In a recent conversation with a collegue we were talking about Artificial Intelligence. He brought up a theory that a computer can never be programmed to build a birds nest, his rational was that the sticks on the floor of the forest were never intended to be made into a nest, and the same goes for pieces of yarn, string, scraps of hay etc… The question is could a computer ever be programmed to build a birds nest?
The twigs were “never intended” to be built into a nest? I’m not quite sure what he means by that, but I don’t think it’s relevant–all that matters is that the twigs can be built into a nest.
As to whether a computer can be programmed to build a nest…I don’t see any reason why not, but we’re probably not close to having the technology to do that. This is just my opinion, though, so take it as you will.
Building a nest requires so little intelligence even a bird can do it. It is definitely not a matter of intelligence. If birds were more intellingent they would buy their nests ready made and there would be a market for nests. Manufacturing bird’s nests would be a piece of cake and you don’t even need a computer as it could be done by purely mechanical means. Machines do things much more complex than this.
Think of this: try to design a machine which will take blades of grass one by one and cut them. You have just invented a lawnmower. Or invent a machine which will take short fibers, one by one, and twist them into usable yarn. I could give you thousands of examples.
I have a basket here which is the closest I have seen to a bird’s nest. I think it was like a present, full of candy. It is not woven but a loose piling of flexible straw held together with some kind of glue and made over a mold.
I can guarantee you that if there were a market for bird’s nests, they would easily be manufactured by purely mechanical means. No intelligence needed.
Think of this: try to get a computer to sew. The answer is you do not need a computer to sew. You need a sewing machine which was invented long before computers. The stitch of a sewing machine is much more difficult to invent and carry out than building a nest.
Sailor: Everything you said is true, but I think the OP was merely asking whether it’s possible to do this, not whether there’s a demand for it.
You do make a good point–there are currently machines that can build things like a bird’s nest. Assuming these machines are computer controlled, then the answer is yes, computers can be programmed to build a bird’s nest.
I took the OP to be asking whether a computer could be programmed to find twigs in the forest and build a nest from that. Which question were you asking, Phlosphr?
OK actually After reading these responses I called my collegue. And he told me that the question? can a computer be programmed to build a birds nest is an old question posed by early inventors and programmers for AI, they can not fugure it out to this day!! Basically the people on the forefront of designing “intelligent” machines can not at this time (2001) develope a program that can build a birds nest out of the miscellaneous materials that birds usually build their nests from i.e. yarn, hay, grass, plastic, rubber, leaves, mud, etc…etc… simply because those items were never intended for that purpose. There for a computer program could not be “taught” to identify something that does not have a qualifiable/quantifiable parameter. Computers can be programmed to have logics for managing structured knowledge, but they cannot be programmed to work out side this realm. I want to know why.
the lawn mower and sewing machine answer is not what I am looking for.
Well, I took it they were talking about AI and how much would be needed to build a bird’s nest. My point is that a machine with zero intelligence can build nests much better than birds can.
Same goes with flying. Airplanes fly much better than birds but they do not fly like birds.
If you visit any factory which builds anything you will see plenty of mechanical machines which do things much more complex than build a nest and these machines have little or no intelligence.
Suppose I have to put those unfillable tops on liquor bottles. The tops are cylindrical and about as tall as they are wide. It takes a human intelligence to grab one, orient it right and put it on the bottle. And yet machines do it without looking at them or thinking about anything. You dump a box of these tops into a machine and it vibrates, pushes, drops, etc along certain passages so that all the tops come out correctly aligned at one end. Using any intelligence on this task is not a very good use of intelligence.
Same goes for building nests. You need to have a bird brain for that as you are better off if you build FHA subsidised housing.
Since the first question’s answer is obvious, let’s give the OP the benefit of the doubt and assume it was the more interesting question. Being able to move around in a natural setting, select a nest site, interpret your surroundings to find likely materials, and then put them together into a nest is a good challenging problem.
This does not markedly differ in degree of difficulty from taking a walk in a forest, surfing, snowshoing, skipping rocks, fashioning a spear, etc. All of them are learning to move around and manipulate some aspect of the natural world. Nest building is a good example, I personally don’t see it as any different from the ones I gave.
IMHO: They are all in principle solvable. Given the feats that various AI/AL systems have performed (in lab, semi-natural, and wild settings), I see no reason to think in principle they couldn’t be combined into one system that can do all that is needed. I don’t think we’re at that point now, nor do I think we will get there in less than 20 years, but I don’t see any insurmountable difficulties.
Well, here’s my two cents again. A large amount of the difficulty of designing a computer to build a bird’s nest has to do with the problem of getting the computer to process things visually. This is very difficult.
I don’t see why the fact that the twigs weren’t intended for this purpose is relevant, but I’m no expert, so I can’t debate that. Can you get your colleague to provide a cite for this argument? Anyone else care to chime in?
The machine that builds bird’s nests (or tops liquor bottles) does have zero intelligence, but the humans who designed and built it were presumably intelligent. I agree, this is not a good application of AI, but it’s a great research area. After all, doing any sort of manual labor is pretty complicated (mainly with regard to the sensory processing, but some tasks require some creativity as well), so if we can get a computer to do something like this, it would be a major leap forward.
