NEW PARTNERS, OLD HABITS

The hypegeist missed it but a few mice and a dish of bacteria have almost perfectly described our constant struggle to use AI to the best of our human abilities.

Want to know how? Let’s start with the mice. They had found a behaviour strategy that worked:

Exposed to a short burst of an aroma. Then a long one. If they licked after the long burst, they got water. And the sweet satisfaction of success.

But then the task expanded. Sometimes the long burst came first. Now the rule that had worked often made them lick too soon. They had to wait for the second burst to finish.

They struggled where other mice didn't. And eight days later, they were still struggling. This recent paper in Nature Neuroscience showed just how hard a once-reliable strategy can be to abandon. Especially, perhaps, one that still sort-of works.

And, in a recent preprint, researchers studied bacteria that supplied each other with nutrients. For E. coli, swimming was an advantage when its partner could swim too. With a partner that couldn’t, swimmers lost ground to relatives that couldn’t swim.

What do these two insights mean together?

Successful strategies can depend on your partner. And when your partner changes, a strategy learned through earlier success may be surprisingly difficult to change with it.

The relevance? Well, as we learn to work with AI, we develop routines: what we delegate, which answers we check, when we take over. Those routines may serve us well with a particular system.

But the systems keep changing. Their capabilities may improve while our habits remain.

An improving partner doesn’t necessarily make an improving pair.

The hypegeist has been goggling and wincing at growing AI capabilities.

But for those at the coalface, as the models improve, are we getting better at working with them? And how quickly can we recognise when a way of working that succeeded yesterday needs to change?