[Previous] | Home | [Next]

Algorithmic Animal Behavior

If studied closely, animals can be seen (at least in some cases like these examples) to follow simplistic, algorithmic behavior patterns (like software where the designer didn't think about some potential cases).

source

The digger wasp, for instance, seems to display highly intelligent brood-tending behavior. Having dug a nest, it flies off in search of a caterpillar, overpowers and kills it, drags it into the nest, and lays eggs on it. The emerging young are thereby provided with the nourishment they need and find protection in the nest, which the wasp seals. Interrupt the sequence of partactions, however, and it soon becomes clear that no form of intelligence is at work here. Returning to its hole with the caterpillar, the wasp first deposits it in the entrance and inspects the interior, then reappears at the entrance, head foremost, and drags its quarry inside. If, while the wasp is inspecting its hole, the caterpillar is removed and deposited some distance away, the wasp will continue to search until it has rediscovered the caterpillar and then will drag it to the entrance again, whereupon the whole cycle-depositing, inspecting, etc. – begins all over again. Take away the caterpillar ten or twenty times, and the wasp will still deposit it at the entrance and embark on a tour of the hole, with which it is thoroughly familiar by this time. The insect continues to be guided by the same commands, in computer fashion, and evidently finds it hard to make any change in the overall sequence. Only after thirty or forty repetitions will the wasp finally drag the caterpillar into its nest without further inspection. Yet the digger wasp shows a great aptitude for learning where other procedures are concerned. While in flight, it memorizes the route which it must take on the ground when returning to the nest with its prey – a very considerable feat of learning. On the other hand, the burial of its prey is an instinctive action and, thus, strongly programmed. The wasp is almost incapable of influencing or altering this part of its behavior by learning, because it is controlled by an innate and extremely incorrigible mechanism.


and

Once stimulated, whole cycles of action can proceed by themselves. In the squirrel, food storing consists of the following part-actions: scraping away soil, depositing the nut, tamping it down with the muzzle, covering it over, and pressing down the soil. A squirrel reared indoors will still perform these actions in full, even in the absence of soil. It carries the nut into a corner, where it starts to dig, deposits the nut in the (nonexistent) hole, rams it home with its muzzle (even though it merely rolls away in the process), covers up the imaginary hole, and presses down the nonexistent soil. And the squirrel still does all these things even when scrupulous care has been taken to ensure that it has never set eyes on a nut before or been given an opportunity to dig or conceal objects.

Elliot Temple on May 21, 2003

Messages (16)

Another example of a simplistic animal algorithm from the same book.

https://web.archive.org/web/20040912010930/http://www.hans-hass.de/Englisch/Human_Animal/1_03_Innate_Recognition.html

> The toad reacts just as unselectively at mating time when faced with the task of finding a mate. The male leaps indiscriminately at any moving body and embraces it. Should the object of its attentions be another male toad, the latter emits a rapid series of cries, whereupon the former releases its hold. The mating-minded toad sooner or later encounters a female, whose spawn it fertilizes, but it has no innate "image" of a prospective mate. Waggle your finger in front of a male toad and it will mount and embrace it in exactly the same manner.


curi at 1:52 PM on November 26, 2019 | #14595 | reply | quote

https://web.archive.org/web/20040912010930/http://www.hans-hass.de/Englisch/Human_Animal/1_03_Innate_Recognition.html

> How little such reactions are associated with intelligence was shown by experiments with turkeys. To the turkey hen, the characteristic cheeping of turkey chicks is the key stimulus which arouses brood-tending behavior. Conceal a loudspeaker which emits this cheeping sound inside a stuffed polecat – one of the turkey's natural foes – and the turkey hen will take it protectively under her wing. Deprive the turkey hen of her hearing, on the other hand, and she will kill her own young because the appropriate key stimulus fails to reach her IRM.

There are other examples like birds looking for a particular visual pattern and you can make something out of cardboard that looks really quite different but triggers the algorithm. Or lizards that look for a blue stripe on other lizards to identify males vs. females and you can fool them with paint to add or hide the stripe.


curi at 1:58 PM on November 26, 2019 | #14596 | reply | quote

This is a typical example of people's evidence that animals are intelligent:

Cat Saves Toddler From Falling Down Stairs (1min video)

The second comment is notable because it relates to what algorithm is triggering and why it evolved:

> That’s the exact behavior that mother cats display when their kittens go too far. So cute how she tried to lift him by the scruff.

People seem unaware that fairly simple pattern recognition could be responsible.


curi at 5:24 PM on March 19, 2020 | #16046 | reply | quote

Hi Curi,

What are your thoughts on Koko the gorilla? I assume we could program a non-AGI algorithm that was similarly able to communicate through sign language but I wanted to get your thoughts.

Thanks,

Patrick


Patrick B at 10:35 AM on July 29, 2020 | #16929 | reply | quote

I haven't read any research papers on Koko and I don't trust the accuracy of popular stories. Apparently others have had a similar issue:

https://en.wikipedia.org/wiki/Koko_(gorilla)

> Criticism from some scientists centered on the fact that while publications often appeared in the popular press about Koko, scientific publications with substantial data were fewer in number.

One of the things that *might* be going on with Koko and dogs is that they're programmable by human trainers. Being programmable is quite different than learning. This might be due primarily due to lack of selection pressure to evolve good software security systems, but there could also be some advantages e.g. for cooperation.


curi at 11:04 AM on July 29, 2020 | #16930 | reply | quote

> This might be due primarily due to lack of selection pressure to evolve good software security systems

Has there *ever* been evolutionary pressure to evolve this?

ppl usually don't think that there's anything deep going on with birds that have been programmed to say some words. i don't see why it'd be any different for gorillas besides parochial misconceptions.

i think DD mentions like rudimentary memes in pre-human species, that might be one reason they're different, but i'm not convinced it's significant or applicable here.


