DRAFT Abstract for DIGITAL BIOTA 2 Conference

Speaker: Aaron Sloman
School of Computer Science and Cognitive Science Research Centre
The University of Birmingham

Title: What sorts of brains can support what sorts of minds?

Slide presentation from the conference

The abstract below was prepared in advance of the conference. For the conference itself I prepared a set of slides, available in either postscript or PDF, here

Filename: Sloman.biota.slides.ps
Filename: Sloman.biota.slides.pdf


There seem to be two kinds of people. Some like to find interesting abstract levels at which things are the same, and they search for general principles at these levels. Others are more concerned to emphasise the interesting details to be found in special cases.

For the first type of person, it is obvious that:

o artificial life and biological life can be construed as essentially the same sort of thing,

o there is no essential difference between intelligent software systems, intelligent robots and intelligent animals

o simulated environments and physical environments are not importantly different.

For the second type of person, it looks like a serious intellectual sin to lump together such obviously diverse phenomena, ignoring all the important details in which they differ, and which require different principles and behavioural laws for their description and explanation.

Is any kind of synthesis of these two views possible?

Perhaps not where the clash is driven by emotional and motivational differences or theological concerns (e.g. a strong *desire* to separate animals from machines, or humans from everything else).

But if we disregard our concerns about what we *wish* to believe is the case, and simply try to find what *is* the case, then I think a synthesis is possible, based on two principles.

(a) The principle of multiple levels of reality.
(b) The principle of multiple realisations of intelligence.

These will now explained.

(a) The principle of multiple levels of reality.

This draws attention to such facts as that besides atoms, quarks, or whatever physicists (present or future) happen to regard as the building blocks of the universe, there are also such things as plants, animals, tidal waves, poverty, crime, economic recessions, brains, beliefs, experiences, emotions, etc.

A tidal wave can be seen as a large collection of water molecules mostly moving approximately up and down, or as a huge amount of potentially destructive energy moving horizontally. The latter depends on the former, but is no less real, and the causal relations have a kind of circularity.

Likewise there cannot be poverty without physical conditions in which human needs are not met, but poverty is real and can be a cause of crime, of political restlessness and physical events like damage to property and injuries to people.

An animal is composed of physiological components (blood vessels, lungs, stomach, brain, nerves, etc.), which in turn are composed of physical atoms, molecules, etc. But, like the tidal wave, or a boat that is completely rebuilt over many years, the animal persists over a long time even if the "bottom level" physical components are replaced. Moreover, besides physical components, the animal can acquire information, skills, and motivation, and such information processing may cause it to produce new kinds of behaviour with many effects in its environment, including physical effects.

Our world contains machines of many types, at many levels of physical scale, at many levels of abstraction, with many different kinds of causal powers, and many different kinds of laws. This is one reason why we have many sciences, not only physics, but also chemistry, electronic engineering, software engineering, geology, meteorology, biology, psychology, sociology, anthropology, ....

When a huge asteroid (or comet?) hit the earth millions of years ago it caused devastating movements of matter and energy all over the earth. When Diana died it was information that flowed around the planet causing much grief, consternation, curiosity, re-scheduling of television programs, and huge amounts of printed material.

Precisely how to describe the different levels and their relationships is often very difficult. In the case of a tidal wave and its molecules, it's fairly easy, whereas the relation between experiences and brain processes is somewhat harder, and the relation between poverty and experiencs perhaps a mixture of the two kinds of difficulties.

A fairly new class of supervenience, or implementation, relationships is to be found between so-called "virtual machines" in computers (e.g. word-processors, compilers, theorem provers, email systems) and the digital electronic infrastructure. Only a philosopher would say that the software components (like a spelling checker) and its causal powers (correcting spelling in an editor buffer) are unreal because they are not physical. Are poverty, crime, economic recession, news reports, and their causal interactions unreal?

We have much to learn about how different levels and types of reality can co-exist, how one can supervene on or be implemented in (or realised in) another, but that they do is not in doubt.

