Rational Behavior models vs' Learning

1
Rational Behavior modelsvs. Learning

• David CAYLA
• INAPOR Meeting
• LUISS, ROMA
• May 11, 2009

2
Introduction

• What rationality model should we use in order to
  apprehend learning?
• Simonian bounded (or procedural) rationality?
• Two learning models
• Adaptive learning
• Teleological learning (strategic)
• Organizations vs. market the role of strategy in
  learning
• It is necessary to develop a rationality model
  that is compatible with a teleogolical learning.

3
1. Rational Behavior

• Walliser (1995) two sorts of rationality
• Cognitive rationality
• The appropriateness between the agents beliefs
  and the information he received.
• It is used to build the agents expectations

Cognitive rationality
INFORMATION
BELIEFS REPRESENTATION
4
1. Rational Behavior

• Instrumental rationality
• The appropriateness between the agents
  opportunities and his preferences.
• It is used to build a strategy that matches his
  expectations.

Instrumental rationality
Opportunities
Preferences
5
1. Rational Behavior

• Instrumental rationality

Chess games are perfect illustations of pure
instrumental rationality problems. (here the
question is how to mate in one?)
6
1. Rational Behavior

• Perfect rationality and learning
• Perfect cognitive rationality
• Perfect instrumental rationality
• Learning is just a process of information
  gathering Batesons level zero learning.
• Learning is an eductive process (Walliser,
  Binmore) agents behavior change
  instantaneously when they get new information.
• If we are in a perfect information world
• Perfect coordination by the Lewisian hand (all
  agents have common knowledge) (Walliser 1998).
• No uncertainty on behavior, no dynamism, no
  process and no strategy each action is
  independent from the others

7
2. Learning and Bounded Rationality

• Limits in instrumental rationality
• Preference for a satisfacing outcome through
  specific procedures (or routines) instead of
  solving an optimization problem.
• Limits in cognitive rationality ?
• For Simon, representations are imperfect we
  must distinguish between the real world and the
  actors perception of it and reasoning about it
  (Simon 1986, p. 211).
• We cannot be perfectly informed. Information have
  a cost and is hard to find.

8
2. Learning and Bounded Rationality

• Limits in cognitive rationality?
• In Simons approach, human beings and ants are
  data processing units, trying to adapt their
  behavior to their environment.
• An ant, viewed as a behaving system, is quite
  simple. The apparent complexity of its behavior
  over time is largely a reflection of the
  complexity of the environment in which it finds
  itself (Simon 1996., p. 52).
• Human beings, viewed as behaving systems, are
  quite simple. The apparent complexity of our
  behavior over time is largely a reflection of the
  complexity of the environment in which we find
  ourselves (Ibid., p. 53).

9
2. Learning and Bounded Rationality

• Three consequences
• Learning procedures are universal
• For Simon, an ant learns like a human being by
  adapting to its environment.
• Learning is a memorization process (chunk
  acquisition)
• 50,000 chunks are needed to master a discipline
  like chess (Simon 1996).
• Reality is objectively perceived
• Even if there are limits to their sensorial
  capacities, there are no systematic cognitive
  biases.
• If two agents face the same environment, their
  representations will spontaneously converge.

10
2. Learning and Bounded Rationality

• So the Simonian model of rationality implies a
  perfect appropriateness between the agents
  beliefs and the information he received.
• In Simons bounded rationality conception, what
  is imperfect is not the interpretation of a
  received information (the way it becomes belief)
  but its perception (the way it is acquired).
• Agents live in different worlds and have
  different representations only because
  information is costly (and long) to acquire, not
  because they interpret information differently.
• So Simonian Bounded rationality model rests on a
  perfect cognitive rationality. It follows a
  cognitivist approach (Varela 1997).

11
2. Learning and Bounded Rationality

• Simonian learning model rests on
• Imperfect instrumental rationality
• Costly and imperfect information
• Perfect cognitive rationality

Learning
Cognitive
Instrumental
Improve representations by acquiring information
Improves procedures of action
12
2. Learning and Bounded Rationality

• The Marshall Plan example (Simon 1996)
• Creation of the Economic Cooperation
  Administration (ECA) in 1948
• Six main conceptions and objectives
  (representations) that were mutually
  contradictory.
• Q How to coordinate the ECA actions?
• Simons answer
• Adaptive learning (by feedbacks) is better than a
  planned process (more flexible and compatible
  with our limited cognitive capacities).
• A unique conception will naturally emerge in the
  field (but he does not say how).
• Howhever, adaptive learning is not relevant to
  understand the ECA coordination problem.

