Gunnar Sommestad
June 2005

Social Science, Ontology and AI - A Pragmatic View


Before my 15-year period as a software developer and my work with The Literary Machine software, I worked ten years in government, calculating the Gross Domestic Product and running studies to project future trends in the energy sector. Since then, some basic questions about social science methodology have stuck in my mind:


I will arrive at the answer after a tour over the fields of digital computing and Artificial Intelligence.

The "Synthetic apriori" choice

Where to start?  There are many philosopher start-ups in ontology (the nature of being and existence). Aristotle, Plato, Leibniz, Kant, etc all have their specific agenda for primary assumptions. Among them, Immanuel Kant really underlines the difficulty of postulating the "bootstrap" platform for all our understanding.

It is certainly an uneasy situation. As a philosopher you probably have to add some pragmatic intuition to get around it.
 

1. Demarcations

"Demarcation" - alias "system" comes to my mind. It seems important to wonder about how things are formed into objects. Indeed, the philosophical debate since the Greeks has been about the nature of "substance" and "things".

With the emergence of Artificial Intelligence research the actual perception of reality is focused in a more practical way. It is our perception that in fact delivers the "substance". And substances are not only things like stones or planets; we also perceive organic beings, or even social contexts and phenomena. ("Perceive" to underline that the process is automatic and non-deliberate.)

Our perception delivers contact with reality. Without it, the interest in consciousness appears senseless.

The pragmatic view here is that a critical analysis of perception in a broad sense is needed. What kind of perception and system formation in mind gives us understanding and prediction capability?  

The "substances" and systems that we find are basic entities in our intelligence at work.

2. Symbols

Second selection is "symbols". Understanding more of cognition and language we all increasingly postulate symbols, connecting links and some form of computer-like processing.

We experience the information flow from perception to the condensed language and memory. In psychology this journey has gone from William James in the nineteenth century up to the attempts today to get on with digital computer or neuronal-inspired models.

While not really succesful in creating artificial intelligence, we see the program code traces around us: biochemistry, DNA, language. It shows the back of something going on. I will come back to the fact that the "code" (e.g. DNA) is physically located. It might be of importance for the understanding of the "cybernetic" natural laws.

3. Time

Time of course. Time is basic just like system and symbol. Time is reflected in the physical universe (expansion, relativity theory, the law of entropy).
 
Here I will point at "innovation", which develops with time. While not irreversible as the second law of thermodynamics, it adds new options for causation as we proceed along a timeline.

Features

So much for the "synthetic apriori" platform of things that just is there. Let's move to some specific features that could illuminate the Social Science vs. "Physics is the king" scenario.

Non-determinism

Determinism is not a reasonable choice. Let's split the problem into two aspects.

  1. Our knowledge and our ability to predict deterministically

  2. The "Block Universe" view - Is everything deducible from start?

The totally deterministic universe has been a heated topic in physics. Albert Einstein never gave it up. However, quantum physics points in other directions. Experiments indicate "non-local" effects and that some processes can be described only statistically.

Even if not so, the actual prediction task seems insurmountable. Physical laws are conditioned on a clean local environment. When calculating the orbit of a planet, you could probably never predict all gravitation effects from distant objects or effects of interstellar collisions. What remains is that we know of states of the world and have "rules of thumb" for causation from time-1 to time-2.

The general conclusion is that causation is (at least) conditioned on stability of assumed local conditions. Further down I will point at innovation, which considerably increases the problem of stable conditions for causation.

Induction

Induction starts immediately for us. We automatically prefer "no change"/"same procedure". This goes for simple perception and it goes for complex prediction in our mind.

Three aspects of induction:

  1. Learning.

  2. Prediction.

  3. Preconditions

Induction proof and Induction as prediction are of course two sides of the same coin.
Aspect 3 is of interest here. When making a forecast, a certain aspect or process may either be internal and carefully studied or it may be (implicitly) referred to the outside.

To be in fault assuming business as usual is of course common in forecasting. However note that it seems difficult to believe in the theoretical possibility of covering all factors. This applies to physics proper (see above) and physics proper in turn has but little to say about societal development.

