Pigeons taught to diagnose breast cancer on X-rays
newscientist.com
newscientist.com
I've an amateur ornithologist (it's a hobby), but never been a huge rock pigeon fan. However, I respect their abilities and how much we underestimate them.
[1] https://books.google.com/books?id=HxU-9UeDCI0C&pg=PA497&lpg=...
[2] http://www.amazon.com/Pigeons-Fascinating-Worlds-Revered-Rev...
Another interesting article: https://en.wikipedia.org/wiki/Project_Pigeon
I wonder if it's useful to think of pigeon learning like a machine learning algorithm -- after being shown examples of images that are classified according to some relatively simple but unknown boolean function, the pigeons can predict the value of that function on new examples.
Not sure about pigeons unless the images were simple enough, but crows, jays or parrots might be able to do that. My concern with pigeons is they might not have enough of the abstract/critical thinking abilities the other 3 bird species have. I could be underestimating them, but I'd probably put my bets on crows figuring it out before a pigeon. Though crows, jays and parrots are also smart enough to get bored after a while, so they might not be up to the task
New Caledonia Crows are smart enough to use "meta tools" where the tool they're using is only a step towards their end goal. They've been observed using smaller sticks to reach for a larger stick that will allow them to reach a piece of food. Crows (in general) also have awesome memories and will remember the faces of people who were mean or nice to them years later while also passing it on to their offspring (who have never seen this person before).
In a lab study, jays were shown to cache food if they knew they won't find food early in the morning the next day. If they knew they have a meal waiting for them the next day, they would cache less or not at all.
I know there's parrots that have been trained to open up locked boxes where certain randomly assigned locks are redundant. The parrots will ignore the redundant locks and only open the ones that are being used.
http://www.washington.edu/news/2012/09/10/crows-react-to-thr...
link to the paper on crow facial recognition: http://www.pnas.org/content/109/39/15912.full
As in, the pigeons are rewarded (or not) based on their choices within only a few moments, given only the pigeons' input to the system.
If pigeons were isolated and input was fed to them in parallel, then two or more inputs were taken to mean consensus and interpreted as evidential truth, would the distribution between individual pigeon responses mean that false positives would be evened out across the input distribution, or would such a technique result in progressive drift of accuracy?
I'm guessing some form of drift would be quite likely, and that for a real-world system a firehose of already-known outcomes (go vs. no-go) would need to be fed to the pigeons to ensure accuracy. Taking it further, I wonder if the drift could be modelled and accounted for...