Notes: Mike Cole, October 19, 2001

 

Pre-talk discussion

Don Norman: Founder & 1st head of cog sci department

Author, The Psychology of Everyday Things (POET), subsequent editions called The Design of Everyday Things; Things That Make Us Smart

Worked, among other things, on interaction of people & artifacts

1st section of Twelve Issues for Cognitive Science.  Cog Sci 4, 1-32 (1980)

            Predecessors oversimplify (at best) cog sci

Mind as “center of universe”; want to study in isolation

Rather, the mind is not an abstract process, but simply a by-product of the processes that keep us alive.

            People don’t think in isolation; they are influenced by environmental and social factors.  The classroom as an example of this fact.

At this time, Norman is finding out that abstract cognitive systems need to be studied as embedded in these two ways.

He thinks Fig 1. (p. 3) is largely correct, but that it does not represent all of cognition.

(note: ten years later, Norman added 4 more issues to his 12)

The paper is not really radical nowadays.

            People are saying many of the same things now (does this represent a lack of progress?)

Difficulty of building a model of an intelligent being in a social environment?

Brooks is an example of a more radical view than that presented in the paper.   

            Rodney Brooks: MIT AI lab director, best known for his layered/subsumption architecture

                        e.g.: make a robot that can move, add a layer that allows it to avoid stuff.          

It is necessary for both lower and higher layers to be able to intervene in each other’s activity.

Distinctness of each level becomes less clear.

Late ‘80s, early ‘90s his approach was to ignore cognition; just make simple robots that interact.  But this doesn’t scale—way too much programming had to be done, and very little learning was involved.

The idea of intelligence w/o representation

            Compare with perception and Gibson

At the time of Norman’s first article, everyone disagreed with Gibson.  Two extremes.  Cole and Balzano (who was trained in Gibsonian tradition) did common cause, but neither could be heard until the mid ‘80s

Early work on PDP.  Rumelhart, McClelland

Note the writing style of Norman’s paper—it sounds like an expanded version of a talk

Discussion of each of the 12 issues:

Belief systems: Norman is proposing a philosophical enterprise (with a nod to the anthropologists as well?)

Look at the reference list

            No philosophers

            Bernstein, early cyberneticist and “patron saint of Gibsonians”

            Most of them are West coast

West: take embodiment, embeddedness, neuroscience seriously

Contrast w/ East: symbolic modularity, pure cognitive system, higher abstract stuff, e.g. chess playing machines

Norman at one point incorrectly says figuring out body state control would be easy

Consciousness: similar to what he says about belief: “now is too early; we don’t know enough to adequately explore it”

Most of what happened at the start of cog sci was east coast and Sloan money

Development: we need to consider where the nervous system comes from and how it develops.

            Contrast with nativism and its lack of regard for developmental matters

Some of what Gardner says isn’t important (history, concept, culture, emotion), Norman says is.

Emotion: there is an absence of current cog sci research on emotion

Is there a need to consider computational issues in order to make emotion research cognitive?

            If not, this raises the question of what cog sci is.

Comments made on the so-called core of cog sci.  The fields that compose it are differentiated by their methods, whereas cog sci as a whole is not?

One needs a clientele for a program to succeed.  Communication and cog sci will, Cole says, but UCSC Consciousness program will not. [cf. Manovich talk]

It is also important to have a goal.

Learning: “little study and not considered a central part,” Norman says (!)

Norman gives advice on research topics to choose (p. 28):

Select which area to focus on minute topics in, but periodically stop and reconsider what you’re doing [Norman has apparently done this several times]

                        Though this is conventional wisdom, it is often overlooked.

 

The talk

Gardner: Gives an account of cog sci from ca. 1940-1975

Technological advances that set the stage include Turing, von Neumann; McCulloch, Pitts; Weiner; Shannon, Weaver, Miller; Chomsky; failure of stimulus-response behaviorism

Sloan Foundation conferences and funds

SF commissioned a report.  The “cognitive hexagon” described relationships between philosophy, linguistics, neuroscience, AI, psychology, and anthropology

            This report was never published: everybody in the six disciplines disagreed

These divisions of opinion led to the formation of variously constituted cog sci departments, representing different sub-sets of the hexagon.  (note the overall absence of anthropology, however)

Gardner said cog sci’s central achievement (as of 1984, anyway) was “the clear demonstration of the validity of positing a level of mental representation.”