Anonymous at 10:45 PM on August 4, 2020 | #16974 | reply | quote

>ppl usually don't think that there's anything deep going on with birds that have been programmed to say some words.

Repeating words doesn't imply any explanatory knowledge on what the word means though. Like a dog can recognize the pattern of: Hear the sound "sit" + "put my butt on the ground" = reward. Obviously we don't know what it's like to be a dog but I think you get my point.

Communication using words requires an understanding of the definition of those words as well as an understanding of the context in which they should be used.

So if we have evidence of a gorilla using words in rudimentary communication wouldn't that imply the gorilla has explanatory knowledge of the meaning of those words?


Patrick B at 9:02 AM on August 26, 2020 | #17619 | reply | quote

> So if we have evidence of a gorilla using words in rudimentary communication wouldn't that imply the gorilla has explanatory knowledge of the meaning of those words?

Depends what "rudimentary communication" is. You've presented no evidence, no models or explanations to consider that may account for the evidence, no critical analysis to rule some out, etc.


Anonymous at 9:16 AM on August 26, 2020 | #17620 | reply | quote

Communication is the transfer of ideas through the use of language.

We could teach a gorilla some hypothetical words A,B,C,D,E, etc. independently of each other (easier said than done). One day we might observe the gorilla use the word E, then D, then C and then A in a way that clearly conveys an idea to the observer.

Beings with explanatory knowledge create new ideas by rearranging, combining, altering and adding to existing ideas with the intention of improving upon them.

In this case the original ideas are the meanings of the words A,B,C etc. In order for the gorilla to convey the idea implied by the sequence E-D-C-A they must have had a conjecture that if they combined those words in that order it would convey a new idea. Conjecture by definition requires creative thought.

Of course the gorilla might just have gotten lucky. We would have to record them for say 1 year and keep track of each attempt they made at communication with sign language.

Then we could classify their attempts as either linguistically appropriate or not linguistically appropriate.

We could also classify all possible sequences (say under 7 words total) of the words A,B,C,D,E, etc. as either linguistically appropriate or not linguistically appropriate.

If the gorilla's attempts were linguistically appropriate much more often than what we obtained from the random combination of all possible sequences it would suggest that the gorilla achieved that success rate by using creative thought and explanatory knowledge.


Patrick B at 12:42 PM on August 26, 2020 | #17621 | reply | quote

> it would suggest that the gorilla achieved that success rate by using creative thought and explanatory knowledge.

No, results like that could come from e.g. a "reinforcement learning" algorithm.


Anonymous at 12:44 PM on August 26, 2020 | #17622 | reply | quote

#17622

Yep, good point. I guess they could keep trying different combinations until they achieved a desired outcome and then repeat that combination. That would still require them to have many wrong guesses which they refine into a right guess which is then reinforced. If we found that they never made wrong guesses and could intuitively make the right guess on their first attempt wouldn't that imply creative thought and explanatory knowledge?

I guess the best method would be for the observer to have limited or no interaction with the gorilla which would probably make the experiment quite difficult.


Patrick B at 1:03 PM on August 26, 2020 | #17623 | reply | quote

>>> it would suggest that the gorilla achieved that success rate by using creative thought and explanatory knowledge.

>> No, results like that could come from e.g. a "reinforcement learning" algorithm.

> Yep, good point.

I don't think you could have learned about reinforcement learning from my sentence (the ">>" quote). My sentence doesn't explain enough. And I don't think you would have written your first sentence (">>>" quote) if you already understood reinforcement learning.


Anonymous at 1:09 PM on August 26, 2020 | #17624 | reply | quote

Can you point me to a good resource?

Thanks for your replies


Patrick B at 1:24 PM on August 26, 2020 | #17625 | reply | quote

#17625 If you share what you already know and a tentative plan of your own, I could comment. I doubt you should read about reinforcement learning unless you're already a skilled programmer, which I'm guessing you're not. And also a skilled book reader who reads regularly, reads fast, and finds it easy. And you need to know stuff about how to evaluate ideas, debate ideas in your head or with others, learn, etc. Most people aren't nearly good enough at those things for it to make sense to start on issues like animal intelligence.


Anonymous at 1:31 PM on August 26, 2020 | #17626 | reply | quote

>If you share what you already know and a tentative plan of your own, I could comment.

I didn't really have a plan, just thought of a hypothetical experiment that would indicate gorillas could have creative thought.

>you're already a skilled programmer

I'm not very skilled but I did study programming in university (mechatronics engineering).

>And also a skilled book reader who reads regularly, reads fast, and finds it easy.

I work in cybersecurity so I'm constantly reading and absorbing new information.

>And you need to know stuff about how to evaluate ideas, debate ideas in your head or with others, learn

I am currently reading BoI (actually rereading the early chapters over a few times to make sure I didn't miss anything, there are some pretty revolutionary ideas in there)

>Most people aren't nearly good enough at those things for it to make sense to start on issues like animal intelligence.

Whether I'm good enough at those things or not doesn't matter to me. My goal is to beat curi in a debate on animal intelligence so if I'm not good at them now I will become good at them.

Also is there something I could call you other than "Anonymous"? Thanks again for your help


Patrick B at 1:47 PM on August 26, 2020 | #17627 | reply | quote

> My goal is to beat curi in a debate

> Also is there something I could call you other than "Anonymous"? Thanks again for your help

I also go by curi. lol


Anonymous at 4:17 PM on August 26, 2020 | #17633 | reply | quote

Want to discuss this? Join my forum.

(Due to multi-year, sustained harassment from David Deutsch and his fans, commenting here requires an account. Accounts are not publicly available. Discussion info.)