Lecture 11 Algorithms

Question 1

algorithm

Answer 1

any evidence based forecasting formula or rule. thus, the term includes statistical models, decision rules, and all other mechanical procedures that can be used for forecasting.

important is execution by both humans and computers

Dietvorst et al., 2015

Question 2

netflix example simplified algorithm

Answer 2

1. viewers are clustered in taste groups
2. shows are coded based on content
3. the algorithm predicts how much a certain group wil enjoy different content

Question 3

why do we need algorithms

Answer 3

• people can be irrational and make mistakes
• people don’t know what they don’t know -> Dunning-Kruger effect
• experts might also have incentives to be entertaining, not only accurate

Question 4

Dunning-Kruger effect

Answer 4

people with limited competence in a certain domain overestimate their skills

Question 5

Castelo, Bos & Lehman (2019)

Answer 5

trust for algorithms
- reduced for consequential tasks
- increases when consumers are more familiar with the use for certain tasks
- increases for tasks that seem objective

facebook ads on dating or finance with either human or ai.

Question 6

Obermeyer, Powers, Vogelu & Mulllainatham (2019) on algorithmic bias

Answer 6

found that black patients had to display more chronic conditions to be eligible to end up in high-rick care management programs.

Question 7

Algorithms in special circumstances (VAM)

Answer 7

value adding model is used to evaluate teachers in the US. but has some issues
- changes in students’ circumstances
- teachers of top students recieve worse scores
- pushes teachers to help students cheat

similair programs are used for consumer products or ranking universities

Question 8

downsides of algorithms

Answer 8

• unintentional discrimination
• unfair in the way they handle special circumstances
• steer consumer demand (LLM)
• “social consequences” to relying on algorithmic judgement

Question 9

steering consumer demands by Rahwan, Soreperra and Werner

Answer 9

had participants learn about two different ebooks with an ai-powered shopping assistant. then got to choose their preferred product and guess wich direction they were steered in
- found that LLM steer preferences, even when people know they are steered.

Question 10

Awad et al., (2018) the moral machine experiment

Answer 10

collected data on what people thought a self-driving car should do in different conditions
- people would spare more characters than fewer
- people were most likely to spare a stroller
- least likely to spare a cat