Shortly thereafter, Gardner noted two trends: reconnecting cog psychology with the neurosciences, and connectionist modeling.  (oddly, he doesn’t mention the increasing influence of anthropology, though he commented on it elsewhere)

Other, imperceptible changes were also going on at the time (elaborations of PDP ideas, embodied cognition)

Cole describes his work in anthropology, and mentions the marginalization of much of anthropology during the Sloan years

He came to UCSD in 1978; there was no cog sci department then, just the Center for Human Info Processing (CHIP)

Interdepartmental interaction included a ’78-79 seminar, and Norman’s creation of the cog sci program.

Norman publishes “12 Issues”

Culture figures prominently in issue 1, belief systems

Figure 2: physical systemsà pure cognitive systems (acting upon and affected by regulatory systems)àOutput

Cole was unhappy with the characterization of the model’s basic constituents: Norman also needed to consider the influence of the environmental system, consisting of physical and social parts. 

This would tie in the relevance of evolutionary neuroscience, developmental psychology, cognitive anthropology

Got him to tack on a discussion of “the environmental system and cultural knowledge”

Several years later, Norman took more seriously the role of the culturally organized environment of cognition.

Norman and Hutchins on distributed cognition.

Artifacts: cultural-historical activity theory

According to his view, an artifact is an aspect of the material world that has been modified over the history of its incorporation in goal directed action.

            From this perspective, human thinking is artificial

Herbert Simon: The Sciences of the Artificial

Provided a canonical AI account of the study of human thought within the disciplines of artificial intelligence.

Physical symbol systems: “almost the quintessential artifacts, for adaptivity to the environment is their whole raison d'etre” (1981, p. 27).

Artifactual things: are synthesized by man; may imitate appearances in natural things while lacking their reality; can be characterized in terms of functions, goals, adaptation; are often discussed in terms of imperatives as well as descriptives

Norman’s “Cognitive artifacts” 1990

Goal was to “emphasize the information processing role played by physical artifacts upon the cognition of the individual…to integrate artifacts into the existing theory of human cognition” (p. 2)

His approach differs from Cole’s: artifacts as somehow external to human thought, and no intrinsic relationship is given between thought and artifact.

Hutchins’ approach: functional system, adapted from Luria.  Artifacts do not simply act on the individual, they are constitutive of the information processing mechanisms.

“Incestuous connections” between cognition, cognitive artifacts, tasks, and the broader socio-cultural environment

Designing a reading activity

Reading viewed as a top-down and bottom-up process

Early models of lexical rec. represent top-down processes as a “trident form the hand of Zeus”

Question-Asking-Reading

5th dimension - Work with children classified as learning disabled.  Establish a procedure to help them around disabilities.

To be coordinated systems of mediation

Create a route that allows kids to interpret the world through print

How to bootstrap?

Q-A-R
Scripted activity: goal talk, roles, silent reading, fulfill roles, give quiz, discuss quiz

Program is good both for diagnosis and learning to read.

Designing complex educational activities

            Creating an after-school activity that promotes learning and development

            Special design features:

Use computers as mediators of interaction

Involve undergrads as key personnel

Combine play, learning and affiliation

Make activity voluntary

Got good results

Basic UC Links Model

Creating a shared educational activity

Layering structurally equivalent programs

Every single 5th dimension is a culture

 

Question and Answer:

What about artifactual things being synthesized by robots?

            Open question; depends on whether you can get self-modifying robots

What have we learned in cog sci that is useful for teaching kids?

Not just one way to skin a cat; learning is not just bottom up; we know certain principles (his study was to show that such principles can be instantiated across various systems)

Why won’t the NSF fund research like his?

The experiment is hard to control because participation is necessarily voluntary; it is sloppy, but within certain conventions.