Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Cognitive architectures and what they can provide to psychological applications

a presentation for an applied cognitive psychology course
by

Tuuli Pöllänen

on 25 October 2012

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Cognitive architectures and what they can provide to psychological applications

Cognitive architectures and their potential for applied cognitive psychology Newell, 1973; "You can't play 20 questions with nature and win" Do you know what "20 questions" is?
Would you like to play a game? 1973 W.G. Chase - symposium on visual information processing Newell - commentary on the articles

Great experiments and interesting results, but microscopic problems and no synthesis (not even within the domain and publications of an individual researcher)

Problem: binary distinctions
insufficient constraints of the experimental situation
"divide and conquer"-strategies
no efforts to create complete cognitive models

Desperate situation - where would the science be in 30 years? Newell's master plan Two injunctions on how to approach cognitive experimentation and to save the world: KNOW THE METHOD your subject is using to perform the task.

Behavior is programmable - under the subject's control in achieving his own ends.

Behavioral prediction = known the goals, the task environment, and the structure of the processing mechanism

NEVER AVERAGE OVER METHODS - this leads to "garbage, or even worse: spurious regularity"

In plain English: unrepresentative results. Then what would you have us do. Mr. Newell? Discover the "structure of the subject, which is fixed and invariant, so we can theoretically infer the method it would choose in solving a task"

Knowing the parameters, generate a collection of methods likely to be used, then find out which method was used

no more correctional experiments, no more divide and conquer

Control the entire selection of methods the participants can use, provide a frame, a control structure, which would constrain what methods can be used to perform the same task.

Newell: programmed algebraic structures work best for making natural constraints, since they create a space where different methods can be tested if they conform to the limitations of space and time thus, control structures would be best illustrated in programming languages Newell: how to exit the crisis Newel suggested three paradigms:

1. complete processing models - make complete models of human cognition with psychologically relevant constraints
2. analyze complex tasks - focus on a single complex task instead of multiple simple experiments and make sure each task contributes to a totality
3. one program for many tasks - construct a single system to perform a diverse collection of small experimental tasks.
the birth of cognitive architectures and psychometric artificial intelligence spoiler alert There were already computational models before Newell´s critique (inc. Newell & Simon, 1959, 1963)

Newell had agenda

The article did trigger new lines of research and modeling - product systems

Later, new architectural classes emerged
1990´s: commercial applications cognitive models and cognitive architectures cognitive model = model of human cognition
computational cognitive model = more or less automated computer program, modeling human cognition and most likely based on a cognitive architecture
enable us to:
inspect the agent's behavior
predict behavior and its outcomes
observe the process
cognitive architecture = "blueprint" system for generating a cognitive model of some agent considered intelligent under some definition
Byrne: AI to simulate "human performance, in a human-like way"
compare to engineering-approach to AI - "by any means necessary" A broad theory of human cognition, based on extensive empirical data, transformed into a running computer program Anatomy of a CA-based cognitive model One part provided by the generic CA, one part supplied by the analyst to make a specific model.
Different models, same architecture -> similar features (architecture provides constraints)
Different cognitive architectures have different features
create systems with different characteristics
e.g. focus on creating functional behavior, or replicating experimental data; different sensory modules; focus on a different cognitive process...

New: embodied models! Add "sensory organs" -> direct interaction with the environment, enable direct observation

Especially if the system has metacognitive processes and
machine learning capabilities! attempt to model the entire outlying infrastructure - components, processes, characteristics... properties and capabilities of cognitive architectures modeling aspects that are relatively constant over time, across different application domains
a) short-term and long-term memory (beliefs, goals, knowledge)
b) elements contained in STM and LTM, their organization
c) functional processes operating on the structures: utilization, learning Some capabilities are maintained on an architectural level - like foundations, walls, roof in an apartment
The architecture has programming potential for other capabilities, which can change over time - like furniture
The extent depends on the architecture - appliances and built in book-cases The absolutely slide of the necessary abilities LEARNING HAPPENS IN ALL COGNITIVE ACTIVITIES - direct instructions, problem-space search, behavioral cloning, reinforcement learning, successfully achieved goals.
recognition - situations as reproductions of familiar patterns - static or dynamic
categorization - assigning objects, situations and events to known concepts and categories
recognition and categorization depend on perception
decision making - "recognize-act-cycle": recognize a situation and behave accordingly; also more complex mechanisms of decision-making
decision making demands ability to represent alternative options and choose from them -> "conflict resolution": assigning score to alternatives based on cost and expected outcome
perception - different modalities, low or high tech sensors, results can be integrated to a total picture
attention is included for perception requiring limited resources
prediction and monitoring - requires experience and memory
problem solving and planning
reasoning (deductive, inductive, heuristic, approximate, abductive), "belief maintenance" - infering new beliefs and deciding whether to maintain old ones
execution and action - primitive skills and complex sequences
communication - agents interact with, and obtain knowledge from one another - generation and understanding language requires nearly all aforementioned skills
remembering - storing and retrieval of knowledge, reflection - processing mental structures in working memory, explanation - reflecting on cognitive abilities, meta-reasoning: explaining the generation of cognitive activities
and TRANSFER - using knowledge learned in one task in a different task properties of cognitive architectures These determine their capabilities

