We need to explain *how things work*, not just *what they do*

Photo credit: Ling App via Unsplash

There’s a clip of Bill Gates being interviewed by David Letterman in 1995 that pops up from time to time. It includes a segment where Gates is trying to explain the Internet:

GATES: So everybody can have their own home page. Companies are there, the latest information. It's wild what's going on. You can send electronic mail to people. It is the big new thing.

LETTERMAN: Yeah, but, you know, it's easy to criticize something you don't fully understand, which is my position here --

GATES: Go ahead.

LETTERMAN: -- but I can remember a couple of months ago there was like a big break-through announcement that on the Internet or on some computer deal they were going to broadcast a baseball game, you could listen to a baseball game on your computer, and I just thought to myself, does radio ring a bell? You know what I mean?

You can find the full interview here: https://www.youtube.com/watch?v=fs-YpQj88ew

When this clip is posted today it’s pretty clear that, while we can appreciate Letterman’s joke, we are supposed to pity those poor people who couldn’t imagine our glorious digital future, a future in which, while radio broadcasts still exist, our radios are computers, and ‘broadcast’ means something like ‘the ephemeral transmission of content whose real existence is as a part of our giant, global, digital library’.

However, I think that the most useful thing to learn from the interview is that Gates does such a bad job of explaining the Internet. We shouldn’t judge Gates too harshly: late night interview shows were not really his natural environment. Furthermore, Gates does what many technologists do when attempting to explain technology: he describes what it does rather than how it works.

That is, rather than saying that computers talk to each other through things called networks, and that the Internet joins networks up so that computers can talk to each other over the whole world, he describes home pages and email. Practical examples are useful, but if they are all that we offer, then we don’t help people gain a mental model of the technology we are talking about. We don’t give them the insight they need to figure things out for themselves.

This phenomenon has been endemic throughout the history of technology, and is not just limited to conversations between technical and non-technical people. Indeed, if you are an enterprise technologist, you have probably spent a lot of time listening to sales people tell you about all the wonderful things that their company’s technology can do for you, while bursting internally with the question: but how does it work?!

We are making the same mistake today in the way we talk about AI. Most coverage about AI, whether from the press, from AI companies, or from organisations adopting AI, says a lot about what it does (Streamlines workflows! Automates customer experience! Outperforms benchmarks!) and little about how it works (Billions of parameters! Millions of chips! Acres of data centres!) Similarly, much AI training focuses on how to interact with AI products, not what is going on behind the scenes. It is not surprising that many people feel bewildered and confused by AI.

I fell prey to this phenomenon myself recently when I realised that it was time for me to do something about my own understanding of AI agents. I thought that I knew roughly what they were and how they worked, but not enough to answer some difficult questions. Where do they fit in a company’s technology architecture? How do you manage them? What identity and permissions should an agent have, and how does this differ from a human? A couple of years ago, when agents first started to appear within commercial products, I used to ask the product companies how agents should be represented in a configuration management database (As a single item? As a combination of a model and some software? Does it need to reference the context, prompts and training data? Does it fit in a CMDB at all? If not, how is it managed?). I didn’t get any good answers.

When I set out to improve my understanding of agents, I expected to find a lot of hype and, hopefully, some useful detail. I was surprised however, by the depth of the divide which I found in the content, between general purpose articles, brochures and websites which simply repeated the same lines over and over again (‘Agents are autonomous and can automate your business processes . . . etc.’), and books, courses and academic papers which, while useful to someone who works in technology, would be daunting to non-technologists.

For what it’s worth, my rough mental model of an agent now comprises a piece of traditional software, which captures natural language inputs, probably containing goals the user wants to achieve, passes them to an LLM, which generates text looking something like a plan, or like machine instructions which the software can execute or call through APIs, and these two components keep passing the ball backwards and forwards to each other until some condition tells them that they are done. And there’s some persistent data in their too. Other mental models and, I am sure, better explanations, are available. But this one helps me think about all the moving parts and how to manage them.

If we want people to trust and understand the technology we are building, then I think that we have to do better. Many technologists emit silent screams when they see the confusing, mangled, garbled, alarming accounts of technology which currently dominate the headlines (I know that at least one person does). We will continue to be frustrated until we convert those silent screams into audible explanations. It appears that technologists have a lot of explaining to do, not least to each other.

As a coda, the interview between Gates and Letterman concludes on an interesting note:

GATES: . . . Eventually we may figure out how to make the computer think, but that turns out to be a very tough problem. In fact, there's been almost no progress made on it. So nobody knows when that will happen. Some people think it will never happen.

LETTERMAN: We don't want them to think, do we? Not really, I wouldn't think.

GATES: Well, it's a scary thought.

LETTERMAN: Yeah. It's too bad there's no money in this too, isn't it? It's a shame.