Cognitive Bias Final Flashcards
CONFIRMATION BIAS
We tend to look for/notice/remember evidence that confirms our opinions. Favoured theories while ignoring/forgetting/explaining away/disortring disconfirming evidence
Heuristics
Short cuts our brain uses to get to conclusions.
Can be very useful twords survival but also can lead our logicl astry.
Ingroup Bias
We tend to place more weight/ Higher Value on the views of/ and members of our immediate group than on those outsiders (can be at the outsiders expense)
We tend to oversitmite the abilities and virtues (and morals) of peers.
A manifestation of our innate
tribalistic tendencies.
NEGLECTING PROBABILITY
This is where we we don’t gage proper sense of peril and risk. We overstaet the risk of relativily
harlmess actvites, when we underate the dangous of norml activtes.
OBSERVATIONAL SELECTION BIAS
Noticing things we didn’t notice that much before — but we wrongly assume that the frequency has increased.
Car Example.
STATUS-QUO BIAS
Can lead to ‘From Tradition Fallacy’
How humans can be relcuatant twords change. How we like our routines and the secuity of normal. This leads us to choices that guarantee that as littel change as possbilbe occures or that things stay the same.
Change is risky- we like settled down things.
We tend to be fearful of change, even when the current circumstances arn’t great great.
Apprehensive of change,
NEGATIVITY BIAS
We unevenly weigh negtive outcomes with postive ones.
Rationality equals to responding to negtive and psotive outcomes in the same manner.
We tend to place more weight on negative information than positive information.
Dwelling on negativity at the expense of genuinely good news.
Bandwagon effect
Can lead to the ‘Appeal to Majroity Fallacy’
We tend to follow the crowd, Sometimes unconsciously.
Often even when the crowd is obviously wrong.
To lose invdual idenity and opinion to the gorup consensus, even if the larger belief is pernicious.
PROJECTION BIAS/ FALSE CONSENSUS EFFECT
We tend to overestimate how typical or normal we are and therefore we assume that most others think like us and believe what we believe
We jump between A and B, assuming other follow the same thinking. I get it, therefor others do too.
That we tend to project our belifes outward and assume there is a consensus on certin
Issues when there may be none.
THE CURRENT MOMENT BIAS
We are terrible at judging our future behavoir and tend to be more optomistic.
We are not good at imganing our selfs in the furture.
Current moment is cyestal clear, we are bad at excersing prudance.
(Procrastanation can stem from this)
- discoutning future costs/benifits
- Underestmating change
ANCHORING EFFECT
We tend to let initial values affect our appreciation of subsequent values(even when the initial value is arbitrary.
To show somthing we are less likely to interact with, then show it to us again but watered down.
THE ACTOR-OBSERVER EFFECT
We tend to let initial values affect our appreciation of subsequent values(even when the initial value is arbitrary.
To show somthing we are less likely to interact with, then show it to us again but watered down.
HINDSIGHT BIAS
We tend, after something has happened, to believe that we knew that it would happen (the “Knew it all along”)
We tend to ignore the times when we “Knew something” would happen and it didn’t.
(Linked with how people think they have ESP becuse they had some sort of premonition)
THE AVAILABILITY HEURISTIC
Our minds tend to think that information recalled easier is more important becuse it is more available.
We tend to give a lot of weight to evidence that is easily accessible. If we can easily recall something then we think it must be common or important. Information particularly memorable.
Being influcned to overestimate how likely an event can occure is.
Causation of AI biased decisions
Bad data
biases transformed from the Human maker to the AI
Weak/ Narrow AI
Narrow more singular tasks, but done at a high level.
Strong AI
Could do what humans could (Muiltiple tasks across the board) but at a higher level even better.
Automation Bias
Contradictory information in light of a computer-generated solution that is accepted as correct.
(trusting a computer more than a human on the grounds that it is a computer)
We tend to over relay on AI.
We can really use it, but must not over use it and still trust Human decisions.
Historical Bias
(ii) Example
AI is often built using Historical data. Any previous bias in the History will carry through the data into the AI.
Using older data that may hve Bias, or may not refelct the current status.
(ii) COMPAS
Algortithm that US used as a producer of risk scoring for convicts re-offending- used for parol and sentancing decisons. Affected balck people more.
(ii) Predpol:
Made a feed back loop in which police car’s patroled minioity neighberhoods more.
Bias through Interaction
(ii) Example
Bias can develop as a result of AI interacting with the world.
This is common in chatbots, where they learn how to answer questions, based on the interactions they have with the public/chatbot users.
Fooling around with friends Youtube, will interact with the AI, shifting their recommendations to certain videos.
(ii) Microsofts tay
Learnt throught interaction of the internet. Absorbed the racist, mysgonisitc side of twitter and then becomae that it’s self.
(ii) Youtube or spoitfy Altering
Sample Bias
(ii) Example
When a Sample is not representative of reality. The AI conclusion will be representative of the sample but not the world.
Training data feed, not appropriate for task at hand.
(ii)
Googles photo disaster.
Lack of diversity left google imges pulling Gorrillas under the catgory of Affericans.
(ii) Amazon Hiring Mess
Historical Biase of Men being better than woman at work made a unreprsetive sample- letting woman to be disadvantaged in the hiring proccess.
Biased Input Selection
(ii) Example
We Humans can also chose factors as inputs that are biased
In the case of risk prediction algorithms, it often leads to minority populations being disadvantaged.
(ii)
Alleghney Country Algorithm.
USed Biased factors like Hot line report as a comomponent for child abusers.
Black Families are more likely to be reaported due to racil Biases.