Regard (and example):
a) knowledge representation, e.g. coded in one formalism or multiple
b) -II- organization, e.g. hierarchical or flat
c) -II- utilization, e.g. heuristic or pre-made, reactive or deliberate.
d) -II- acquisition, e.g. analytical, empirical, hybrid
Overlap with capabilities, but even more technical, so I will not burden you by listing them So, what are these things good for? models built around the same architecture share characteristics
combine different models to a complete model of human cognition
create libraries of "behavior idioms" for future model building projects CA provides a framework by posing constraints
constrained model allows identification of information needed to perform the given task
differentiate what information is from the environment and what belongs to the agent CAs contain mechanisms relevant to the task at hand and reveal their role in successful task performance
Example: using a cognitive architecture for interface optimization, the CA may reveal:
working memory load
time taken to learn the interface
types and frequency of errors
eye movement pattern when operating the interface (in an embodied model)
exact metacognitive process during operation (metareflective model)
etc etc relationship with applied cognitive psychology Part of the long interplay between cognitive science, psychology and AI
Can be considered an integral part of the field
Support both applied and theoretical goals
applied:
embody vast amounts of empirical data that can be used to create a model for virtually any applicative purpose,
provides a supplement to some non-computational applications (e.g. expert models)
theoretical:
support the creation of complete cognitive models,
enable testing of theoretical constructs,
help in compacting existing models
etc etc Next I was going to present to you a detailed account of a cognitive architecture, but then I thought you'd prefer the world's cutest robot: MIT, Leonardo - social cognition robot
http://robotic.media.mit.edu/projects/robots/leonardo/socialcog/socialcog.html Problems with cognitive architectures Newell (1973) do NOT average over methods -> but you must average somehow if you are to make one architecture based on 100 years of empirical data
Leads to an architecture nearly flat on individual differences
not a problem in creating agents
problem when modeling a phenomenon influenced by ID
What about when you want to model individuals proven to differ from the average population? Jones et al., 1999 and fighter pilots

How to include affective variables?

Does it contain any more information than your expert system already did?
This is is important, computational modeling costs money!

Newell (1973): "stop fiddling with your own little business and cooperate"
AI and cognitive science are fragmented into well-defined sub-disciplines with little interest in collaboration and different evaluation criteria -> make creating complete cognitive models difficult
Practical applications require increasingly integrated intelligent systems!

Benjamin, D. P., Lonsdale, D. & Lyons, D. (2004). ADAPT: A Cognitive Architecture for Robotics. International Conference on Cognitive Modeling, Pittsburgh PA, July 2004.
Duch, W., Oentaryo, R.J. & Pasquier, M. (2008). Cognitive architectures: where do we go from here? Frontiers in Artificial Intelligence and Applications, 171, 122-136
Gärdenfors, P. (2004). Conceptual spaces as a framework for knowledge representation. Mind and Matter (2)2, 9-27.
Jones, R. M., Laird, J. E., Nielsen, P. E., Coulter, K. J., Kenny, P., & Koss, F. V. (1999). Automated intelligent pilots for combat flight simulation. AI Magazine, 20(1), 27-41.
Kieras, D. (2005). A survey of cognitive architectures – pros and cons of existing architecture applications. University of Michigan.
Kieras, D.E. (2007). The control of cognition. In W. Gray(Ed.), Integrated models of cognitive systems. Oxford University Press.
Langley, P., Laird, J.E., Rogers, S. & Sun, R. (2008). Cognitive architectures: research issues and challenges. Cognitive Systems Research, doi:10.1016/j.cogsys.2006.07.004
Magerko, B., Laird, J. E., Assanie, M., Kerfoot, A., & Stokes, D. (2004). AI characters and directors for interactive computer games. In Proceedings of the sixteenth innovative applications of artificial intelligence conference (pp. 877–884). San Jose, CA: AAAI Press.
Meyer, M., & Kieras, D. (1997). A computational theory of executive control processes and human multiple-task performance. Part 1: Basic mechanisms. Psychological Review, 104, 3–65.
Newell, A. (1973). You can't play 20 questions with nature and win: Projective comments on the papers of this symposium. In W. G. Chase (Ed.), Visual information processing (pp.283-308). New York: Academic Press
Ritter, F. E. & Young, M. R. (2001). Embodied models as simulated users: introduction to the special issue on using cognitive models to improve interface design. International Journal of Human-Computer Studies, 55, 1-14
Sammut, C. (1996). Automatic construction of reactive control systems using symbolic machine learning. Knowledge Engineering Review, 11, 27–42.
Sun, R., Merrill, E., & Peterson, T. (2001). From implicit skills to explicit knowledge: A bottom–up model of skill learning. Cognitive Science, 25, 203–244.
Sutton, R. S., & Barto, A. G. (1998). Reinforcement learning: An introduction. Cambridge, MA: MIT Press.
Van Merrienboer, J. J. G., Paas, F.G.W.C, & Sweller, J. (1998). Cognitive architecture and Instructional Design. Educational psychology review, 10(3), 251-296
Promote good experimental methodology! Think critically: quite the leap from poor experimental design to only using computatioal models
Newell was driving his own agenda!
Full transcript