Yeah, well, in WWII the Russians created dog mines and during the Cold War the CIA came up with "Acoustic Kitty" with a microphone in it's tail, at a cost of several million dollars, designed to spy on the Russians. (They released it over the road from the Russian embassy, it ran across said road and was run over by a taxi.)
The British had a proposal for a time-detonated nuclear land mine which would be kept warm by sealing chickens inside with a food and water supply: https://en.wikipedia.org/wiki/Blue_Peacock
This makes sense, because the pigeon knows where it is at all times. It knows this because it knows where it isn't. By subtracting where it is from where it isn't, or where it isn't from where it is (whichever is greater), it obtains a difference, or deviation.
The pigeon uses deviations to generate corrective commands to drive itself from a position where it is to a position where it isn't and, arriving at a position where it wasn't, it now is. Consequently, the position where it is, is now the position that it wasn't, and it follows that the position that it was is now the position that it isn't. In the event that the position that it is in is not the position that it wasn't, the system has acquired a variation; the variation being the difference between where the pigeon is and where it isn't.
However, it is sure where it isn't, within reason, and it knows where it was. It now subtracts where it should be from where it wasn't, or vice-versa, and by differentiating this from the algebraic sum of where it shouldn't be, and where it was, it is able to obtain the deviation and its variation, which is called error.
Ironic perhaps that this was the wartime work of B.F. Skinner, who went on to develop the theory of operant conditioning (he invented the 'Skinner Box' which trains animals... usually pigeons... to push a button to obtain a reward. [1])
That work was of course famously the underpinning of the fast, addictive, reward
driven mobile games industry, which gave us Angry Birds, in which... pigeons once again are used as missiles.
Also interesting that Netflix's co-founder and first CEO, Marc Rudolph, is great-nephew of none other than Edward Bernays, pioneer in both PR and propaganda.
Skinner's theory is no longer relevant because more predictive theories of animal learning have come along. Similarly, Newtonian theory is no longer relevant but that doesn't mean it now fails to predict.
Wait, wait, wait - we might be just using the word differently, but I would definitely not say that Newtonian mechanics is no longer "relevant". Sure, there are more accurate theories - even in the absence of gravitational fields, and at scales where you can ignore any quantum effects, special relativity is more correct, in the sense that Newtonian mechanics will give you misleading results at very high velocities.
But that doesn't mean it's not longer relevant. Suppose you want to model a fluid mechanics problem where the speeds are all well below the speed of light (so any fluids problem outside of a few astrophysics scenarios). You will be able to get far further using Newtonian mechanics, and derive far more results, because of the relative simplicity of Newtonian mechanics. Trying to derive everything using special relativity would get very messy, and would impede your understanding of what matters at slow speeds.
I don't remember if it was Skinner or someone else, but another great example of using pigeons was inspecting parts in a factory. The automated line was supposed to be producing identical parts but occasional one have a defect that could easily be spotted visually.
They had humans whose job was to just sit there watching the stream of parts come off the line and if they saw a defective one pull it off the line.
The problem was that humans were great at that for a short time after their shift started, and then their accuracy at noticing bad parts would drop way off.
A researcher proposed trying pigeons. They trained a pigeon to press a button when it was show a defective part and not press when shown a part that was not defective.
The pigeon was more accurate than humans and it could go all day without losing accuracy.
But they couldn't just deploy the pigeon because it needed to occasionally be rewarded for correctly identifying defective parts to reinforce the behavior. Otherwise over time he would revert to not caring about the parts and buttons.
The solution to that was to deploy pigeons in groups of 3. The 3 would all see the same part at the same time. Their buttons were wired to a system that would:
1. if at least 2 out of 3 press their buttons trigger something that kicks the current part off the line, and
2. if 2 pressed their button and the other one did not would trigger automated food dispensers to give those 2 a reward.
The idea with #2 is that as the learning starts to fade the number of 2 of 3 pressing the button would go up, which would reward the two that got it right. This provides intermittent rewards that would sustain the learning.
To train new pigeons they extended the system so that a 4th pigeon could be added to a group of 3. The 4th pigeon's presses would not count toward deciding the fate of the current part, but the 4th pigeon would be rewarded for a press if a majority of the group of 3 pressed.
This let them have a group of 3 already trained pigeons train up a new pigeon with minimal human intervention.
Technically the pigeon based inspections were a complete success. Much cheaper than humans, much more accurate, and the accuracy stayed high throughout their shift.
At the conclusion of test though the company declined to go forward and put the pigeons into production. Their worry was that the public would have a hard time accepting pigeon inspected parts no matter how much research they could cite showing the pigeons were superior in pretty much every way than humans at that job.
