary of Swift’s history and philosophy, but there was something missing. I wasn’t doubled over laughing, as you would expect from the author of “A Modest Proposal.” (Google it if you want to laugh yourself sick.) I could certainly see how it might help a student cheat in writing an essay about Swift, but it’s not helping me write this column.

Still, I had some fun with Silver City history. I asked: “Was Harrison Schmidt born in space?”
ChatGPT answered: “No, Harrison Schmidt was not born in space. He was actually born on Earth in Santa Rita, New Mexico, on August 3, 1935.”

It’s Schmitt, not Schmidt, and the month of birth is wrong, but more important, it completely missed the point. In fact, what was Santa Rita now sits on air, above a big, deep hole, and people who were born there, including Schmitt, like to brag that they were born in space. Apparently, humor is not a strong point for chatbots.

I also got unreliable answers about John Bullard, Billy the Kid and Madam Milly. Both chatbots wrote horrible poems about Geronimo.

Finally, I asked about a local issue I am really interested in. I asked: “Does removal of juniper trees lead to grassland regeneration?” This is a topic of general interest around the West. One theory is that the oak, juniper and piñon forests on many of our hillsides weren’t here before White settlement. In some states, juniper removal is being done, and proponents (mainly ranchers) say that it does restore grasslands. The scientific papers are overwhelming. But ChatGPT didn’t help. It quoted both sides, reached no conclusions and had no insights.

I may write a column about this topic someday, but I’ll have to do the work myself. I can’t count on help from a lying, stupid chatbot.