Computational Discovery of Communicable Knowledge

1
Experimental Studies of Integrated Cognitive
Systems
Pat Langley Computational Learning
Laboratory Center for the Study of Language and
Information Stanford University, Stanford,
California Elena Messina Intelligent Systems
Division National Institute of Standards and
Technology Gaithersburg, Maryland
Thanks to David Aha, Michael Genesereth, and
Barney Pell. This work was funded in part by
DARPA IPTO, which is not responsible for the
points made herein.
2
Experimentation in Artificial Intelligence

• Controlled experiments are the primary evaluation
  tool in modern AI, including the subfields of
• supervised learning and reinforcement learning
• generative planning and scheduling
• computational linguistics and text processing
• but not for work on integrated cognitive
  systems.
• Extending experimental methods to the latter is
  crucial, since it deals with the ultimate goals
  of artificial intelligence.

3
Challenges for Experimentation

• The reasons that experiments with integrated
  cognitive systems have lagged behind are clear
  from the phrase itself
• systems are harder to evaluate than component
  algorithms
• cognitive methods involve complex, multi-step
  reasoning
• integrated software relies on interactions
  among components.
• Together, these factors have slowed the
  development and wide acceptance of an
  experimental framework.
• In this talk, we propose the key elements of an
  experimental method for the study of integrated
  cognitive systems.

4
Dependent Variables Basic Measures

• Dependent variables in an experiment measure
  system behavior.
• Some basic measures of integrated cognitive
  systems include
• success or failure on a given problem
• speed or efficiency of the systems response
• desirability or quality of the systems
  response.
• Such metrics provide the building blocks for more
  sophisticated and informative measures of
  behavior.

5
Dependent Variables Combined Measures

• Statistics tells us we should not draw
  conclusions from one case.
• Collecting multiple samples supports combined
  measures like
• average behavior of the system
• cumulative behavior of the system
• variance of the systems behavior.
• Combined measures also partly cancel variation
  due to unknown or uncontrolled factors.
• However, this requires some population from which
  samples are drawn, which one should always
  specify clearly.

6
Dependent Variables Higher-Order Metrics

• Combined measures present only a small window on
  behavior.
• However, one can also derive higher-order
  measures such as
• the slope and intercept with respect to a
  control system
• the intercept, rate, and asymptote of a
  learning curve.
• Such metrics let one summarize behavior even when
  variation across samples is not systematic.
• Conclusions about higher-order measures are more
  important than ones about basic or combined
  variables.

7
Independent Variables Task Characteristics

• Independent variables in an experiment reflect
  factors thought to influence system behavior.
• An important class of factors are domain or task
  features like
• the complexity of the environment
• the difficulty of achieving a given task
• the resources available for pursuing the task.
• Experiments that vary these factors reveal how
  the intelligent systems behavior depends on
  them.
• Synthetic domains let one alter such variables
  systematically, but it is crucial that they be
  similar to natural domains.

8
Independent Variables System Characteristics

• Another important class of variables involves
  system features.
• Varying these factors leads to different types of
  experiments
• parametric studies (altering system
  parameters)
• lesion studies (removing a system component)
• replacement studies (replacing one module with
  another).
• Such experiments suggest ways that the
  intelligent systems behavior depends on its
  parameters and components.
• Studies that vary two or more factors can reveal
  interactions among them.

9
Independent Variables System Knowledge

• A third class of factors concerns the knowledge
  and experience of the intelligent system.
• One can adapt lesion and replacement studies to
  examine
• the presence or absence of types of knowledge
• the amount of knowledge about a given subject
  
• the amount of experience with a class of tasks.
• Such experiments let one plot behavioral measures
  as a function of knowledge and experience
  (learning curves).
• They also let one compute higher-order measures
  such as rate of improvement and asymptotic
  performance.

10
Repositories for Cognitive Systems

• Public repositories are now common among the AI
  subfields, and they offer clear advantages for
  research by
• providing fast and cheap materials for
  experiments
• supporting replication and standards for
  comparison.
• However, they can also produce undesirable side
  effects by
• focusing attention on a narrow class of
  problems
• encouraging a bake-off mentality among
  researchers.
• To support research on cognitive systems, we need
  testbeds and environments designed to evaluate
  general intelligence.

11
Desirable Characteristics of Testbeds

• Testbeds that are designed to support research on
  integrated cognitive systems should
• include a variety of domains to ensure
  generality
• be well documented and simple for researchers
  to use
• have standard formats to ease interface with
  systems.
• However, these features are already present in
  many existing repositories, and more work is
  necessary.

12
Desirable Characteristics of Testbeds

• In addition, testbeds for integrated cognitive
  systems should
• contain not data sets but task environments
• which support agents that exist over time
• at least some of which involve physical domains
• provide an infrastructure to ease
  experimentation with
• external databases (e.g., geographic information
  systems)
• controlled capture, replay, and restart of
  scenarios
• methods for recording performance measures
• Also, environments should have little or no
  dependence on sensory processing.

13
Physical vs. Simulated Environments

• For domains that involve external settings, one
  can either a physical or a simulated environment
  for evaluation.
• Simulated environments have many advantages,
  including
• ability to vary domain parameters and physical
  layout
• ease of recording traces of behavior and
  cognitive state.
• One can make simulated environments more
  realistic by
• using simulators that support kinematics and
  dynamics
• including data from real sensors in analogous
  locations.
• This approach combines the relevance of physical
  testbeds with the affordability of synthetic
  ones.

14
Some Promising Domains

• A number of domains hold promise for the
  experimental study of integrated cognitive
  systems
• urban search and rescue (Balakirsky Messina,
  2002)
• flying aircraft on military missions (Jones et
  al., 1999)
• driving a vehicle in a city (Choi et al.,
  2004)
• playing strategy games (Aha Molineaux, 2004)
  
• general game playing (Genesereth, 2004).
• Each requires the integration of cognition,
  perception, and action in a complex, dynamical
  setting.

15
Goals of Scientific Experimentation

• Science aims not to show that one method is
  better than another, but to understand the
  reasons for complex behavior.
• This goal can best be achieved through
  experimental studies that
• ask clear questions or test specific hypotheses
• examine relations between behavior and
  independent factors
• move beyond descriptions to explanations of
  phenomena
• Good experiments provide insight into the reasons
  that underlie system behavior.
• Also, whether or not they support an hypothesis,
  they do not end the story, but rather suggest
  ideas for further studies.

16
Concluding Remarks
In this talk, we considered the experimental
study of integrated cognitive systems, including

• challenges posed by their distinctive
  characteristics
• dependent measures that describe their
  behavior
• independent variables that influence this
  behavior
• the need for environments and testbeds that
• exercise the full capabilities of integrated
  agents
• evaluate their behavior at the system level
• support studies of interactions among components.

Taking these into account will transform the
study of integrated cognitive systems into a
well-balanced experimental science.
17
End of Presentation