• Keine Ergebnisse gefunden

Increasing knowledge will always improve modeis, with no upper bound observable so far

Knowledge and Information

2. Increasing knowledge will always improve modeis, with no upper bound observable so far

2 The Artificial Life Route to Artificial Intelligence, Luc Steels & Rodney Brooks eds, LEA 1995 3 There are those who will disagree with this

4 Schopenhauer, A., "The World as Will and Representation"

5 Wittgenstein, L. J. J., "Philosophical Investigations"

On Knowledge

Knowledge is closely linked to learning. In fact, knowledge is that-which-has-been-learnt6 The word knowledge is also commonly used to refer to what I have termed Information (knowledge-base for example, actually refers to a computerised störe of information), and strictly speaking the two are completely different, and so we need to make a distinction for the purposes of this paper.

The key to understanding what knowledge is, is that it is species-specific - i.e. different sen-tients with different physical characteristics and environments will, in order to survive, build complex modeis of their experiences; these modeis are knowledge, and depend on the envi-ronment and physical characteristics of those sentients who are learning.

The more complex the modeis they can build, the more sentient they are. What we must remember is that at every step of our learning and model building we have to examine the hidden assumptions our humanness has built in to our modeis, and be sure that we are aware of the consequences on the potential accuracy of our knowledge.

On Motivation

Humans, because of the biological imperative to reproduce, are forced to survive: to eat, sleep, breathe, build shelter, organise tribes, build supermarkets. It is incredible to think that something as simple as a biological imperative can result in societies, religions, learning, language - but the biological imperative is one of the few absolutes that I presuppose in this paper. For an excellent discussion of the role the biological imperative has to play in science, see7

I link the assumption of the existence and validity of the biological imperative to that of sen-tience, and based on these foundations attempt to shed some light on how we currently or-ganise information, how we communicate and how we may be able to do these things better infuture.

On Communication

Communication is what we do with other sentients. Or is it? We have something of a problem here, because we must choose somewhere between the two extremes of "gestalt" and

"mathematical" communication.

A gestalt definition of communication refers to the idea that everything we perceive contrib-utes to communication, including for example the passive parts of our environment.

"Mathematical" communication is that which can be performed using symbol Systems, nor-mally also between humans.

Both are extreme definitions: gestalt because it is undirected, and mathematical because it is context-f ree and cannot be used to teil us anything about the act of communication.

What is then communication? Once more we have to turn to the biological imperative, and define communication as that which sentients use to exchange parts of their worldmodels -this definition includes the case of two neighbours gossiping overthe garden fence, the sci-entist trying to talk to killer whales, but perhaps more importantly it also includes everything we communicate in a social sense, i.e. it includes society as a means of communication.

6 This is an example of a folk-based axiom, which can be paraphrased as "someone always has to learn the hard way"

7 Dennet, Daniel C; "Darwin's Dangerous Idea"; Simon & Schuster, 1996; 0-684-82471-X

On Metaphor

Metaphor, traditionally considered to be solely a linguistic phenomenon, has undergone something of a revolution in the last two decades. Starting with the work of Eleanor Rosch and George Lakoff8' modeis of how humans think are now emerging in which metaphor is not just linguistic colouring, but a core cognitive process. These modeis of thought are helping us to understand how we think and how we communicate.

As infants we learn to communicate and interact within the real world by learning what OB-JECTS are, what CONTAINERS are, what a BOUNDARY is, the difference between INSIDE and OUTSIDE, and our entire life is accompanied by a process of assimilating new informa-tion into existing modeis, creating new modeis where necessary, and using those modeis to predict how our environment will behave.

Classification and ordering of sensory information within human thought has been shown by Lakoff to be based on a few such simple idealised cognitive modeis. We use these modeis from the moment we are born, and all subsequent classification and categorisation are based on them. We are forced to look for similarities and differences, and at a level that is directly meaningful through our senses. This leads to a hierarchical approach to classifica-tion, although the tendency to hierarchically categorise sensory information is itself

ulti-mately a direct result of embodiment.

On Learning

Building these modeis of how the world behaves is the process of acquiring knowledge. The process of adding to knowledge is continuous, with all sensory experience reinforcing one or the other of our modeis of the way the world works.

