AI does not understand anything

AI is a tool and scientists are doing some amazing things with it, but it doesn’t know anything on its own. It is important to me that you know this so I wrote a very long newsletter today

I get asked a lot whether my job (science writing/communication) is threatened by AI. So far, not really. The work I do usually relies on interviewing scientists and that requires understanding their work and asking relevant questions. AI doesn’t really understand anything, so I’ve got an advantage there.

But we are hearing much more about AI and many writers are now using it to transcribe interviews. That’s really helpful. It would take me hours to type out the audio from an interview, but AI software can do it in seconds. It’s never perfect, but it’s good enough for me to use as a reference.

I’ve added a statement to my website that describes how I use AI in my work, because I probably use it differently than other people. I definitely don’t use it to generate the text of my articles, but I don’t really use it for research either (and others probably do). I just don’t think I would get much out of its research. Sure, AI can summarize a paper, but as I described in the last newsletter I can find ideas for stories in throw-away sentences that probably would never have ended up in a summary. I want to read what the researchers wrote!

Derpy robot doesn’t know what you’re talking about. Photo by Rock'n Roll Monkey on Unsplash

Below the paywall today I have some more thoughts on AI and talk about how it’s actually being used for good in research. (It even won a Nobel already!) I’ll give you the important info up front: AI does not understand anything. It doesn’t know words. It doesn’t understand the information it gives you, even if it happens to be very helpful.

But first, some interesting links!

---

Interesting links

• Somewhat sticking with the AI theme, a new orchestral piece Lim Cosmic Rhapsody for Piano and Orchestra imagines “a collaborative journey between humans and AI in space, seeking solutions to Earth’s environmental crises” From BBC Music magazine

• Riley Black has a new book out and she was also on Science Friday to talk about how plants powered prehistoric giants millions of years ago

• Arts at CERN collaborates with Tabakalera and DIPC on the exhibition “Quantum Visions”.

• The Notre Dame sounds different after the restoration. Acoustics researcher Brian Katz has the receipts. By Richard Gray in BBC Future

• The ‘elegant’ math model that could help rescue coral reefs. By Max. G. Levy in Quanta Magazine

• Age and migration influence bird groups’ song repertoires, study finds. By Nicola Davis in The Guardian

---

What AI doesn’t know could kill you (so it always needs human help)

Earlier this year I read about a study that showed that people are more open to AI when they understand less of it, and that was fascinating. It explains why AI slop does so well online but also why I hate it. I know that it doesn’t understand what it writes so I just can’t take it seriously!

One of the clearest examples of how stupid AI really is is that it doesn’t understand clear danger unless it’s specifically built in by humans.