Some thoughts on AI

Most of what I read about AI these days is bad. Does it have anything valuable to offer, and can it be used responsibly?

It's difficult to read anything about generative artificial intelligence (which I'll refer to henceforth as AI) without being horrified. Here are a few examples.

• It steals intellectual property on an enormous scale.
• It frequently makes mistakes but then lies about them. It makes frequent errors in basic math and literacy.
• It consumes vast quantities of resources and is harmful to the environment.
• It fosters ersatz friendships with lonely people all over the world. There are even accounts of it encouraging people to commit suicide.

The trajectory of new technology

Technological innovation typically follows a path codified over 40 years ago as Amara's Law

We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run

The peak of excitement is followed by a trough of disillusionment, at which point many people decide that the new technology was little more than hype and feel deceived. But the trough is followed by a gradual rehabilitation as the new technology’s real benefits are appreciated, the technology itself matures and is eventually integrated into society, and its real value is appreciated.

Where are we now?

I believe we're somewhere around the peak of the cycle described by Amara. I believe we'll see a steep fall in AI's stock in the near future, followed by a slow rehabilitation.

The question that's worth asking is: what will that rehabilitation look like, and how can it inform how we use AI now?

Ways in which I've found AI useful and beneficial

I've been using AI for a couple of years now. I've come to appreciate its strengths and am learning to live with its limitations. Over time, I'm using it to solve a wider range of problems, but in a narrower and more disciplined way.

Here are a couple of ways in which I find AI useful in my daily work

• Writing first drafts of small programs or CMS plugins

  A Content Management System (CMS) such as WordPress consists of a core program interacting with a database to generate the content you see in your browser. It can be extended almost without limit by self-contained mini-programs known as plugins.
  Writing and maintaining custom plugins is a near-daily task for me.

  If a plugin is very small I can often generate the outline using a similar plugin as a starting point. If it's a little bigger it's often quicker to describe the plugin in detail to ChatGPT and have it write the first draft.
  More often than not, ChatGPT will produce this draft to an acceptable standard, and more quickly than I would do it unassisted.

• Solving problems in areas that I have only limited knowledge of

  I have a working knowledge of MySQL, the language WordPress uses to extract content from the database.
  Occasionally I'll need to generate a complex query involving multiple joins and subqueries, or involving complicated logic that I'm not sure MySQL can deal with. I find ChatGPT can handle these quite well.

  Sometimes I'll write a query that throws an error. MySQL will simply say "there's an error in your query," without offering much useful insight. ChatGPT will often identify the error and recommend a fix.

  Compared with the near-magical abilities often attributed to AI, these don't seem like a big deal but if and when we get to the far side of the trough of disillusionment, they'll make AI seem genuinely indispensible.

My AI Code of Conduct

Here is my unofficial AI Code of Conduct. Note that all of my rules begin with Never

1. Never use AI to produce creative output

   AI can do quite well when it's asked to combine input from multiple sources and solve a well-defined problem. Producing creative output on a blank canvas is a different matter.
   Asking AI to, for example, write a short story on a given topic is asking it to do a passable impersonation of a human being.
   It's a slap in the face to people who produce creative writing or any kind of art that involves technology, and it's insulting to your audience. Just don't do it.

2. Never let AI do any task unsupervised

   This is the big one from which most others flow. Agentic AI is a buzz phrase I've been seeing a lot lately. The idea is that you can turn AI loose on a problem and it will handle it, sometime combining multiple tools, with minimal or even no supervision.

   I can't think of a single task that I would trust AI to handle unsupervised.

3. Never use AI in an area you don't have at least some understanding of

   I've read stories of people writing programs and scripts in languages they have no knowledge of. If they're lucky, the program will work on the first try and they'll feel like they've accomplished something.

   The problem is that even the simplest program is likely to have bugs. A program built by what's called (another buzz phrase) vibe coding just might solve a simple, transitory problem but, even if it does, from the user's perspective, it's more of a parlour trick than a clever solution.

   Even if AI suffers a spectacular fall from grace in the near future, I'm confident that, in some form, it's here to stay. Per Amara's Law, it will become both ordinary and indispensible.
   I'm reminded that the word robot derives from the Czech robota, meaning "serf labour." Used responsibly, AI can become a labour-saving, drudgery-reducing device.

More information

• Roy Amara (Wikipedia)
• Rossum's Universal Robots (R.U.R.), the play that introduced the word robot (Wikipedia)