Jordi Pons on what AI is really changing in music

Jordi is a friend and a doctor who works on artificial intelligence applied to music.

We've looked after his website for years. Which mostly means he writes to me when something breaks and I pretend it's not our fault.

I keep asking myself questions about AI. This time I wanted to ask them to someone who actually builds it.

Explain what you do to someone who knows nothing about AI or audio.

I teach machines to listen to music and, sometimes, to make it. Listening means things like separating the voice from the instruments in a recording, or recognising what's playing. Making means generating sound. Behind both there's the same idea: turning audio into something a computer can work with.

How did you end up researching AI for music?

I liked music and I liked engineering, and at some point I realised I didn't have to choose. Research was the place where both fitted.

What is AI changing in music that most people still don't see?

Most people look at songs generated from a sentence. That's the loud part.

The quiet part is in the tools: cleaning up a recording, separating tracks, finding one sound in a library of thousands. It doesn't make headlines, but it changes how people work every day.

A lot of people fear AI will replace musicians and creators. What do you think?

Some jobs will change, and it's honest to say so. But music isn't only sound. It's who made it and why.

A machine can produce something that sounds fine. Fine is not what makes you listen to something twenty times. I'm more worried about a flood of mediocre content than about musicians disappearing.

You've worked in research. How is building something for a paper different from building it for people to use?

A paper has to be new. A product has to work on Monday morning.

In research you can set aside the cases that break your method. In a product, those cases are your users. Both are hard, but they're hard in opposite directions.

What advice would you give a small company that wants to use AI without making a fool of itself?

Start with a boring problem you have every week, not with the tool. And check the results yourself before trusting them. AI is very good at sounding sure.

What would you like someone to take from your path?

That curiosity is a career plan. Not a very clear one, but it's held up for me.

I'll keep looking after his website. He'll keep explaining things I only half understand.

Seems like a fair trade.