Children learn a core set of modeis very early on in their lives. This core set of modeis is responsible for all subsequent higher level modeis - for example mathematics is fundamen-tally based on the modeis of CONTAINER (sets), OBJECT (numbers), UP_DOWN (order-ing) and LINK and SYMMETRY (modell(order-ing).

I have listed some of the core modeis which humans learn - although this list is not complete.

8 Lakoff, G., "Women, Fire and Dangerous Things", Chicago Press

Model Name Depends on Frame of reference OBJECT BOUNDARY Embodiment at the direct

physical level CONTAINER OBJECT, BOUNDARY, Feeding

INSIDE-OUTSIDE

AGENT SOURCE-PATH-GOAL, Action at a distance, OBJECT relationships SOURCE-PATH-GOAL LINK Motivation INSIDE-OUTSIDE CONTAINER Seif Awareness MIRROR LINK, FRONT-BACK Seif Awareness LINK OBJECT Relationships PART-WHOLE CONTAINER Building, Destroying FRONT-BACK SYMMETRY, NEGATION Motion

BOUNDARY OBJECT, INSIDE-OUTSIDE Consciousness UP-DOWN PATH Locomotion, Danger CENTRE-PERIPHERY INSIDE_OUTSIDE Family

1. OBJECT is anything we can touch. We learn to identify and manipulate OBJECTS, later extending this model of an object through differentiation to CONTAINER and INSIDE-OUTSIDE. In its turn, OBJECT becomes the basis for scientific modeis.

2. CONTAINER is learnt implicitly during feeding, but its first explicit use comes with the recognition of seif. CONTAINER borrows from OBJECT, INSIDE-OUTSIDE and STRUC-TURE.

3. AGENT is learnt as the model for "providers". AGENT is also a model that finds great Utility in our social interactions - and it is also one of the modeis that is most clearly mis-used. Nevertheless is has very great Utility, because it allows us to attach attributes of OBJECT and CONTAINER to situations where there is no clear OBJECT or CONTAINER - it acts as a type of multiplier.

4. SOURCE-PATH-GOAL is, next to OBJECT, one of the earliest modeis we learn. We use it when we Start to move, and it provides us with a model that we can use to obtain OB-JECTS.

5. INSIDE-OUTSIDE is learnt indirectly during the process of feeding, although it does not become fully developed until a sense of SELF develops. It is closely related to CON-TAINER, since most CONTAINERS have an inside and an outside, and it finds Utility deep in the scientific disciplines.

6. MIRROR is learnt relatively late, and does not necessarily result from direct experience with a mirror. Children who are conf ronted with a mirror at an early age are able to grasp this model immediately - a facility which humans share with surprisingly few other sen-tients.

7. LINK is closely related to SOURCE-PATH-GOAL, although different in important re-spects. In contrast to SOURCE-PATH-GOAL there must not be an associated purpose to the relationship modelled by LINK. LINK is also the basis for further extension mecha-nisms within the hierarchy of cognitive modeis, underlying such mechamecha-nisms as meta-phor and metonymy.

8. PART-WHOLE draws on CONTAINER, extending the model to include such aspects as internal structure, and when used in conjunction with LINK can provide modeis of the in-ternal behaviour of OBJECTS.

9. FRONT-BACK is closely related to SOURCE-PATH-GOAL, because a sense of forward direction is associated with both.

10.BOUNDARY is available to us as a model after we leam to move, and becomes fully de-veloped when we perceive seif.

11.UP-DOWN is synonymous with our experience of weight and gravity, and is used to model extent and danger.

12.CENTRE-PERIPHERY, initially used in social situations becomes a powerful tool when applied to the process of categorisation itself.

Orthogonality

One of the most important points about these modeis is that they are not orthogonal. Ortho-gonality means sets of things that cannot be expressed in terms of each other, and which can be combined in some way to produce every type of thing possible.

The cognitive modeis listed above do not have this property, but ultimately they are all that we have.

