Another interesting test with pigeons was in a factory. There was a step where parts were on a conveyor belt and a human was supposed to press a button if a part came by that had certain defects. They were all defects that you could see just looking at the part as it went by so the inspector didn't have to interact the part. Think things like a part that was supposed to have been painted was not, or a part that is supposed to have a cap is missing that cap.
It turns out that humans aren't very good at this. They start off great at the start of their shift but by the end of the shift they are making a lot of mistakes.
Experimenters found that they could train pigeons do this. The birds were as good or better than start of shift humans and they could maintain accuracy over their shift.
There's a problem though. You still need to occasionally reward the pigeons when they correctly identify a defective part or the training eventually fades. It would seem you would still need to have a human around to do that.
Or maybe not. They found that they could deploy the pigeons in groups of three, with their buttons wired up to a circuit that would reject a part if at least 2 pigeons pressed their buttons.
In the case where 2 out of 3 pressed they would also give those 2 a reward. These rewards happened often enough to keep the training from fading.
Finally they considered the problem of training new pigeons. Could they do that with minimal human involvement?
It turns out they could. They made it so that they could add a 4th pigeon to a group of 3 already trained inspector pigeons. That 4th pigeon's button was rigged up so that it did not affect the inspection outcome, but if it was pressed when the majority of the 3 inspector pigeons also pressed the 4th pigeon would get a reward.
The experiment was a success. The groups of 3 inspector pigeons were much more accurate than humans, they stayed accurate all day, the were cheap, and it was straightforward to train new ones.
In the end though the factory decided not to actually switch to pigeon inspection. The thought that no matter how much research they had showing that pigeon inspection was better the public would not believe it.
Sorry I can't be more specific, but I remember this from a psychology class I took over 40 years ago.
off-topic but that reminds me of the “Twitch plays $someGame” phenomenon of about a decade or so ago. I remember some games being nearly impossible to play because the next control came through by chat and basically cancelled out the previous due to how messy that all was. No thanks!
This is kind of a fun exercise in hindsight but for a successful WW2 project that was pretty high tech for the time and successfully deployed - the proximity fuze is pretty hard to beat, the basic physics is simple enough for most people to grasp it but this problem stumped the Nazis and the Japanese only figured out a solution too late in the war to deploy.
Really with a lack of resources (and the relative safety & time available), off the wall thinking became the standard for British researchers throughout the war – to the point where it practically became policy. Not all of it worked, but considering the challenges faced it's amazing how novel some of the solutions were.
It turns out that humans aren't very good at this. They start off great at the start of their shift but by the end of the shift they are making a lot of mistakes.
Experimenters found that they could train pigeons do this. The birds were as good or better than start of shift humans and they could maintain accuracy over their shift.
There's a problem though. You still need to occasionally reward the pigeons when they correctly identify a defective part or the training eventually fades. It would seem you would still need to have a human around to do that.
Or maybe not. They found that they could deploy the pigeons in groups of three, with their buttons wired up to a circuit that would reject a part if at least 2 pigeons pressed their buttons.
In the case where 2 out of 3 pressed they would also give those 2 a reward. These rewards happened often enough to keep the training from fading.
Finally they considered the problem of training new pigeons. Could they do that with minimal human involvement?
It turns out they could. They made it so that they could add a 4th pigeon to a group of 3 already trained inspector pigeons. That 4th pigeon's button was rigged up so that it did not affect the inspection outcome, but if it was pressed when the majority of the 3 inspector pigeons also pressed the 4th pigeon would get a reward.
The experiment was a success. The groups of 3 inspector pigeons were much more accurate than humans, they stayed accurate all day, the were cheap, and it was straightforward to train new ones.
In the end though the factory decided not to actually switch to pigeon inspection. The thought that no matter how much research they had showing that pigeon inspection was better the public would not believe it.
Sorry I can't be more specific, but I remember this from a psychology class I took over 40 years ago.
https://www.rfc-editor.org/rfc/rfc1149
https://en.wikipedia.org/wiki/Department_of_Miscellaneous_We...
https://en.wikipedia.org/wiki/Bouncing_bomb
https://en.wikipedia.org/wiki/Explosive_rat
https://en.wikipedia.org/wiki/Panjandrum
https://en.wikipedia.org/wiki/Hobart%27s_Funnies