Two types of intelligence and the current state of AI


These pictures were painted by a robot called Pix18. He or she (it doesn’t feel right) is a decommissioned factory robot trained to paint by picking up a paintbrush and painting on canvas. That’s pretty remarkable and shows how advanced AI is getting.

However, despite what we see here creativity is still largely beyond non human intelligence. Singularity Hub just published a great interview with Hod Lipson, professor of engineering at Columbia University and the director of Columbia’s Creative Machines Labs, who is pushing the next frontier of AI.  He posits that there are two types of intelligence:

  1. Convergent intelligence – Taking in large amounts of information and making a decision – e.g. whether to invest in a stock or pull out at a roundabout
  2. Divergent intelligence – Starting with an idea, or with a need, and then diverging to create many new ideas from it- e.g. designing a new robot different to anything we’ve seen before

You’ve probably guessed the punchline already. Machines are good, and increasingly better than humans, at convergent intelligence but not so good (yet) at divergent intelligence. That’s why we have automated trading and self driving cars, and will soon see robots flipping our burgers and slowly taking over all routine tasks, but still need humans to solve messy problems where there’s fuzziness around goals – e.g founding a company.

More interestingly Lipson contends that to crack divergent intelligence our AIs must become self-aware and reflective. That’s the key to the creative process. And once machines are self-aware and reflective they are arguably conscious. Some of them will certainly seem conscious.

The moment when we have divergently intelligent machines is perhaps the moment when the AI explosion occurs, with all of it’s massive potential upside, downside and uncertainty. I’m hugely excited by this future. If you are I recommend reading Hod Lipson interview in full.