Arguing about which level of machine is the right one to study in any particular case is pointless: different levels have their own properties and need to be studied in different ways.

Sometimes focusing on an interesting and important level means ignoring many detailed lower level differences between instances (which does not mean others should not attend to them).

Finding new levels of abstraction at which interesting things happen can be a major advance in science.

Examples include studying information processing architectures independently of their implementation in living or non-living matter, and studying developmental and reproductive processes independently of their implementation in physical or information structures. As for whether these systems are "really" thinking or "really" alive, who cares? Only people attracted to rigid dichotomies?

For more on this see this draft paper:
Filename: Sloman.supervenience.and.implementation.ps
Filename: Sloman.supervenience.and.implementation.pdf
Title: Supervenience and Implementation: Virtual and Physical Machines

(b) The principle of multiple realisations of intelligence.

One application of the previous ideas is the notion that there is a very general and abstract concept of intelligence which involves the ability to acquire and use information. In that sense even an amoeba is intelligent and so are very simple robots and many software systems.

Arguing about whether they are "really" intelligent is futile: the important task is to understand what sorts of intelligences there are, how they differ and what the implications of those differences are.

Even elementary knowledge of biology shows that in nature there are myriad different types of intelligence, all using information but using it for different purposes and in different ways, requiring a host of different sorts of information processing architectures, sensing systems, processing mechanisms, processing strategies, forms of representation, divisions of labour between parts of one organism or between different organisms in a colony, etc.

Within a complex organism there can be many different kinds too: for instance the difference between acquiring information about a predator and taking evasive measures and finding information about damaged cells and repairing them, which can happen in our bodies without our knowledge.

Yet more types of intelligence are being developed in many ways using computers, e.g. designing them, evolving them, letting them bootstrap themselves by learning, etc.

Can we hope to understand this diversity? Perhaps not in complete detail, but I suggest that one way to impose structure is to explore the structure of ``design space'' the structure of ``niche space'' the relationships between them and the trajectories that can occur in those spaces, as individuals develop, species evolve, or machines are repaired or modified.

In particular, instead of wasting time on fruitless debates about whether a particular type of mechanism (e.g. condition action rules, theorem provers, or neural nets) is or is not necessary for intelligence we can explore the trade-offs in using or not using it, and analyse the effects of those trade-offs on options available during evolution.

In many cases differences in interactions between trade-offs and niches caused divergence, so that different solutions developed in parallel. For instance, we find both systems where large numbers of small, expendable, individuals with relatively simple combinations of reactive behaviours suffice, and systems where there are relatively few, much larger, individuals with far more sophisticated information processing mechanisms, for instance planning and reasoning capabilities.

My talk will summarise some of these issues and if there's time I'll try to show how the architecture of human minds can be traced back through some evolutionary design choices leading to at least three importantly different concurrently active types of processing layers responsible for different aspects of our mental life, e.g. three different sorts of emotions.

For more on this see the papers listed in the Cognition and Affect project directory Especially:

this paper introducing the main concepts:
Filename: Sloman.kd.ps
Filename: Sloman.kd.pdf
Title: Architectural Requirements for Human-like Agents Both Natural and Artificial. (What sorts of machines can love? )

also this one,
Filename: Sloman_iberamia.ps
Filename: Sloman_iberamia.pdf
Title: The ``Semantics'' of Evolution: Trajectories and Trade-offs in Design Space and Niche Space.

and this discussion note on the relations between AI and Alife.

I've prepared a short paper for careers advisers attempting to answer the question "What is AI?"

Also very relevant:
D.C. Dennett, Kinds of minds: towards an understanding of consciousness, Weidenfeld and Nicholson, London, 1996.

A very readable online document written in question-and-answer form attempting to answer the question "What is AI", written by the person who invented the name for this field, can be found at John McCarthy's web site at Stanford University, California.

Another useful online source, provided by one of the founders of AI, is Marvin Minsky's web site at MIT in Cambridge Massachusetts. He has a useful online paper discussing various approaches to AI and the need to combine them.

Last updated: 25 Jun 2002