What do 1960s LISP programs for natural language understanding, 1980s Prolog programs for expert systems, and today’s use of large language models have in common? Nothing, really, except they’ve all been referred to as Artificial Intelligence.

AI is not a technology. It’s a marketing term used by tech world people discussing many different technologies over the last 60 years. What’s different now is this scale of the use of this marketing term to sell us services. People like Sam Altman talk about generalized artificial intelligence because it gets them lots of media coverage that helps them to market the services that are their real goal: to have us all sign up to give them money every month so that their artificial intelligence can write our emails for us and read the emails that other people send to us so that we don’t have to. (And of course, to get lots of ad revenue as well there; look what it did for Google!)

We could reduce a lot of the silliness out there if these discussions — especially the many overly optimistic and overly pessimistic ones — mentioned the specific technology that they’re talking about instead of calling it “AI”.

University of Washington linguistics professor Emily Bender describes the term’s marketing role well:

[AI] is a marketing term. It’s a way to make certain kinds of automation sound sophisticated, powerful, or magical and as such it’s a way to dodge accountability by making the machines sound like autonomous thinking entities rather than tools that are created and used by people and companies. It’s also the name of a subfield of computer science concerned with making machines that “think like humans” but even there it was started as a marketing term in the 1950s to attract research funding to that field.

(Interestingly, when I try to find out more about AI as a marketing term, I mostly just find pages — especially advertising — about using AI to help you do marketing).

Bender goes on to say that “discussions of this technology become much clearer when we replace the term AI with the word ‘automation’”. She makes some nice points to support that, but I find it too simplistic to globally replace “AI” with a single term. We’d all communicate our ideas better by naming which specific technology that we’re marketing by calling it AI, and like I said above, it has been a set of different technologies since the term was coined. (I do highly recommend her Mystery AI Hype Theater 3000 podcast with sociologist Dr. Alex Hanna. The title is a tribute to Mystery Science Theater 3000, a late twentieth-century television show in which two hosts added their own hilarious commentary to bad old science fiction movies; Bender and Hanna analyze bad academic papers on AI technology and add their own sarcastic commentary.)

I see the current meaning of AI as being “the use of generative text chatbot interfaces to work with popular large language models”. It’s only about five years ago that AI usually meant the use of vector embedding models with neural networks (with so many layers that they were “deep,” so it wasn’t just learning; it was “deep learning”!) to identify patterns that could help people make predictions: was that chest x-ray anomalous? Was that series of financial transactions unusual enough to maybe indicate fraud? Instead of “AI” I would call that “machine learning”; the biggest, most successful versions of it led to the Large Language Models that dominate our AI discussions today as they predict which words might make the most sense after a given set of input words.

To complement a historical approach to the different things that AI has meant over the decades, at the recent Graphwise AI Summit I learned from Alan Morrison about the excellent Where Does ChatGPT Fit in the Field of AI? Venn diagram by consultant Jeff Winters:

Compared with my linear, historical approach, Winters’ diagram is more top-down. By describing the various technologies that have been associated with Artificial Intelligence and showing their relationships, this diagram can also help us all focus our discussions on what we can get out of which of these technologies rather than just referring to everything in the diagram as “AI”.

Many references to the term remind me of how non-technical people refer to “the” cloud as if there was just one. There are multiple clouds out there, and a cloud-based application is using one or more of them: AWS, or maybe Azure or Google Cloud or one of the others. There are many AI-related systems out there, and referring to all of them with one term poses the additional danger —worse than the vagueness of describing “the” cloud—of letting people think that there is one thing that combines the capabilities of all the different “AI” technologies. It’s dangerous because it feeds the panic about superhuman “generalized” AI.

In life in general, the use of a larger vocabulary helps us to express ourselves better. With all the panic and blather about AI these days, using the available broader vocabulary to discuss these technologies can help everyone to better appreciate the potential good and bad attributes and then plan for appropriate usage. Saying “LLM” is more specific; you can be even more specific by saying which LLM: ChatGPT, Claude, Gemini, or whatever. Maybe even better: say the name of the relevant tool. Don’t say “I’ll send you the AI summary of the meeting you missed” when you can say “I’ll send you the Copilot summary of the meeting you missed”.

Recently, as I chatted with someone behind a counter while we waited for my credit card to be approved, he told me that his son was studying machine learning at a German university. I thanked him for saying “machine learning” instead of “AI”. Being more specific about what we mean makes for clearer communication, especially if we’re using an alternative to a term whose meaning keeps changing and means different things to difference audiences.

Comments? Reply to my Mastodon or Bluesky posts announcing this blog entry.