On Philosophy

The major goal of western philosophy forthe last 2500 years9 has been more or less a search for such an orthogonal set of things. In contrast, what is new here is that we acknowledge that what we have learned by observing

• learning

• pathological exceptions

• physiological dependencies

is that any set of human modeis cannot be orthogonal. The new modeis described here nei-ther claim to be a set of things which cannot be expressed in terms of each onei-ther (a CON-TAINER is an OBJECT with an INSIDE-OUTSIDE for example), nor do they Claim to be able to provide a mechanism for deriving all things. Indeed, it is fair to say that as far as embodied thought goes, this new view of thought teils us quite clearly that such a mechanism is a holy grail and will by definition always remain out of reach.

Thanks to the improving Performance of Computers and the means to communicate between them, information is becoming a commodity resource10. The costs associated with being in-formed about a particular topic are dropping rapidly: the example of the world wide web, with 60000000 static documents11, represents a formidable amount of information. The major problem with using such vast amounts of information is that it is very difficult to leverage it: in a given Situation, the "needle" of information one wants is hidden beneath a planet-wide hay-stack of biblical proportions12.

Most approaches to this problem have been reductionist or objectivist13, viewing thought as consisting of distinct areas separated by well-defined boundaries. The solution to the

Infor-9 without trying to over-generalise

10 This means that the effort we expend is rapidly diminishing for access to raw information, but re-mains fairly constant for filtered information.

11 As of July 1996. Growth rates in this figure are at least 10%permonthcurrently.

12 Mattingley-Scott, M. A., "Information Explosion and the World Wide Web", IEEE Multimedia News-letter Vol.4 No. 2 May 1996

13 Brooks often uses the term "abstract"

mation Explosion problem has often been translated into a search forthe "true" underlying rules which would allows a final classification of all information14.

The Web is a counter example of an "object" which does not fit the assumptions of these approaches: it has no boundaries, has a small number of simple rules, and is capable of Standing forthe humans who use it.

Nothing in the real world (as opposed to modeis of the real world) is fixed however: When something comes into existence it redefines reality, extending what humans can do, ulti-mately leading to often unimaginable consequences. Reality then may be better viewed as a continuum: continually under pressure from social constraints on one side to conform, stand still, be defined, be quantifiable, manageable, safe; on the other side it is continuously being redefined and extended by our existence and human inquisitiveness. This conflict is very real, unavoidable in society, and of course sometimes very dangerous.

On Humans

Underlying our consciousness is a continuous matching of our current Situation with modeis that our memory makes available to us. This matching results in a given behaviour for a par-ticular Situation, based on the Utility of a parpar-ticular scenario. The timescales involved may be very Short (fractions of a second) or very long (over a lifetime).

The important point is that we are continually re-assessing our behaviour and modifying it in the light of our experience. Feedback occurs of course, because one of the things we do very well is to observe the way our behaviour changes our environment. Environment in this sense is not just things like the effects of ten's of years of pollution on German forests, but also the effects of a smile on one's mother. In short, we are continuously doing the following:

1. Observing viaoursenses

2. Under intense pressure due to limited resources (principally time), looking for a model to help us understand what is happening

3. Under intense pressure due to limited resources, modifying the model to try to make it better fit the actual observed information

4. Applying our learnt behaviour for that Situation to the new Situation Two effects result from this continuous process:

We stay alive

We learn, or acquire knowledge

This process, and incidentally the results of it, we seem to share to a greater or lesser degree with all other animals. In humans however, the speed and agility with which we are able to find modeis of how the world functions and use them to help us survive is unparalleled, we assume.

References

Shannon, C. E., "Communication in the Presence of Noise", Proc. I.R.E., Vol. 37, Jan. 1949, pp.10-21

The Artificial Life Route to Artificial Intelligence, Luc Steels & Rodney Brooks eds, LEA 1995 Schopenhauer, A., "The World as Will and Representation"

Wittgenstein, L. J. J., "Philosophical Investigations"

Dennet, Daniel C ; "Darwin's Dangerous Idea"; Simon & Schuster, 1996; 0-684-82471-X 14 A classic red-herring is the CYC-O project, which set out to classify all human knowledge.

Lakoff, G., "Women, Fire and Dangerous Things", Chicago Press

Mattingley-Scott, M. A., "Information Explosion and the World Wide Web", IEEE Multimedia Newsletter Vol.4 No. 2 May 1996