March Flashcards
(84 cards)
Utilitarianism author date
John Stuart Mill 1863
Hello world author date
Hannah Fry 2018
4 types of algorithm
Prioritisation
Classification
Association
Filtering
Kadoodle experiment scientists year conclusions
Robert Epstein
Ronald Robertson
2015
After a few minutes participants 12% more likely to vote for favoured candidate.
‘The search engine manipulation effect (SEME) and its possible impact on the outcomes of elections’, Proceedings of the National Academy of Sciences, vol. 112, no. 33, 2015, pp. E4512–21, http://www.pnas.org/content/112/33/E4512.
Linda J. Skitka, Kathleen Mosier and Mark D. Burdick, ‘Accountability and automation bias’, International Journal of Human–Computer Studies, vol. 52, 2000, pp. 701–17, http://lskitka.people.uic.edu/IJHCS2000.pdf.
We end up believing that algorithms have superior judgement and we are unaware of our own biases that result.
Russian military officer who avoided nuclear war
Staislav Petrov 1983 26 september
Alton towers tragedy
2015
2 girls lost legs
Engineers had just fixed a fault. Left test car on track. Algorithm detected a dangerous obstacle and set off alarm, stopping the ride. Engineers overruled.
Algorithm aversion
Though we tend to trust algorithms we don’t understand, as soon as they make a mistake we tend to overcriticise them. People are less tolerant of a machine’s mistakes than their own, even when their own mistakes are bigger.
Insurance company used data from supermarket loyalty scheme to discover that people who buy what are less likely to claim on home insurance
Fennel. Home cooks are more houseproud and careful.
Peter Thiel data broker company
Palantir 2003 $20bn
Sesame credit
Chinese scoring system based on internet data. Score between 350-950. Over 600 gets you a special credit card. 750 is a fast tracked visa for Europe. Mandatory in 2020
Elizabeth Fry point about judges and algorithms
Algorithms imperfect and currently amplify bias. However judges are biased and manipulable. Algorithms have potential to be more objective and fairer.
An algorithm designed for specificity will have…
Few false positives but some false negatives.
An algorithm designed for sensitivity will have…
Few false negatives and some false positives.
Competition between algorithms and pathologists
CAMELYON16
HUMAN: 96% accurate no false positives.
Neural nets 92% considerable false positive rate.
Nun Alzheimer’s study
Over 600 nuns studied Whether a nun developed Alzheimer’s or not easily predictable by the literacy demonstrated in letters written in their early 20s
Kant dates
1724-1804 (79)
Kant three formulations of the categorical imperative
1 Universal law of nature formula
2 Humanity formula
3 Autonomy formula
Lady Godiva
Rode naked through Coventry as part of a bargain with her husband so he would remove the taxation on the city. Phrase peeping tom originates as he looks even though she asked the whole town to stay indoors and shut their curtains.
Marx. Man is…
… a creative labourer.
Lisanne Bainbridge 1983 psychologist big idea
If you build a machine to improve human performance it will lead to a reduction in human ability
R D Laing study of family
Applied game theoretical approach to analyse and quantify the effects of a range of behaviours, concluding that even ostensibly loving actions are tools for exerting control and power over others.
PredPol
Algorithm that predicts crime hotspots based on patterns in reported crime.
Study of novelty in films
Sameet Sreenivasan 2013 analysis of IMDB tagging. Showed that increased novelty up to 0.8 led to higher success. But after that it was certain failure.