13
3. Solving the ECA problem

• We need alternatives to cognitivism that allows
  imperfect cognitive rationality (subjective
  representations).
• Connexionism
• Sub-symbolic approach knowledge depends on the
  whole structure of the cognitive system.
• Information is interpreted when it is acquired

Cognitive rationality
INFORMATION (objective)
BELIEFS, REPRESENTATION (subjective)
Interpretation
14
3. Solving the ECA problem

• Enaction (Varelas approach)
• A circular relation between the world and the
  mind.
• The world properties are not predefined, they are
  enacted. They are the product of actions.
• The perception you have depends on the
  perspective you follow your Intentionality
  (Searle 1985)
• Information is not given, but constructed

One of the lead of this book is that ordinary
life necessarily implies embodied agents, who are
continuously confronted to the problem of action,
whereas their different sensory-motor systems are
engaged in parallel activities. The
corporality implies that the cognitive entity has
by definition a perspective. It means that
its links with the environment are not
objective, independent from the situation, from
attitudes and from the system history. On the
contrary, these links are closely related to the
perspective established by properties that keep
emerging from the agent itself, and to the role
played by these redefining in the coherence of
the whole system (Varela, 1997)
intention
Mind
World
enacted
15
3. Solving the ECA problem

• An embodied agent
• circular relation between knowledge and behavior.

• What you know is what you do
• What you do is what you know

• A cognitive entity with a perspective
• circular relation between knowledge and intention.

• What you know is what you seek
• What you seek is what you know

16
3. Solving the ECA problem

• SIMON learning is limited to a memorization
  process that aims to adapt the organism to a
  given reality.
• VARELA learning is a construction that emerges
  in the relationship between the individual and
  its environment.
• A color or a smell are not the objective products
  of nature, but emerge in the perception
• Information is a modern phlogiston. It does not
  exists by itself.
• Knowledge is the product of a subjective
  creation. Its meaning depends on the
  characteristics of the cognitive system.

17
4. Intentionality and Identity in Learning

• Varelas cognitive learning represents a
  cumulative process that is torn between its need
  for internal coherence and the necessity it has
  to be flexible
• What you learn from the outside world helps you
  to deepen your intentionality
• But everything you learn must be coherent with
  this intentionality.

18
4. Intentionality and Identity in Learning

• In imperfect cognitive rationality, what would be
  a rational learning?
• RUSSEL Rationality signifies the choice of the
  right means to an end that you whish to achieve.
  It has nothing whatever to do with the choice of
  ends.
• But in Simons view, learning is substantive
  what is important is the capacity you have to
  adapt yourself to your environment.
• Our proposition (following Varela and Russel)
  Rational learning signifies the choice of the
  right knowledge accumulation to an intention you
  wish to adopt. It has nothing whatever to do with
  the choice of intentions.

19
4. Intentionality and Identity in Learning

• Identity as the driving force of learning
• How does the intention evolve? How do we choose
  an intention?
• Existentialist view Identity does not exist, we
  only make rational choices that are related to
  our preferences.
• Substantialist view We discover rather than
  choose our identity (Sen 2004, p. 10).
• Philip Grill Identity is conquered, it is a
  way an agent gives a sense to its own actions,
  trying to gather its life under the form of a
  story (Grill 2004, p. 90).

20
4. Intentionality and Identity in Learning

• If intentionality is defined as the relationship
  that an individual develops with its environment
  and on the basis of which he constructs his
  cognitive system, then
• identity can be defined as the relationship that
  an individual develops with himself and on the
  basis of which he creates his own intentionality.

• What you are is what you seek
• What you seek is what you are

21
4. Intentionality and Identity in Learning
22
5. Conclusion
Subtantive rationality
Perfect rationality
Simons bounded rationality
Connexionism approach
Varelas approach