School of Computer Science Ghost Machine

What's information?
An Answer from Physics, Biology, MindScience and Philosophy

Aaron Sloman
School of Computer Science, University of Birmingham, UK

A subset (not yet decided) of these ideas will be presented on Thursday 2nd June 12:00-13:00 at the Hebrew University of Jerusalem. at:
The 30th Annual International Workshop on the History and Philosophy of Science
Information and information-processing in science:
Biology, Physics, and Brain & Cognitive Sciences
Research Workshop of the Israel Science Foundation
Monday-Thursday, 30 May - 2 June, 2016

(DRAFT: Liable to change)

DRAFT EXTENDED ABSTRACT
(critical comments and suggestions welcome)

It is no more possible to define the notion of "information" under discussion here than to define "matter" or "energy".
There is a pseudo-definition wrongly attributed to Gregory Bateson, namely "Information is a difference that makes a difference", which might do credit to an advertising copy writer but lacks both substance and accuracy. I have criticised both the attribution and the definition here, and elsewhere: http://www.cs.bham.ac.uk/research/projects/cogaff/misc/information-difference.html
(Bateson wrote something different, and much less sweeping.)
Instead, we have to develop a theory or collection of theories specifying what information can do in various contexts. A good theory will stimulate research of a type that leads to new discoveries including discoveries of new problems, new phenomena to be explained, and possibly also gaps or other deficiencies in the theory that need to be remedied. In short we need a concept of "information" that plays a key role in a progressive scientific research programme -- contrasted with degenerating research programmes by Lakatos (1980).

Moreover concepts implicitly defined by a complex theory can survive minor changes in the theory, which makes it possible to say, for example, that we have found a new more accurate way of measuring the charge on an electron: the measurement method can change while the role of the concept in the theory (in this case "charge on an electron") is mostly preserved. The theory implicitly defines the concept. A change in the theory may amount to a relatively minor change in the concept, which would not be the case if the concept were defined by a method of detection or measurement. The concepts of "matter", "energy", "location", "speed", "space" and "time", received a bigger battering at the hands of Einstein around a hundred years ago.

Similarly a good theory about the role(s) of information in the world will implicitly define the concept of "information" as explained in Sloman, 2011. There are many different kinds of information with slightly different roles, not discussed here.

The Turing-inspired Meta-Morphogenesis (M-M) project arose from the question: "What would Alan Turing have worked on if, instead of dying two years after publication of his morphogenesis paper (1952), he had lived several more decades?"

My conjectured answer is: he would have tried to explain how a lifeless planet (or universe) could generate all the forms of life, and all the (evolved and still evolving) forms of biological information processing (including mathematical information processing resulting in Euclid's discoveries) that have existed on Earth.

The key feature of information is not what makes it relevant to information engineers, e.g. the amount of storage needed or the physical requirement for reliable transmission, but its content.

The science of information content and its biological uses is still in its infancy. The content is related to what the information can be used for. The notion of use of information is multi-faceted and its extension (the variety of types of use of information) changes as a result of evolution, one of many forms of "feedback" involving information.

Many biologists are already studying transitions in morphology, behaviours, environments, genetic codes, and epigenetic mechanisms and processes. I suspect that if he had survived several more decades Turing was likely to have investigated transitions in biological information-processing competences and mechanisms, including the many forms of information-processing required for evolution by natural selection or produced as side-effects. That includes human uses of language for communication and also the much older and more wide-spread uses of internal languages for control, perception, learning, desiring, intending, planning, formulating questions, etc., which must have evolved first, contrary to popular opinion Sloman 1978BBS.

This study can benefit from expanding and re-deploying ideas and theories, developed in the last 70 years, about varieties of types of information, uses of information, information processing mechanisms, information processing architectures, and, above all, the crucial importance of complex virtual machines made of interacting virtual and physical machines of many types, now used every day unwittingly by philosophers, scientists and even pre-school smart-phone users Sloman (2013a). A closely related, but slightly less complex analysis is presented in Maley and Piccinini (2013).

One of the earliest things learnt as a result of development of computing (though there were precursors in music boxes and Jacquard looms) was that control should not merely be thought in terms of changing physical states and processes with numerical values. A great deal of what is known as "Control Theory" is concerned with more or less complex systems with measurable states and state changes, in which everything that is changed, and everything that produces change takes on numerical values, usually related by equations, including differential equations. (Compare the common connotation of "cybernetics", and the collection of ideas known as Perceptual Control Theory associated with William T Powers.) This was undermined by the fact that control in early computers was not associated with continuously changing physical measurements, but with discrete sequences of instructions producing discrete changes in various parts of a potentially very large collection of memory items.