Innovation

Organic life, human cognition and social phenomena add a dimension of complexity above the conventional natural science explanation.

"Dynamic systems" or "symbol-program systems" could be the name. We see a continuum of advanced systems in biochemistry, organic life with DNA, human knowledge, the written word, and organisation of society. This is a layer of  "substance and things" above the physical science layer.

We also see an evolutionary process and accumulating processes of innovation.

Innovation brings about that a possibility X2 is available for causation at time T2, but not at T1, T2>T1. That is, a certain persistence for X2 is assumed.

The kind of innovation that we know about is manifested in code: DNA, knowledge learned, written text, computer systems.

Although persistent, it is not irreversible. Since it is locally bound to organic life, machines and their coding, it can be destroyed.

Summing up

The cybernetic reality of organisms and machines offers an alternative "hard core" model for theory, a model without the usual kind of mathematics as the final word. Thus, knowledge about the weak position of classical natural science and its mathematical paradigm should have an impact on Social Science work planning and presentation of result.

Artificial Intelligence may be very far away in the sense of reproducing human cognition performance, but on the other hand the psychology of man shows every sign of being similar to man-made computers. Our cognition cooperates with all these lower level neural mechanisms, which appear to be some sort of computer programs. For the moment it seems as a good guess that human cognition itself is some kind of cybernetic process.

Cognition should probably be viewed broadly: There are many competing activities within us humans and among individuals in a society. The result is apparently an evolutionary pressure manifested in innovation development.

Still, we do not know very much. The brain and the DNA seems to be a physically located coding in the process, but of course an extravagant hypothesis is that "non-local" immaterial phenomena may add other things to it.

What this boils down to is that the premises

  1. Non-determinism
  2. Impossibility of a total prediction based on causation
  3. Reality of organic and man-made cybernetic systems

imply the we should

View organic life and other cybernetic systems as equally suitable for model-based  prediction as physical objects with mathematically described laws of physics.


Applications

Causation based on cybernetics in nature is as potent as is traditional "hard-core" natural science theory ultimately relying on physics. This insight corresponds to the actual scientific practice: Many disciplines compete in overlapping research areas. Cybernetic processes in nature (and ultimately intelligence) are introduced by innovative organic life. This calls for new "low-level" scientific disciplines not yet existing.

Social Science

The importance of innovation and change should not be underestimated. Social science often strives to prove what there is in a stable society. However, societies are not just stable, they evolve and innovation brings change.

Theory constructs for social science that in one way or another connects to psychology, linguistics and AI research could be successful.
 

Forecasting methods and future studies

I now arrive to my intital own question about future studies. Best practices? What comes below is about the same as we did in Sweden with "energy futures" in the 1970's. What I can add is a conviction that

A. It is possible to do a good job

B. The task to predict the world in the strict sense of the word is impossible.


Tips:

  1. Find those environments and systems that have the best chances to offer a prediction method for the outcome.

  2. If the objective is to explain developments and highlight options for having an impact on events, select systems with that criterion instead.

  3. Discuss the selection of systems studied.

  4. Apply similar research to history: The analysis of causation is equally hard even if the outcome is known. However, some things happen again and this should be used for what it is worth.

  5. Include studies about how innovation can bring about changes at a future point of time.


What does this mean?

a) You should be open-minded in the selection of areas studied and methods.  
b) Empirical validation of development indicators is useful for the fine-tuning of short-range forecasts.
c) You should observe that alternative processes actually compete in reality.

Forecasts for World population in the 1960's are examples of really bad forecasting; they did not allow for social adaptation of a well-known type. Other examples can be found where the potential for mental change is overestimated.

Dynamic social and economic processes with feedback are common in reality and difficult to master. Economics, being an old social science, has some fine capabilities in that respect.

Brainstorm-like scenarios are often added to cover unexpected events. They need not be economic in an economic forecast. Earthquakes, climate change and ? could happen.


Gunnar Sommestad