cognitive science
and more
Breeding the perfect visual-search display

Tens of thousands of psychology students have spent hundreds of thousands of hours in stuffy little lab cubicles doing visual-search experiments. They have searched for diamonds among squares, red lines among green crosses, smileys among frowneys, and so on. You would think that by now every conceivable visual-search experiment has been done. But no, there's still cool stuff left.

In a study that just appeared in Journal of Vision, Erik van der Burg and his colleagues used a genetic algorithm to breed the best visual-search display. That is, they used evolution through 'natural' selection to create a display in which a target object was super easy to find. The results are a little surprising, which makes this experiment extra cool.

Natural selection applied to visual-search displays. The fittest displays from generation 1 are crossbred to create the displays from generation 2. The target is the horizontal red line segment in the center.

Van der Burg and colleagues started with random displays that consisted of tilted red, green, and blue line segments. There was always one horizontal red line segment: the target. Participants had to find the target and indicate whether or not there was a small gap in it. The (r)evolutionary aspect of their experiment was that each display had a fitness, which simply corresponded to the speed with which participants found the target. Next, they used the three fittest displays to create a second generation of displays, which were random mixtures of their fit parents, with a little random mutation thrown in. The fittest displays from this second generation where then used to create a third generation. And so on, for 14 generations.

Below you see the pinnacle of evolution: a display in which it is really to find a red horizontal line segment. (Strictly speaking this is a 'dominance map' that summarizes the outcome, which was slightly different for each participant.)

The optimal visual-search display. The horizontal red line segment in the top left is the target. Source: Van der Burg et al. (2015).

I suspect that every visual-search expert would have predicted that the optimal display would be all blue or all green (except for the target), so that the red target would 'pop out' because of its unique color. I certainly would have.

But surprisingly, the red target is surrounded by other red line segments. When you think about it, this makes sense: There is a big blob of red, which immediately attracts your attention. Then, within this blob of red, there is only one horizontal line segment. In other words, there is a two-step pop out: First the area around the target pops out because of its unique color; Next, within this area, the target pops out because of its unique orientation.

I wonder though: What would have happened if evolution had gone on for longer? After all, fourteen generations is not much. Is the display shown above really fitter then a display consisting of one red target among all blue (or green) distractors? An experiment waiting to happen ...

  • Van der Burg, E., Cass, J., Theeuwes, J., & Alais, D. (2015). Evolving the stimulus to fit the brain: A genetic algorithm reveals the brain’s feature priorities in visual search. Journal of Vision, 15(2), 8. doi:10.1167/15.2.8