Another strand of change came from logic and linguistics and later on the design of computer languages with their analysers, compilers, interpreters, and even automatic program generators. In these cases control was associated not with fixed "control loops" but with constantly changing activation of rules, of logic or grammar or arithmetic: this broadened minds of some, but not all researchers. Many remain who attempt to understand a brain, for example, only in terms of a fixed collection of numerical values that can change with various sorts of causal linkage causing changes to propagate. That style of theorising made it hard (or impossible) to think about how human languages, and various kinds of logical and mathematical reasoning work, and less obviously, how perception of complex structures, such as animals, flowers, trees, buildings, etc. might be done by brains.

The portia spider, under some conditions, sits and looks at physical environment surrounding its prey, and after several minutes sets out on a sometimes complex route through plants and other structures available until it is above the prey and then drops down onto it. This suggests strongly that the initial perceptual process uses perceived information about the structures in the environment to created a structured executable plan to get to the desired point, sometimes being forced to use routes that temporarily leave the prey out of sight.

So although some control theorists and brain scientists continue to work only with complex sets of equations linking numerical values, others think also about discrete information structures of varying complexity, interacting discontinuously with other discrete structures: e.g. a parser program taking in a string of words and building a parse tree in the computer; or a planning program taking in a specification of a goal state, an initial state and a collection of facts (possibly facts acquired by perception, like the Portia spider's planning facts), and using the facts to build a plan for action to achieve the goal. In the 1960s some computer vision researchers [e.g. Max Clowes(REF)] realised that the structure-manipulation approach rather than the variables and equations approach was required for major functions of vision, in humans and other intelligent animals, so that visual perception became more like language understanding than anyone had previously realised.

My suggestion is that a very long time ago evolution "discovered" the power of discrete forms of structure manipulation and in the process produced new languages for use by brain subsystems concerned with perception, reasoning, planning, and plan execution.

But that might have been impossible if evolution had not, even earlier, evolved structure-building mechanisms for assembling body parts or food stores of varying complexity.

All of this implies that for a very long time (presumably starting long before the evolution of Portia spiders) evolution had produced mechanisms concerned with discrete assembly of physical parts of bodies, and perhaps later produced derived mechanisms for manipulating not the things of interest themselves but internal information structures referring to or representing things in the environment.

We have also learnt that production of such systems requires a huge diversity of supporting "construction kits" of various kinds (concrete, abstract, and hybrid) and various kinds of scaffolding that are essential to the processes whether included in the products or not, all of which are also products of evolution.

Many of these mechanisms make use of mathematical structures, including geometric and topological structures, and somehow evolution acquired, and systematically, used information about those mathematical structures long before there were human mathematicians: a simple example being pervasive use of negative feedback loops in homeostatic control mechanisms. (I call evolution The Blind Mathematician.)

Types of control, and by implication, types of information, can vary in spatial and temporal scope, in types of effects, in energy requirements, and in what is controlled. For example, in simple cases information is used in control of simple physical systems, but what is controlled can vary in spatial and temporal scale. Moreover, one of the major effects of evolution is to modify the use of information to control the use of information: i.e. meta-control.

The earliest, simplest, organisms (I suspect) had a sphere of control that was very local in both space and time: e.g. allowing or not allowing a molecule to pass through a membrane. In such cases the information is both acquired and deployed (used in control) in a small time-interval and spatial region.

In more complex cases the space and time gaps between where and when information is acquired and when it is used can increase, and the the spatial and temporal scale of the effects can increase: e.g. a herd of elephants using (shared?) previously acquired information to control migration to a different source of water.

Other changes are concerned with the heights of "towers" of control: control of control of control, etc. which will also be related to the heights of towers of supply of information: information about information about information, ... etc.

Yet more changes are "architectural", i.e. involving increasingly complex coexisting components acquiring and using information about one another and their environments.

One strand in all of this that particularly interests me was the evolution of mathematical competences required for the achievements of Euclid, Archimedes, and other great ancient mathematicians. Their competences are not yet replicated in computers and it is not clear when they will be, in part because we still do not know exactly how to describe the cognitive functions that were required.

