This document discusses cognitive systems and distributed cognition. It provides examples of cognitive systems including the human brain, computer systems, software programs, and social groups. It also discusses how cognition is distributed across internal mental representations, external environmental representations, and social interactions between people and technologies.
10. Sensory Memory
Short Term Memory
Long Term Memory
Sensory memory is like a scratch pad (for
vision), or a bit of recording tape (for hearing),
that can be overwritten with new things. It is re-
written every few seconds by new perceptions.
Short Term Memory is a
temporary store. Some
things from it end up
becoming long term
memories.
Probably stored here forever, though we might
have trouble with retrieval.
10
12. We don’t know much
about where declarative
long term memories are
stored.
The hippocampus
appears to be important
for transforming short
term into long term
memories.
Procedural memory
appears to happen in the
cerebellum, the basal
ganglia, and the motor
cortex.
12
Image courtesy of HowStuffWorks.com
13. Computer Memory
• Disk
• Flash Memory
But who cares? This
description is not at the
cognitive level.
13
Software Memory
• Hash tables
• Schemata
• Frames
• Scripts
• Logical sentences
• Bitmaps
• Activation patterns
• Weights in links
To a cognitive scientist, this
is interesting stuff.
16. Learning is changing memory with the
purpose of preparing a mind for better
action in the future.
We’ll go over several kinds of learning.
16
17. Diminution of a behavioural response with
repeated stimulation.
E.g., the first time you hear a loud noise,
you might jump, but afterward you stop
jumping.
17
18. The more someone pokes you, the more it
bugs you.
At first you get very hungry when you skip
lunch, but after a few weeks you aren’t.
After living in the old house for a while, you
learn where to avoid stepping so that the
floorboards don’t creak.
After tasting lots of Indian food, you come
to expect it to be spicy.
18
19. When a behavioural response is amplified
by repeated exposure to a stimulus.
E.g., at first you can barely feel a vibrating
phone in your pocket, but eventually you
become very sensitive to it.
19
20. Learning to associate two previously
unrelated stimuli.
Typically, this means that you learn to
behave similarly to stimulus B as you do to
stimulus A.
E.g., the dog learns that when you pick up
the leash, a walk is soon to follow. Also,
Pavlov’s salivating dogs.
20
21. A child learns not to touch the stove
because it burns her.
After hearing the same joke a few times,
you stop laughing at it.
You learn that the capitol of France is Paris
in a textbook.
You eat a new food and then get sick
because of the flu. However, you develop
a dislike for the food and feel nauseated
whenever you smell it.
21
22. Positive reinforcement (the strongest)
• Someone smiles at you when you hold the door for
him or her.
Negative reinforcement
• A baby screams until you give them candy. Then she
stops. (Adult learning NR, baby learning PR)
Positive punishment
• You get burned by touching the hood of a car in the
sun.
Negative punishment
• Your parents cut off your allowance because you lied.
22
23. Making a behaviour more or less likely to happen in
the presence of a stimulus depending on reward,
punishment, or taking away a reward or aversive
stimulus.
Mnemonic:
Positive = Present,
Negative = Remove,
Reinforce = Make more likely,
Punish = Make less likely.
23
Adding feedback Removing feedback
Encouraging behaviour Positive reinforcement Negative reinforcement
Diminishing behaviour Punishment
(positive punishment)
Extinction
(negative punishment)
24. Practice involves doing something over and
over and learning how to do it better. It uses
reinforcement and punishment to hone the
skill.
Play is theorized to be a form of practice for
future events (chasing, fighting, caretaking,
manipulating objects, etc.)
Motor skills (that is, moving your body) gets
easier because of automatization.
• It becomes “instinct,” fast, unconscious, and
automatic. For most of you, tying your shoes is
automatized.
24
25. A time-sensitive learning in an animal that is insensitive to
behavioural outcomes.
E.g., a goose learning who its mother is happens 13-16 hours
after hatching.
25
26. Learning that happens by observing
another individual do something.
E.g., copying Michael Jackson dance
moves on YouTube.
Some have tried to get robots to do this.
Much of cultural learning (called
“enculturation”) is observational, but some
is explicitly taught.
26
27. When someone tells you something.
E.g., how to start a web browser, or that
orcas are mammals.
It can be read or heard aloud.
We also get facts by figuring them out, but
this is better described as “reasoning” or
“inference” than “learning.”
27
28. Genetic
• When environments change very slowly
Cultural learning (imitation)
• When environments change relatively quickly
• Content Bias
Imitate the best idea
• Prestige bias
Imitate the most successful
• Conformist bias
Imitate the most common ways of doing things
Individual learning (figuring out for yourself, perhaps
influenced by the Baldwin effect)
• When environments change very rapidly, there is reduced cultural
transmission
Of course, cultures are subject to natural selection as
well. Evolution is substrate-neutral.
28
Editor's Notes
The Space Shuttle Atlantis, backdropped against clouds over Earth, is pictured after it undocked from the International Picture: Space Station at 7:50 a.m. CDT, Sept. 17, 2006. The STS-115 astronauts completed six days, two hours and two minutes of joint operations with the station crew. Image from Wikimedia Commons.
We’ll mostly be talking about human beings in this class, but for many of the things we will talk about the issues will be the same for many animals.
Eye picture: By Woodwalker, with a retouche by Poxnar [CC-BY-SA-3.0 (www.creativecommons.org/licenses/by-sa/3.0)], via Wikimedia Commons
Camera picture: By Justin Smelley [Public domain], via Wikimedia Commons
We can describe memory, as many things in cognition, at different levels of explanation.
In animals, including humans, we can describe memory’s location, biologically (e.g., in the hippocampus).
We can also describe if functionally (e.g., long-term and short-term).
Image: http://en.wikibooks.org/wiki/File:Overview_Memory.jpeg
Working Memory is Baddeley’s refined theory of short term memory.
The working memory for facts is commonly thought to have 5-9 slots, one for each “chunk.”
A chunk is a bit of working memory. The letters M, L, K, J, F, K each might be a chunk, but if you noticed that they were initials (MLK, JFK) then you could store them as two chunks.
Some research suggests that there are more like four slots in WM.
Baddeley thinks there are separate working memories for audition (hearing) and vision.
Some people think that WM is like what we said short term memory—that is, a temporary buffer that holds it before it goes into LTM. This is the gateway model.
The workspace model thinks that WM is a separate buffer from LTM. Things get shuffled in and out of it as the task demands. That is, it holds copies of information in LTM.
The unitary model holds that WM is simply the activated parts of LTM.
Image courtesy of HowStuffWorks.com
(permission granted in June 7, 2011 email.
For memory, most cognitive science is interested in the structure of knowledge at the cognitive level.
It’s the level above the implementation level (e.g., neurons, or disks).
Pictured is famous scientist Konrad Lorenz, leading around geese who learned through imprinting who their mother was at an early age.
Atkeson tried to get a robot to learn to balance an object on a pole by watching a person do it. In 1997, anyway, it was really hard.
Atkeson, C. G. and S. Schaal (1997). Robot Learning From Demonstration, Machine Learning: Proceedings of the Fourteenth
International Conference (ICML '97), Edited by Douglas H. Fisher, Jr. pp. 12-20, Morgan Kaufmann, San Francisco, CA, 1997.
For more information on prestige bias and conformist bias, see Richerson and Boyd 2008, esp. kindle location 2187