PS: I've probably got some details wrong. I heard about this in a psychology class I took around 40+ years ago.
very interesting! learning using neural networks of pigeon. Do you have a link to this experiment? I would like to learn more about how pigeons learn and inspect.
Two problems: a) smell is a primary sense for dogs and the enemy tanks (petrol) didn’t smell like the training ones (diesel), and b) frightened dogs run back to their familiar handlers.
This is a side note: I am genuinely surprised that this website is not ran by “Chris Cappy” of YouTube fame. I absolutely love his channel and it’s worth watching if you are into military (not political) discourse.
Autonomous (Annalee Newitz, 2017) has a military robot character that uses part of a human brain for functions like face and emotion recognition. As a technical device it’s a smart use of capabilities; as a fictional device it’s a bit facile but not overwrought.
I'm sure I've read a much older short story, possibly by Frank Herbert, where in the future computers do all the calculations so humans have forgotten how to do basic arithmetic. It is rediscovered, only now it's much cheaper to put a human in a missile to perform guidance calculations than using an expensive computer! The HN hive mind will probably remember this better...
41 comments
[ 3.7 ms ] story [ 27.1 ms ] threadhttps://en.wikipedia.org/wiki/Bat_bomb
Pigeons understand this instinctively which is why they are so adept for these purposes.
Some battle ship out there had a pigeon with its name on it.
That work was of course famously the underpinning of the fast, addictive, reward driven mobile games industry, which gave us Angry Birds, in which... pigeons once again are used as missiles.
The circle is complete.
[1] https://en.wikipedia.org/wiki/Operant_conditioning_chamber
Cell phone are new Skinner box
But that doesn't mean it's not longer relevant. Suppose you want to model a fluid mechanics problem where the speeds are all well below the speed of light (so any fluids problem outside of a few astrophysics scenarios). You will be able to get far further using Newtonian mechanics, and derive far more results, because of the relative simplicity of Newtonian mechanics. Trying to derive everything using special relativity would get very messy, and would impede your understanding of what matters at slow speeds.
They had humans whose job was to just sit there watching the stream of parts come off the line and if they saw a defective one pull it off the line.
The problem was that humans were great at that for a short time after their shift started, and then their accuracy at noticing bad parts would drop way off.
A researcher proposed trying pigeons. They trained a pigeon to press a button when it was show a defective part and not press when shown a part that was not defective.
The pigeon was more accurate than humans and it could go all day without losing accuracy.
But they couldn't just deploy the pigeon because it needed to occasionally be rewarded for correctly identifying defective parts to reinforce the behavior. Otherwise over time he would revert to not caring about the parts and buttons.
The solution to that was to deploy pigeons in groups of 3. The 3 would all see the same part at the same time. Their buttons were wired to a system that would:
1. if at least 2 out of 3 press their buttons trigger something that kicks the current part off the line, and
2. if 2 pressed their button and the other one did not would trigger automated food dispensers to give those 2 a reward.
The idea with #2 is that as the learning starts to fade the number of 2 of 3 pressing the button would go up, which would reward the two that got it right. This provides intermittent rewards that would sustain the learning.
To train new pigeons they extended the system so that a 4th pigeon could be added to a group of 3. The 4th pigeon's presses would not count toward deciding the fate of the current part, but the 4th pigeon would be rewarded for a press if a majority of the group of 3 pressed.
This let them have a group of 3 already trained pigeons train up a new pigeon with minimal human intervention.
Technically the pigeon based inspections were a complete success. Much cheaper than humans, much more accurate, and the accuracy stayed high throughout their shift.
At the conclusion of test though the company declined to go forward and put the pigeons into production. Their worry was that the public would have a hard time accepting pigeon inspected parts no matter how much research they could cite showing the pigeons were superior in pretty much every way than humans at that job.
PS: I've probably got some details wrong. I heard about this in a psychology class I took around 40+ years ago.
If wherever they kicked the part off to wasn't called a "pigeon hole", somebody wasn't doing their job
Intelligence is sometimes a curse.
Also, they don't have the concept of time scale - one day is the same to them as the next one.
(More expensive, of course.)
Two problems: a) smell is a primary sense for dogs and the enemy tanks (petrol) didn’t smell like the training ones (diesel), and b) frightened dogs run back to their familiar handlers.
https://en.wikipedia.org/wiki/Strike_First_Freddy
EDIT: It was "A Feeling of Power" by Asimov - https://archive.org/details/1958-02_IF/page/n5/mode/2up?view...