Hacker News Reader

Comment by robg | Hacker News Reader

What a fantastic post.

One thing I'd like to add: It's extremely difficult to overwrite or unlearn things that have already been learned. You can teach users something new, but if it conflicts with what they already know, they're going to run into trouble and much effort will be expended to overcome the multiple messages.

One silly example from my research: We found developmental traces of pre-school knowledge in biology professors. The professors acted like children even though they've had at least 20 years to overcome their early confusions. The simple, bastardized story: The brain doesn't have a delete button and overwrite appears to be very hard.

Better to conform to users, because they're not likely to conform to you. At most, pick one thing you want them to learn and the more novel, the more likely they'll "get it"..

View on HN

"We found developmental traces of pre-school knowledge in biology professors. The professors acted like children even though they've had at least 20 years to overcome their early confusions."

Details!

Sorry, it's still rare for me that folks outside my research area want to hear more. Two people makes a quorum!

The abbreviated background: Young children tend to confuse movement and naturalness with life. So if you ask a typical four-year-old if a cloud is alive, they'll tend to say yes. Same deal for things like a mountain and a truck. They don't know yet that life involves mechanistic things "inside" the organism. Similarly, they'll deny life to plants becuase they don't do anything. There's been a ton of research in this area because it addresses why young children are initially confused and how they can best be brought along toward mature knowledge. Even still, it takes us years to understand basic life processes. You might learn what's alive or not alive, but obviously the knowledge drills much deeper than that.

Results: Any case, the very first study I conducted as a grad student, I found that undergraduates, when tested under speeded conditions, have much more difficulty answering, for example, that a cloud, a mountain, and a truck are not living things than saying the same for a table. Same difficulty for calling plants living things as compared to animals. Just to show how hard it is to see the forest for the trees (pun intended), it took me five years to even think about testing bio profs. And wouldn't you know, they replicated the effects but with one important caveat. They showed improved performance for living things, as compared to the undergrads, but still much worse for plants than for animals and no improvement for the moving and natural non-living things. Even after 20 years of advanced biology training, they are still "thinking" like little kids, on some level.

Of course, I'm interested to hear complaints, concerns, or questions. But my take-away is that if something as simple as living or non-living can't be overwritten once much more advanced knowledge is acquired, and for many years at that, then learning that involves change is really, really hard. Better to rely on common schemes as much as possible, and expect users to learn as little as possible.

I wish I could cite a paper, but this startup stuff has had me happily distracted from the joys of publishing. But the manuscript is in decent shape and once my co-author looks at it, we'll send it out for review.

Interesting. But this does not necessarily say much about unlearning - as the distinction between living and non-living may be hard-wired into our brain. (Something along the lines of Steve Pinker's "the brain as a box of tools".)

It's an interesting point. A few quick reactions -

1. I don't know how to test that hypothesis (and indeed most neuroevolutionary "conjectures"). A mutation is one thing, but linked to very specific aspects of cognition is something else. For instance, how can we run a controlled study while leaving evolution free to vary? Some animal models may help, eventually, with certain aspects of knowledge (e.g., tools), but abstract thought is a whole other beast.

2. We've done some preliminary developmental work, and it seems with a specifically targeted instruction we can radically improve performance in young kids, and quickly too. So one answer might be to start early and never let kids (or users) go too far afield. The best way seems to be to show kids that only biological things move toward goals (e.g., you go to the sink when thirsty, roots grow towards moisture). But we don't know if that learning sticks or if they fall back into old patterns.

3. An alternative approach is to test a population that may not see animacy in the same way early in development and so may not exhibit the same types of traces later on. One idea is to follow up with autistic kids. To me, that's the difference between evolutionary accounts and a testable hypothesis about neurodevelopment. If the results show less potent traces, then we can use brain imaging to examine how early differences in the neuroanatomy lead to different processing streams.

4. If I were to agree with anything in that direction, an evolutionary foundation may reflect a static versus dynamic division (rather than living or non-living) of the visual system. Problem is, to understand biological mechanisms, static things (e.g., plants) have to be seen as dynamic, and dynamic things have to be seen as illusory (e.g., cloud). So irrespective of the true first cause, learning involves overcoming biases that initially "appear" to be powerfully important.

I read about a study where they found that people noticed moving animals faster than equally large moving cars. (I could look up the source if anyone is interested.) For your point 3: As far as I recall this effect did not seem to rely on whether those people grew up with animals (African subjects) or not (American subjects).

Perhaps if you would find some other task for your volunteers to test you learn/unlearn hypothesis with. Some task that we do not suspect of being hard wired into our brain. But a task people would nevertheless encounter at an early age.

Anyway - thanks for reading through my amateurish ideas.

Reply on news.ycombinator.com