Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
22/11/16
1
1
Lecture 3: The intelligent use of space
729G12
Erik Prytz [email protected]
www.ida.liu.se/~eripr77/
2
Today’s lecture at a glance
• Examples of situated cogniBon ArBcle 1: The intelligent use of space
ArBcle 2: On disBnguishing epistemic from pragmaBc acBons
22/11/16
2
3
Situated cogniBon vs. ”tradiBonal” cogniBon
• QuesBon: How are we to understand human cogniBon? – According to tradiBonalists:
• By looking at processes in the black mystery box between sensaBon (input) and motor acBon (output)
…5 1… …
”CogniBon”
4
Situated cogniBon vs. ”TradiBonal cogniBon”
• QuesBon: How are we to understand human cogniBon? – According to situated cogniBon perspecBve: • No, we have to look at acBvity and thinking in the world. Input and output are invalid separaBons. • We are always situated in a world and context.
”CogniBon”
1 39
+66
..5
…5 1… …
22/11/16
3
5
Situated cogniBon • SBll focuses on the individual • Thinking with things and arBfacts • Impact on ArBficial Intelligence
– Planning and problem solving
• Get out of the laboratory!
• ImplicaBons for cogniBve science – Joint cogniBve systems – InteracBon design – Contextual psychology
6
THE INTELLIGENT USE OF SPACE David Kirsh
22/11/16
4
7
The intelligent use of space • We are always surrounded by space
– At work – At home – …
• As spaBal creatures, can we take advantage of the resources in the environment to aid thinking?
8
The intelligent use of space • TradiBonal AI and planning • We don’t have a perfect search tree in our head and then execute our
acBons. • We plan as we go, interact with our environment and solving problems
using things to think with. – Memory cues – Solving mathemaBcal problems on paper – …
22/11/16
5
9
The intelligent use of space • We organize and re-‐organize workplaces to enhance performance – Time – Space – Energy – Memory
• Main point: – By using space around us… – …we lessen demands on the other resources
• Case in point: ExperBse!
10
Intelligent use of space: ExperBse
• TradiBonal view: – Experts are cogniBvely superior in their domain due to memory ability, quicker inferences and knowledge base
22/11/16
6
11
Intelligent use of space: ExperBse
• TradiBonal view: – Experts are cogniBvely superior in their domain due to memory ability, quicker inferences and knowledge base
• AlternaBvely: Experts structure their domain befer than novices through experience and pracBce – Not always consciously so!
• IntenBonal, but not necessarily deliberate.
• We are all experts in our everyday environments!
12
Intelligent use of space • As experts in our everyday environment we… – … use resources locally in our environment instead of thinking and planning analyBcally as we go.
– … setup our environment in a manner that we have local informaBon available. • Through ”jigging” and informa9onally structuring our environment
• So, how does this setup work?
22/11/16
7
13
Intelligent use of space: Jigging
• Environmental factor that decrease variability – Makes the situaBon more ”stabilized” (e.g. cup holder) – Physical or informaBonal jigging
• E.g. door jam constrains ac-ons • Memory cue on a map facilitates informaBon processing
• ”jigging” makes the environment hospitable for relevant problem solving – Reduce visual search – Things easier to noBce, idenBfy or remember – Problem representaBon
14
Intelligent use of space
• Humans exploit spaBal arrangements in the workplace in three ways:
– i) SpaBal arrangements that simplify choice
– ii) SpaBal arrangements that simplify percep9on
– iii) SpaBal dynamics that simplify internal computa9on
22/11/16
8
15
Intelligent use of space -‐ examples
• Using space to simplify choice: – ”What acBons do I have available?”
16
Intelligent use of space -‐ examples • Using space to simplify choice: – ”What acBons do I have available?”
– Affordances! • Features in the environment ”affords” acBon (Gibson/Norman)
• Depends on skills of an individual • Context and culture • Important in interacBon design
22/11/16
9
17
Intelligent use of space -‐ examples • Affordances can… – …reduce perceived acBons by hiding features – …highligt perceived acBons by cueing afenBon
• Case in point: ProducBon lines – StaBons give rise to:
• Limited tools • Limited tasks • Limited acBons • Less variability and less complex problem-‐solving context • Less planning and deliberaBon needed
18
Kitchen design – StaBons & highlights
22/11/16
10
19
Cueing blocked acBons
• Blocking – spaBal arrangement that says ”don’t do X!”
