accounting for cognitive and perceptual biases in computational models of human movement

41
Accounting for cognitive and perceptual biases in computational models of human movement Thomas J. Pingel Department of Geography University of California, Santa Barbara February 10, 2012 – Department of Geosciences, University of Arkansas

Upload: rey

Post on 24-Feb-2016

36 views

Category:

Documents


0 download

DESCRIPTION

Accounting for cognitive and perceptual biases in computational models of human movement. Thomas J. Pingel Department of Geography University of California, Santa Barbara. February 10, 2012 – Department of Geosciences, University of Arkansas. My Focus. - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Accounting for cognitive and perceptual biases in computational models of human movement

Accounting for cognitive and perceptual biases in computational models of

human movement

Thomas J. PingelDepartment of Geography

University of California, Santa Barbara

February 10, 2012 – Department of Geosciences, University of Arkansas

Page 2: Accounting for cognitive and perceptual biases in computational models of human movement

My Focus

Science that helps us understand and visualizations that help support

spatial decision making.

Page 3: Accounting for cognitive and perceptual biases in computational models of human movement

This includes elements of:

• Spatial Strategy– Navigation– Geospatial reasoning– Cognitive science

• Methodology– Open source

programming– Nonparametric

statistics– Visual analytics

• Geovisualization– Automation

• LIDAR and feature extraction

• TINs and other terrain-related data structures

– Immersion– Dynamism &

Interaction

Page 4: Accounting for cognitive and perceptual biases in computational models of human movement

Cognitive and Perceptual Biases & Patterns of Movement

• How do human biases in slope perception affect patterns of movement in mountainous terrain?

• What role do vision and spatial extent play in determining how humans search an area?

• What role do strategic disposition and attitudes toward risk play in causing asymmetric patterns of pedestrian movement?

Page 5: Accounting for cognitive and perceptual biases in computational models of human movement

The spatial pattern of movement for humans in mountainous areas is different than it is in less rugged terrain.

This is because travel on slopes is slower, requires more effort, and is generally more dangerous than walking on level ground.

Page 6: Accounting for cognitive and perceptual biases in computational models of human movement

The impact of a slope depends somewhat on the direction of travel.

For example, pedestrian travel on slopes is asymmetric with respect to time.

Tobler’s Hiking Function

kph = 6 { exp -3.5 * abs (dh/dx + 0.05) }

Page 7: Accounting for cognitive and perceptual biases in computational models of human movement

Incidentally, it is probably time to update the hiking function using GPS measurements.

Page 8: Accounting for cognitive and perceptual biases in computational models of human movement

Humans greatly overestimate geographic slope.

Following data presented in Proffitt et al. (1995).

Slope / Cost Translation• Uphill / Downhill• Energy, fatigue, safety• Abilities, limitations of agents• Anticipated vs. actual costs

Page 9: Accounting for cognitive and perceptual biases in computational models of human movement

Decisions can be modeled by shortest-path algorithms operating on cost-transformed slope values.

A slope raster is often the basis for the cost surface.

Page 10: Accounting for cognitive and perceptual biases in computational models of human movement

It is important to distinguish between linear and areal slope.

I think linear slope is more important, though both matter.

Page 11: Accounting for cognitive and perceptual biases in computational models of human movement

Observations of existing tracks and trails provide frequency distributions of selection patterns.

Routes may be asymmetric but paths are not.

Page 12: Accounting for cognitive and perceptual biases in computational models of human movement

The frequency distributions are transformed to cost functions.

The cost functions are applied to the linear slope values.

Page 13: Accounting for cognitive and perceptual biases in computational models of human movement

The type of cost transformation function greatly affects the location of the selected path.

Page 14: Accounting for cognitive and perceptual biases in computational models of human movement

The method is adaptable to different agents and constraints, and seems to perform well at a variety of scales.

Model of Rory Stewart’s walk in Places in Between.

Page 15: Accounting for cognitive and perceptual biases in computational models of human movement

I am also very interested in digital navigation systems.

• How do people choose routes?• Preferences• Heuristics• Strategies

• How can we help them find those routes?• Profile development• Expressed• Revealed

• Algorithms• Visualizations & Interactions

Page 16: Accounting for cognitive and perceptual biases in computational models of human movement

If existing methods generate least cost paths (in terms of time or distance)

why would we want to modify that?

1. Purely “economic” concerns don’t capture the full range of criteria that go into route selection.

2. These other criteria are important for overall utility.

Page 17: Accounting for cognitive and perceptual biases in computational models of human movement

Preference rankings depend on mode of travel.

Page 18: Accounting for cognitive and perceptual biases in computational models of human movement

Characteristics of the traveler - in this example, sex – matter as well.

Page 19: Accounting for cognitive and perceptual biases in computational models of human movement

A clearer concept of strategy could be helpful in helping us ask the right questions and to interpret the data.

Strategy has many meanings in the literature.

• Style – Route vs. Orientation– (or Landmark vs. Survey)– (Lawton, 1994)

• Explicit Techniques– Look-back strategy, edge following – (Cornell, Heth & Rowat, 1992)

• Reliance on external aids– Maps or knowledge – (Hutchins, 1995; Ishikawa et al., 2008)– Digital vs. analog

• Task-related– Search vs. Access – (Passini, 1992)

Page 20: Accounting for cognitive and perceptual biases in computational models of human movement

I think the idea of a “strategic disposition” is useful.

Strategic disposition reflects the degree to which an individual cares about reasoning strategically about

a problem, without necessarily suggesting what strategy that individual might use.

It is important to distinguish between strategy, strategic disposition, and performance.

Page 21: Accounting for cognitive and perceptual biases in computational models of human movement

How can we measure strategic disposition?

• Self-report is crucial– Not directly observable

• Instrument– 40 initial items– 101 participants– Factor analysis

• 10 Items• Affinity, frequency,

latency, externalizability, & conditional thinking

Strategic Disposition Index1) I enjoy playing games that involve a great deal of

strategy.2) I am not very good at finding the shortest or

quickest route to a place.3) When driving, I consciously try to find the best

route for the circumstances.4) I don't often play games that involve a great deal of

strategy.5) When driving, I don't typically think about my

route.6) I enjoy activities that involve strategic thinking.7) When walking, I don't consciously try to find the

best route for the circumstances.8) I am not very good at explaining the strategies I

use.9) I often think about route planning in a way that I

would characterize as strategic.10)When parallel parking on a street, I am careful to

park so that as many vehicles as possible can fit in a given block of spaces.

Page 22: Accounting for cognitive and perceptual biases in computational models of human movement

The distribution of SDI is fairly normal.

Page 23: Accounting for cognitive and perceptual biases in computational models of human movement

Men tend to a report a higher SDI than women.

Page 24: Accounting for cognitive and perceptual biases in computational models of human movement

Search Strategy

• Idea from a series of articles– Tellevik, 1992– Hill et al.,1993– Gaunet & Thinus-Blanc, 1996

• Blindfolded or blind subjects search for objects in a small room

• Search Strategies– Perimeter– Gridline (aka Parallel)

• Memorization Strategies– Object-to-wall– Object-to-home– Object-to-object

(Hill et al., 1993)

Page 25: Accounting for cognitive and perceptual biases in computational models of human movement

“Finding the Invisible Animals”

• Audio cues• Movement tracking• Assessment• Object learning• Route efficiency

Page 26: Accounting for cognitive and perceptual biases in computational models of human movement

We reason about space differently, depending on the scale (or extent) in question.

Source: Montello, D. R. (1993). Scale and Multiple Psychologies of Space. In: Frank and Campari (Eds.), Spatial Information Theory for GIS.

• Figural• Projectively smaller than the body• Pictoral and Object subspaces

• Vista• Larger than the body• Can be apprehended without locomotion

• Environmental• Requires locomotion• Learned from direct experience

• Geographical• Learned through symbolic representations

Page 27: Accounting for cognitive and perceptual biases in computational models of human movement

Incidentally, we also think about inside spaces and outside spaces differently.

Creating immersive visualizations that support better inside/outside reasoning is one goal of my

current project.

Page 28: Accounting for cognitive and perceptual biases in computational models of human movement
Page 29: Accounting for cognitive and perceptual biases in computational models of human movement
Page 30: Accounting for cognitive and perceptual biases in computational models of human movement

Gridline searches, localizations, and object-to-object visits

Page 31: Accounting for cognitive and perceptual biases in computational models of human movement

There was a strong spatial signature, supporting the idea of decreasing marginal costs.

Page 32: Accounting for cognitive and perceptual biases in computational models of human movement

There was no association between gridline search propensity and strategic disposition index.

Perhaps there is a difference between being systematic and being strategic.

There is evidence to suggest that strategists differ substantially on the level of detail in their plans.

Page 33: Accounting for cognitive and perceptual biases in computational models of human movement

Attitude toward risk represents another kind of “meta-strategy” that influences wayfinding.

Some people like to play it safe, willing to sacrifice a worse mean result for a lower variability.

Page 34: Accounting for cognitive and perceptual biases in computational models of human movement

Kahneman and Tversky explored this in the context of Prospect Theory.

Would you take:

(a) a sure $20 or

(b) a 50-50 chance at $40 or nothing

Page 35: Accounting for cognitive and perceptual biases in computational models of human movement

Wayfinding questions are somewhat trickier to frame.

Mode of travel, mean, variance, and order effects are all important considerations.

People who gamble in walking contexts are also likely to gamble in driving contexts, but the expression varies.

Page 36: Accounting for cognitive and perceptual biases in computational models of human movement

Let’s use the concepts to shed some light on route asymmetry.

Route asymmetry happens when people take a different route when traveling from A to B than

they do when traveling from B to A.

The same criteria lead to different expressions; local differences and perspective produce different

spatial patterns of movement.

Page 37: Accounting for cognitive and perceptual biases in computational models of human movement

Route Asymmetry Study Design

• Seven legs between four waypoints• Random order according to several

criteria– Flagpole / Psychology excluded– Five unique connections (Routes)

• Position tracked with GPS• Only immediate destination known

– Subjects radioed for the next destination

• Each walk took about 25 minutes• n = 65

Page 38: Accounting for cognitive and perceptual biases in computational models of human movement

Measuring Asymmetry

• Binary (Same / Different)• Gate Coding

– Major pathways & obstacles

– Common sequence length• CHLQ, AFKP, etc.

• Some gates (and Routes) showed more asymmetry than others– Environmental influence– Usually on a subset of gates– High friction sites – Creation or strengthening of

secondary channels

Page 39: Accounting for cognitive and perceptual biases in computational models of human movement
Page 40: Accounting for cognitive and perceptual biases in computational models of human movement

Individual Differences• Risk-takers move through

high-friction sites.– Fast potentially relevant– But not “simple”

• Symmetry connected to– SBSOD– Strategist– Lawton’s Orientation

Strategy• But not

– Risk-taking– Fast / Simple preferences

Page 41: Accounting for cognitive and perceptual biases in computational models of human movement

Summary

We can create better models of human movement if we understand their cognitive and perceptual biases.

1. Humans select routes based on more than time, distance, and effort.

2. Humans make systematic errors in the assessment of these, anyway.

3. More fundamental aspects like strategic disposition and attitudes toward risk may ultimately be more useful as classifiers.