Hi Philo,
I still don’t get the problem. I agree that the sewing machine/ lawn mower weren’t what you were after, but I don’t see why programming a computer to build a nest strikes you as impossible. Sure, if you tell it that yarn is used to knit with, and only used to knit with, and that leaves are for photosynthesis and only for photosynthesis, etc., you will have a problem. But why are you imposing such limitations on the computer?
You grant that a computer could be built with the architectural knowledge of nest-building. In the instructions, rather than hard-coding in the exact type of materials needed (which doesn’t strike me as very A.I.ish) give it a list of properties that each part must have. Then plunk it down on a movable platform with a few sensors and testing equipment and let it loose in the world. When it comes across a piece of yarn, it shouldn’t be too hard for the computer to determine that it is flexible, thin, and has a high tensile strength. Give it enough time to ‘learn’ and it will pick up several items on the forest floor, test them, and find many things that match the material requirements. Mud-wet, malleable, semi-liquid at T[sub]1[/sub] is a hard solid at T[sub]2[/sub]. Looks to me like it will have the same properties as cement, glue or whatever other material you choose for the initial ‘form’ of a nest.
Do we have the hardware to do all of this at the moment? Probably not. If a few tech schools opened up a competition would it be developed over the course of a semester? Probably. Unless I am missing something (which happens quite frequently) I don’t see a theoretical problem with programming a computer to build a nest.
Rhythmdvl
I think you’re all missing the point of the original question, which even the OP may not have been aware of. I suspect that the early AI theorists mentioned above knew of it.
The question about intelligence and bird’s nests was first raised, AFAIK, by Rene Descartes. His question wasn’t so much about the mechanical task of building a nest (which was conceivably a programmable task. Maybe like a Jacquard Loom), but about how the bird knew enough and was able to adapt nests to specific circumstances – to build a semicircular nest against a wall or a quarter-circular nest in a corner. The customizing of the nest to the situation seemed to imply a capacity for at least some degree of rational thought, at least to Descartes.
So the REAL question is: Can a computer be a slightly creative engineer? Not “Can a computer follow directions and build a nest?” (I think the answer is yes. In the limited cases noted by Descartes one doesn’t even need much creativity – just a “library” of solutions to choose from. But I think that computers are capable of much greater “creativity” even today.)
Sailor, note that all of your examples are used in huma-made environments. Your machines already have piles of straw, manmade tarmac, cylindrical supplies etc. available to them in predetermined ways. Planes cannot fly like birds can – they can’t land on twigs for instance. I think a major part of this question is to able to drop this hypothetical computer/machine in the middle of nowhere and have it perform the task successfully in the natural world.
Rhythmdvl has hit the heart of the problem, learning. AI has not been successful as was hoped with the learning process. To solve this OP, I think a machine has to be able to do at least some learning about the environment it’s in, and the materials it has available. It has to learn not to bother testing trees, air, etc., or eliminate those sorts of things quickly. From testing one bit of twig, it has to reason out that all twigs are good candidates, and it has to learn what a “twig” is. It has to learn how to put these twigs of non-uniform length/width/shape/density/etc’s together with any other materials it has successfully found in a pattern that will hold. And so on. Learning is incredibly non-trivial. It’s frustrated AI scientiests for 50 years of so, and will continue to do so no doubt.
However, given a restricted task like this, I think we will get there someday. (By restricted I mean to contrast it with trying to pass as a human in society or something very open-ended. In this case, at least some knowledge can be preprogrammed.)
If you are sincerely interested in the topic, you might try contacting Frederick Crabbe, who as of Aug 99 was a PhD candidate at UCLA researching “adaptive neural-net controlled agents that build nest-like structures out of material gathered in their environment.” I imagine he, or his colleagues, have made some significant strides in this area.
Most computers now-a-days are all out there in factories assembling motors, cars, plastic molds, etc…etc… But a motor is designed to be put together. The debris lying around on the floor of a forest isn’t designed to be made into nests.
Industry needs programs that work now, not programs that point the way to a cognitive science of the future. And the quickest way to make a program accomplish a sophisticated task is to write a sophisticated set of instructions for the computer to follow, step by step, one after another. An expert system takes a particular kind of expert behavior - medical diagnosis, for example - and imitates it, with the help of rules abstracted from the way people perform the same tasks. A diagnosis program might tell the computer to begin with certain questions, just as a doctor would. Then a particular response might guide the computer to a particular set of follow-up questions. The paths of questions and answers can go in an intricate variety of directions.
"This from Marvin Minsky Donner Prof. of Science MIT, so philosophically speaking, to date computers can not be programmed to create a birds nest.
still the answer is vague, and this only proves that technology will never surpass our humanity, or is that wishful thinking.
It ought to be fairly simple to get a computer to model a bird’s nest; I expect if you threw fractals into the mix, you’d even get very lifelike nests; then it’s just a matter of buiding a robot delicate enough to be able to reproduce the model in real life.
It’s not quite that simple, as the following quote shows:
Trying to translate a fractal model into a real-world object would only complicate the process.
Have to agree with ultrafilter, you can not really bring a fractal set into this picture. Maybe a Birds nest is as complex or as simple as a fractal set, [depends on which theory you like chaos or order] If one looks at the Mandlebrot set the famous fractal puzzle of infinitely large and small, it looks chaotic but it is quite orderly. As for the birds nest…we shall see with time. I do not think the answer to that one will come any time soon. and to be honest when it does I hope it does not mean our technology has surpassed our humanity…which is a whole other forum.