20
Intelligent use of space
• Using space to simplify percepBon: – Facilitates acBon decision by speeding up the process – E.g. tomatoes at both sides of the sink etc. – Symbolic marking
• Puong envelope to be posted by the door • Marking an ”X” on the hand • Puong a finger on your posiBon on a map • Puong a bill to be paid on the laptop
22/11/16
11
21
22
Intelligent use of space
• SpaBal dynamics that simplify internal computaBon: – Make computaBons in the world instead of inside the head
• RotaBng a map
22/11/16
12
23
Intelligent use of space • SpaBal dynamics that simplify internal computaBon: – Make computaBons in the world instad of inside the head
• RotaBng a map
– Using percepBon instead of internal computaBon • Frees up internal resources • Solves problems faster • E.g. pocket calculator
24
Intelligent use of space • Some conclusions: – Human beings create and organize their workspace to…
• …Simplify problem-‐solving… • …reducing the complexity • …offload memory resources…
• ImplicaBons for AI: – Planning – problem solving
• ImplicaBons for psychology – Take space and context into account
22/11/16
13
25
ON DISTINGUISHING EPISTEMIC FROM PRAGMATIC ACTION
Kirsh & Maglio
26
Epistemic and pragmaBc acBons • Main points: – Not every acBon is performed to reach closer to the goal
– A criBque on AI
– Two sets of acBons: • PragmaBc acBons: Lead us closer to the current goal • Epistemic acBons: External acBons that yields knowledge about a situaBon – later payoff
• Division between ”Planning” and ”AcBon” not clear-‐cut
22/11/16
14
27
Epistemic acBons
• Improves cogniBon by… – Reducing memory load – Reducing number of steps involved in mental computaBon
– Reducing probability of error of mental computaBon
28
PragmaBc acBons
• Physical acBons that bring an agent closer to the goal
• Planning = series of transformaBons from iniBal state to goal state – Purely through pragmaBc acBons – Using Bme, distance, or energy as metric
22/11/16
15
29
Tetris
30
Tetris as empirical domain
• Fast game with both perceptual and cogniBve load – Time as cruicial factor will provoke strategy deployment
• Every acBon leads closer or farther to final posiBon – Easy to discriminate acBons
• Fun to play – Easy to get parBcipants – Become expert
22/11/16
16
31
Tetris: Some observaBons
• Moves farther from the goal – gain informaBonal certainty as payoff
• Clearly epistemic rather than pragmaBc acBon
32
Tetris: Some observaBons
• Early rotaBons to get informaBon about idenBty
• Save mental rotaBon effort and Bme -‐ matching
22/11/16
17
33
Epistemic acBons
• AcBon to put one in a befer posiBon for computaBon
• Reduces Bme, space, energy or unreliability – PragmaBc costs are offset by epistemic benefits – We intelligently exploit informaBon without even knowing it!
– Hallmark of experBse – automaBzed procedures.
• Some acBons are both pragmaBc and epistemic
34
Classical informaBon-‐processing model
This model presupposes that every acBon is pragmaBc Creates a separaBon between acBon and cogniBon
22/11/16
18
35
Input vs. Output?
36
Essence
“The point of taking certain ac-ons, therefore, is not for the effect they have on the environment
as much as for the effect they have on the agent” – Kirsh & Maglio (1994)
22/11/16
19
37
Essence
• 007 principle: – Know only as much as you need to know to get the job done!
”Evolved creatures will neither store nor process informa-on in costly ways when they can use the structure of the environment and their opera-ons upon as a convenient stand-‐in for the informa-on processing opera-ons concerned. That is, know only as much as you need to know to get the job done.” -‐-‐(Clark, 1989, p.64)
38
Conclusions and implicaBons • CriBque towards AI
– A new set of acBons
• CriBque towards standard informaBon-‐processing – No clear boundary between input-‐”cogniBon”-‐output
• Calls for situated cogniBon in CogniBve Science – Study the interacBon between agent and environment
• CogniBve coupling between agent and world – We structure our own workplace and world – CogniBve systems – CogniBon extends to the outside world?
22/11/16
20
39
Next week
• Monday: guest lecture by Tom Ziemke – Embodied CogniBon
• Tuesday: lecture by Corinna – Ethnography
• Wednesday: lecture on AcBvity Theory • Thursday: first advisor meeBng for the project – Sign up in project groups (3 people in each)!