AI_behaviour

Rodney Brooks

Principles of behaviour based AI (overview)
The robots are situated in the world - they do not deal with abstract descriptions, but with the here and now of the world directly influencing the behaviour of the system The robots have bodies and experience the world directly - their actions are part of a dynamic with the world and have immediate feedback on their own sensations They are observed to be intelligent - but the source of intelligence is not limited to just the computational engine. It also comes from the situation in the world, the signal transformation within the sensors, and the physical coupling of the robot with the world. The intelligence of the system emerges from the system's interaction with the world and from sometimes indirect interactions between its components - it is sometimes hard to point to one event or place within the system and say that is why some external action was manifested - Brooks (1991), pp. 3, 14-16
 * Situatedness**: //"The world is its own best model"//
 * Embodiment**: //"The world grounds regress"//
 * Intelligence**: //"Intelligence is determined by the dynamics of interaction with the world"//
 * Emergence**: //"Intelligence is in the eye of the learner"//

glossary

 * regress** -- a procedure that entails its own reapplication without any limit
 * subsumption architecture** is a way of decomposing complicated intelligent behaviour into many "simple" behaviour [|modules], which are in turn orgnized into layers. Each layer implements a particular goal of the agent, and higher layers are increasingly more abstract. Each layer's goal **subsumes** that of the underlying layers, e.g. the decision to move forward by the eat-food layer takes in account the decision of the lowest obstacle-avoidance layer. (wikipedia, [|subsumption architecture])

A **situated** creature or robot is one that is embedded in the world, and which does not deal with abstract descriptions, but through its sensors with the here and now of the world, which directly influences the behaviour of the creature An **embodied** creature or robot is one that has a physical body and experiences the world, at least in part, directly through the influence of the world on that body. a more specialised type of embodiment occurs when the full extent of the creature is contained within that body (pp. 51-2, Brooks, 2002)

"...an airline reservation system is situated but not embodied. A robot that mindlessly goes through the same spray painting pattern minute after minute is embodied but not situated" (52)

"Brooks has argued strongly against symbolic processing approaches to creating intelligent machines, which had been the focus of AI since the days of [|Alan Turing], directly tracing back to the work of [|Gottlob Frege]. Instead, Brooks has focused on biologically-inspired robotic architectures (e.g., the [|Subsumption architecture]) that address basic perceptual and sensorimotor tasks." - [|wikipedia entry]



human information processing is not like computer processing
"Biological systems run on massively parallel, low speed computation, within an essentially fixed topology network with bounded depth. Almost all artificial intelligence research and indeed nearly all modern computation, runs on essentially von Neumann architectures, with a large inactive memory which can respond at very high speed over a very narrow channel, to a very high speed central processing unit which contains very little state. When connections to sensors and actuators are also considered the gap between biological systems and our artificial systems widens" - Brooks (1991), p. 3

(against the notion that the brain is an electrical machine, against electrocentrism) "... The brain is situated in a soup of hormones, that influences it in the strongest possible ways. It receives messages encoded hormonally and sends messages so encoded throughout the body ... hormones play a strong almost dominating role in determination of behaviour in both simple and higher animals

Real biological systems are not rational agents that take inputs, compute logically and produce outputs. They are a mess of many mechanisms working in various ways, out of which emerges the behaviour that we observe and rationalise ..."

"... we do know that signals are propagated along axonsand dendrites at very low speeds compared to electronic computers, and that there are signicant delays crossing synapses. The usual estimates for the computational speed of neuronal systems are no more than about 1 Kilo-Hertz. This implies that the computations that go on in humans to effect actions in the subsecond range must go through only a very limited number of processing steps - the network cannot be very deep in order to get meaningful results out on the time scales that routinely occur for much of human thought. On the other hand, the net-works seem incredibly richly connected, compared to the connection width of either our electronic systems, or our connectionist models. For simple creatures some motor neurons are connected to tens of percent of the other neurons in the animal. For mammals motor neurons are typically connected to 5,000 and some neurons in humans are connected to as many as 90,000 other neurons ([Churchland 86]).

For one very simple animal Caenorhabditis elegans, a nematode, we have a complete wiring diagram of its nervous system, including its development stages ([Wood 88]). In the hermaphrodite there are 302 neurons and 56 support cells out of the animal's total of 959 cells. In the male there are 381 neurons and 92 support cells out of a total of 1031 cells. Even though the anatomy and behavior of this creature are well studied, and the neuronal activity is well probed, the way in which the circuits control the animal's behavior is not understood very well at all." - Brooks (1991), p. 14

"A long time ago the brain was a hydrodynamic system. Then the brain became a steam engine. When I was a kid, the brain was a telephone switching network. Then it became a digital computer. And then the brain became a massively parallel digital computer. About two or three years ago I was giving a talk and someone got up in the audience and asked a question I'd been waiting for — he said, "but isn't the brain just like the World Wide Web?" The brain is always — has always been — modeled after our most complex technology. We weren't right when we thought it was a steam engine. I suspect we're still not right in thinking of it in purely computational terms, because my gut feeling is there's going to be another way of talking about things which will subsume computation, but which will also subsume a lot of other physical stuff that happens" - Brooks, //[|BIOCOMPUTATION]: A Conversation with// J. Craig Venter//, Ray Kurzweil,// Rodney Brooks

Brooks traces the evolution of both his own thoughts and the development of the AI community on this question
 * what is intelligence?**

GOFAI - play chess, calculus problems, algebra problems - computers have been successful in these domains

"The things that children of four or five years could do effortlessly, such as visually distinguishing between a coffee cup and a chair, or walking around on two legs, of finding their way from the bedroom to the living room were not thought of as activities requiring intelligence. Nor were any aesthetic judgments included in the repertoire of intelligence based skills.

By the eighties most people in AI had realised that these latter problemss were very difficult, and over the twenty years since then, many have come to realise that in fact they are much harder than the former set of problems. Seeing, walking, navigating and aesthetically judging do not usually take explicit thought, or chains of thought-out reasoning. They just happen

At first blush, my decision to leave out a cognition box seemed to indicate that I was giving up on chess, calculus and problem solving as a part of intelligence that I wanted to tackle. In fact this was not my intent. To me it seemed that these sorts of intelligence capabilities are all based on a substrate of the ability to see, walk, navigate and judge. My belief at the time, and still today, is that they arise at the intersection of perception and action, and that getting these right was the key to more general intelligence

... It seemed to me that insects could do much more than any existing mobile robot at the time. They could move at a speed of a metre or more per second while avoiding obstacles, evading predators and finding food for mates. They had only tens of thousands or perhaps a few hundred thousand neurons and each of those computed very slowly when compared to a digital computer.... What was the key about the way in which the nervous systems of insects were organised that let them perform so well with so little computation? ...

Slowly the idea dawned on me. Make the computations simpler and simpler, so that eventually what need to happen in each box would take only a few milliseconds ... The key was that by getting the robot to react to its sensors so quickly, it did not need to construct and maintain a detailed computational world model inside itself. It could simply refer to the actual world via its sensors when it needed to see what was happening ..." (pp. 36-40, Brooks 2002)

reference:
[|Intelligence without Reason] Rodney Brooks (1991) [|Flesh and Machines: How Robots Will Change Us] Rodney Brooks (2002) [|The brain is not a computer] Bill Kerr