Lecture 13 Bounded Rationality & Heuristics

Bounded Rationality (BR)
- Introduced in Models of Man by Herbert A. Simon (1957) #economist
- Rationality limited by Cognitive capacity, available information and time constraints (Limitations)
- Under these limitations, rational individuals will select a decision that is satisfactory rather than optimal ^feefcf
- Decision-makers = satisficers with everything they have at the moment (than optimal solution)
- $\implies$ Humans do not undergo full cost benefit analysis. Instead anything that fulfills their adequacy criteria
- Humans $\approx$ Rational entities (as in RCT and DPA model)
- BR addresses this discrepancy.
- Revise notions of perfect rationality. Perfect rational decisions not possible because natural decision problems are intractable and computing resources are finite.
BR and Satisficing
- Satisficing: consider available options until meeting one the meets/exceeds a predefined threshold (aspiration level) for a minimally acceptable outcome
- Simon on Study of BR: > The behavior of human beings who satisfice because they have not the wits to maximize.
Concept of Satisficing
- Decision-making under certainty in a nutshell
  - Real world
  - Avoid Catastrophic Consequences
  - Heuristics
  - Satisfy & Suffice
- As our efforts increase, values associated with that effort also increase. The perfect solution seems to be impossible. Our aspiration level is the dotted box. Perfectionists are highly ambitious (high aspiration levels) e.g. Sai Sanjay S
- Applications: in machine learning (a good enough solution, e.g. high-dimensional optimization problems)
- We need to specify what's "good enough".
- ~~Optimization objective from EUT of seeking an undominated outcome~~ Picking out an option that meets your aspirations.
BR and BE
- Self-inflicted problems $\Leftarrow$ Capacity of the human brain $\lt$ Capacity needed to solve the problem (in the best way, without creating another)
- Three Cognitive limitations:¹
  1. Brain's info-processing capacity
  2. Working memory capacity
  3. Imaginative capacity
Brain's Info Processing Speed
- Constraint on filtering out irrelevant stimuli
- Focus on pertinent information
- Sight reading a complex piece of sheet music
- Complex choices in a hurry... regret later $\to$ Confound our attempts to meet particular ends
- Cognitive exhaustion increase the risk of these errors
Capacity of Working memory
- Long-term memory might be prodigious... working memory is severely limited
- Miller's Rule: "People can only keep around $7\pm 2$ things in mind when working on a problem" (Famous argument by George Miller #economist )
- Prone to
  - Err in complex mental calculations
  - Forget some options if range of choices is large
  - For those we do remember, we forget some of their features
  - Forget things we intended to refer to ("I will have to take care of this later...") when considering what to do
Imaginative capacity
- This issue is not given much attention in literature... but is a key theme in the writings of George Shackle #economist
- An advantage of this limitation: Don't worry about things that should not be troubling us. "Don't worry about a tiger being in our bathroom"
- But, failures of imagination $\implies$ failing to check out possibilities. Or basing our expectations on assumptions that lead us to be overoptimistic or unduly pessimistic
- $\implies$ We get surprises. "I didn't even dream about such a situation"
Imagination & Innovation
- Even bounded imagination has enough creative power to make innovations
- Innovators focus on dreaming up new products, rather than merely competing (cut costs)
- Problematic for buyers and sellers alike
- Schumpeter #economist (1943): Fruits of innovation is short-lived as firm's success push its rivals to engage in retaliatory innovation.
- $\implies$ Market lifecycle < Operating life of product
  - By the time buyers return to the market, something new came up. Too many new products flooding the market
  - Comparison becomes difficult: They don't have enough time to choose as the menus are being constantly revised. How to get time to experiment or learn from others' experiences?
Role of Heuristics
- One of the Two ways to circumvent cognitive limitations
- Use rules or heuristics: reduce cognitive demand of DM
- Some of them heuristics: hardwired products of human evolution
e.g. Search Heuristics
- Process of acquiring info, such that, benefits > costs of this process.
- $A$ wants to buy breakfast cereal
  - Money left ($x$)
  - Taste quality ($TQ$)
  - Health quality ($HQ$)
- Utility: $u(x, TQ, HQ) = 20\sqrt{ x } +2TQ + HQ$
- Say she has $x = \$100$

Product/choice	Price	TQ	HQ	Wealth	Utilty
No cereal	0	0	0	100	0
Budget	1	1	1	99	202
Nutty	3	2	2	97	203
Honey	4	3	2	96	204
Superior	6	3	3	94	203

Five Alternative Search Heuristics
1. Try everything
  - Try different cereal every week, stick with the one she liked the most. But this can be very costly
  - Suppose she tries Honey in the first week, and tries the rest also... then it is just a waste of time. But she wouldn't know it beforehand.
  - Costly in terms of forgone utility (of consuming honey in the later weeks, by consuming other cereals instead)
  - Also costly in terms of time and money
2. Satisficing
  - Set some aspiration level. "What am I looking for?" Find something within that level. A good enough choice.
  - e.g. something that tastes good and is reasonably healthy
  - Keep trying cereals until she tries Honey or Superior
  - Avoid the cost of excessive searching.
3. Directed cognition
  - Treat each chance to gather info, as if it is the last such chance before they have to make a choice.
  - e.g. suppose she knows only about Nutty, she asks herself: "Shall I try one alternative cereal, and, if so, which one?"
  - Narrowing down... Direct your cognition to one particular variety and just explore around it
4. Elimination by aspects
  - Sequentially eliminate choices that fall below some aspiration level
  - Say $A$'s aspiration:
    - medium taste and health quality
    - under $5.
  - Then she will eliminate Superior based on budget and Budget based on quality. The choice is between Honey and Nutty.
  - So, we compare across aspects such as price rather than across choices such as Nutty or Honey.
  - Conceptually simple, because aspects are easy to order.
  - Problem: assumption that the person has info about all possible choices.
5. A person decides how much time to spend on search ("I am in a hurry, I will choose the best I can find in 5 minutes, not more") and searches for taht long.

Though they do not always result in poor choices even in challenging choice environments ↩