An example was the discovery (by Archimedes?) that a simple and natural extension to Euclidean geometry made it easy to trisect an arbitrary angle.
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/trisect.html

I suspect this had deep connections with biological mechanisms providing abilities to perceive and reason about possibilities (not probabilities) for change in spatial configurations, and limitations (impossibilities) of such changes. It is possible to change any simple planar polygon by breaking one of the sides into two parts forming a new vertex, which is pushed in towards the interior of the new polygon or pushed out. In all such cases the sum of the interior angles will increase by exactly 180 degrees. How can you be sure? What is your brain doing? What mechanisms does it require? What biological functions produced the evolutionary changes that allowed our ancestors (when???) to aqcuire the reasoning competences used? What previously evolved mechanisms made those changes possible? And finally, why has it proved so difficult to replicate those mechanisms on computer-based systems?

By still unknown stages the evolutionary processes, using unknown mechanisms, led to production of many individual intelligent organisms (squirrels, crows, elephants, hunting mammals, etc.) unwittingly using mathematical relationships, e.g. weaver birds using topology. This is totally different from use of statistical learning and probabilistic inference, but seems to have largely gone unnoticed by biologists, neuroscientists, psychologists and philosophers (though Kant and Piaget both had some deep insights, while lacking the concepts required for building explanatory theories).

Later still, apparently as a result of evolution of several layers of meta-cognition, mathematicians such as Euclid, Archimedes, preceeded by generations of perceivers, thinkers and doers without mathematics teachers, began to make, think about, and even discuss mathematical discoveries, about geometry, topology and arithmetic -- long before the discovery of modern axiomatic methods and formalisms. I conjecture that those developments required evolution of new layers of meta-cognition, meta-meta-cognition, meta-meta-meta..., etc. a form of progress also found in simplified forms in the evolution of computer systems engineering.

In humans that included what could be called "other-directed" cognition, i.e. one individual using information (or speculation) and the information processing of other individuals. The biological and cultural evolution of mechanisms and practices of teaching depended on this. But perhaps even more important was the opportunity that provided for forms of reasoning to be challenged and better explanations provided, that later evolved into what Euclid called proofs.

The talk will present some partial results concerning the nature and diversity of biological information and information-processing, along various dimensions sketched above, including connections with other disciplines.

Most researchers seem to focus on a subset of types of information, and information-processing, with major omissions that have bad consequences for science and philosophy. An important first move is to focus on the fact that information is used rather than simply being communicated. Use is of primary importance in determining requirements for and nature of varieties of information and information processing mechanisms. Starting from communication of information will lead down blind alleys in the attempt to understand the role of information in biology, psychology, cultures, etc.

Shannon's metrical notion of information was concerned with requirements for storage and communication rather than use, and vast numbers of thinkers have since been misled by his work -- though I think he understood the difference (and possibly later regretted his choice of terminology?) [Reference?]

The novelist, Jane Austen, in contrast used the older (biologically more important) concept of information in her work:
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/austen-info.html

One of many possible by-products of this research may be future development of machines with human-like (e.g. Archimedes-like) mathematical competences. Current AI isn't even close, as far as I can see. And most brain scientists and psychologists appear not to have noticed what needs to be explained. So that concentrate on trying to explain other things: omitting major aspects of human intelligence and animal vision.

Some links to related parts of the M-M project, and other things.

The Meta-Morphogenesis (M-M) project page is here: http://goo.gl/9eN8Ks

Archimedes knew about an extension of Euclidean geometry that made angle-trisection easy:
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/trisect.html
What sort of mathematical competence is involved in dressing a child?
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/shirt.html
How to see impossibilities:
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/impossible.html
Some of what we have learnt about the powers of virtual machinery
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/vm-functionalism.html
Proto-mathematical information-processing in toddlers:
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/toddler-theorems.html
Evolution's construction-kits (mostly produced by evolution):
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/construction-kits.html
Schrödinger's groundwork, in 1944:
   http://www.cs.bham.ac.uk/research/projects/cogaff/misc/schrodinger-life.html
What are the functions of vision? How did human language evolve?
   http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#talk111

REFERENCES

Installed: 19 Apr 2016
Last updated: 20 Apr 2016

THIS DOCUMENT IS
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/inf-2016.html
     A PDF version may be added later.

A partial index of discussion notes is in
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/AREADME.html

This is part of the Turing-Inspired Meta-Morphogenesis project:
http://www.cs.bham.ac.uk/research/projects/cogaff/misc/meta-morphogenesis.html

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham