- reaction to:
- idea that people should be looking at everything in order to make fully rational decision
- idea that heuristics are bad
- Bounded rationality: rationality is bounded by limited time, limited resources, and limited cognitive capacity
- as vs. infinite rationality
- people will look at a problem until they arrive as a solution that is “good enough”
- Tversky and Kahneman, “heuristics & biases: heuristics are reasoning shortcuts that generate errors
- Gigerenzer: argues that heuristic reasoning is useful and effective
- Fast and frugal heuristics: (as opposed to Cues + Probability)
- fast: does not involve much computation
- frugal: does not involve all available information
- want to capture how humans make decisions in real world situation: under time, resource, and knowledge constraints
- Supernatural / “Demons” Reasoning – reasoning based on everything you would possibly need to consider
- assumes astors with unlimited time, resources, info
- within contraints, still unbounded
- Bayesian reasoning:
- also an unbounded rationality model
- stops when actor has arrived at an appropriate decision
- requires the actor to continually analyze next step to determine whether a stop is appropriate, which itself requires infinite time
- Satisficing”: problem-solving ceases when a solution is generated which satisfies prior criteria for a “good enough” solution
- neither people nor machines can deal with complexity of the real world in the way it is manifested in optimal problem solving; it must be reduced to approximate a solution
- given the limitations of human mind and the complexity of its surroundings, the question is not “if” have to make compromises; any solution will be sub-optimal. The question is now what part of the sub-optimal solution space we choose to deal with.
- fast and frugal model: a “satisficing” method that optimizes correct decision-making
- within approximate solutions there exists a subset that are fast and frugal
- Appropriate decision-making is context-dependent
- if the decision space is constrained, then bounded decision strategies become more effective
- Combining information from different cues requires conversion into a “common currency”, which comes at a time and cognitive cost
- fast and frugal strategies often avoid these conversions, and make decisions on only one or a few factors
Heuristics built around how people “do” make decisions, not how supposedly “should”
“Cognition is the art of focusing on the relevant and deliberately ignoring the rest.”
- Ecological Rationality:
- rationality in a natural context
- social rationality: rationality in decisions between and among people
The visualization community to a degree already understands this: Tufte and data density
- A lot of the flaws in cognition of uncertainty visualizations are flaws in “fast and frugal” reasoning
- e.g., cone of uncertainty, where outside the cone is interpreted as safe, inside the cone as a hit zone
- people made simplified binary model rather than re-creating the intended, more complex model
- 2 options: don’t present it in a way that they can simplify (e.g. spaghetti plots), or come up with fast and frugal visualization that generates the proper mental model
- Individual differences in decision-making: some people choose to invest cognitive effort in more decisions; some people prefer a fast-and-frugal approach to most decisions
- how do we approach this problem from a visualization standpoint?
- some people are more visually fluent than others
- one heuristic: a tendency to transform continuous spaces into binary ones
- e.g. for the cone of uncertainty: if people are going to make a binary decision, a better binary would be “evacuate/don’t evacuate”
A “convertible” model: one that can be interpreted fast-and-frugal, or include more data in the decision-making process
- If we make the visualization complex, will people just abandon it? (Probably…)
- is the goal for uncertainty to force the user to delay decision-making, allowing for greater cognitive interpretation and introduction of new information
- this argues for benefit of uncertainty that is quite different from baysiean utility optimization
- changing decision process rather than just the decision [not so fast, not so frugal]
- Should we be considering this particular type of heuristic problem solving b/c we want to support it, or are the exact problems we are considering ones where these heuristics break down (known exceptions to the “general” rule)
- is the latter an interesting formulation of the question?
- Is there really a spectrum of how people make decisions? Do people think about the space of these models as complex continuum?
- supporting qualitatively different views for different types of user
- user may affect what heuristics “work”
- well chosen defaults with flexibility to change as you want
Tension: the paper is arguing that what we’re doing is not how people think / is hard for them to do
- Does this mean that this is not what we should be doing?