AI won't learn the golf swing for you

There's something both funny and revealing about using AI to learn golf.

Not because the technology can hit the ball for you. It can't. It doesn't stand on the mat, doesn't feel the grip in its hands, doesn't hear the sound of a clean strike, doesn't feel the frustration when the old fault creeps back into the swing. But it can watch a video. It can read Trackman data. It can compare motion, ball flight, posture and intent, and sometimes explain a movement in a way that finally makes sense.

In this episode I talked with Guðmundur Smári Gunnarsson — biologist, firefighter-paramedic, fly-fisherman, golfer, archer, one of those people who seems to get good at almost anything he takes on. We started with golf and ended up in a much larger conversation about the body, rhythm, the nervous system, and AI.

Skill is built in the body

The difference between information and skill matters far beyond the golf course. We live in a time where information gets cheaper by the day — AI can explain, sort, summarise and plan in seconds. But skill isn't built in seconds. It's built in repetition, mistakes, corrections, and that strange patience of continuing to try. You can read about a golf swing, watch the best player in the world explain it in slow motion, ask AI for the perfect practice plan — and then you stand over the ball yourself, and you find out whether the knowledge has settled into the body.

AI as a training partner

I've been using AI as a kind of training partner in golf. I record videos, feed in data, describe my posture and old injuries. Sometimes I get a technical read. Sometimes a metaphor that helps me understand the movement. Sometimes an uncomfortable but correct reminder: you're tired, stop for a few days. What's most interesting isn't that the tech is "smart" — it's that it helps me talk to my own practice. It forces me to put into words what I'm trying to do. But the mirror doesn't hit the ball.

When the system is convincingly wrong

We also went beyond golf to one of my favourite tests of AI: "which endemic species are found only in Iceland?" It looks simple, but it isn't — it needs specialist knowledge, sources, caution. Over the years AI models have answered it with great confidence and sometimes pure fiction. It's a useful test because it reminds us of a fundamental: a system can sound convincing without being right. The same thing happened in producing this very episode — a raw transcript had Guðmundur down as an Icelandic champion in "boxing." It didn't fit. On checking, it turned out to be archery: public records confirm he became Icelandic indoor champion in men's recurve in 2016. That's exactly why human review matters — not because the machine is useless, but because it's almost right, in a very convincing way.

Some things should be hard

One thing I said in the episode was that I'm deliberately adding difficulty to my life because AI is making so much else easy. I said it half-joking, about blade clubs and golf, but the line has stayed with me. Maybe we sometimes need to choose tasks the technology doesn't make too easy. Not to reject the tech — these tools can help us enormously, especially when they level the playing field. But the human still has to do something themselves. Practice. Take responsibility. Know when the answer is too tidy. AI can help us learn faster. But it doesn't learn for us. And maybe that's exactly the point.

Show notes written up from the conversation with AI assistance and edited in my own voice. The raw transcript is available separately — the soul stone, for those who want to dig deeper. Direct quotes should be checked against the audio.