The Public Scrutiny of Algorithms

Jentery Sayers | Unlearning the Internet | Week 9
DHum 150 | UVic English | 4 March 2019
Slides Online: jentery.github.io/150/slides/week9m

What's an Algorithm?

a procedure or formula for conducting a specific action
usually studied through software and programming languages
but doesn't require either
+ surprise! existed before internet + personal computing

The Combinatorial Arts

A term by Gottfried Leibniz (1660s)

Example from Jonathan Swift's Gulliver's Travels
Professor in Laputa creates a device for combining words
the results are both sense and nonsense
but the combination is perceived as wisdom
and it's accessible without knowledge of the code

"the most ignorant person at a reasonable charge,
and with a little bodily labour, may write books in philosophy,
poetry, politics, law, mathematics, and theology,
without the least assistance from genius or study”

Ok, Sayers. Let's fast-forward to the present.

Issues with Scrutiny of Algorithms

Autonomous . . . as if: entwined with society and culture
Such bias!: the cooker's habits are in the recipe + food
Observation: effects not discernible directly or via an instance
Discrimination: further enable systemic issues (see COMPAS)
Accountability: no standard or mechanism for it
Regulation: need for a public interest observatory?

Accountability

Transparency: open data, code, and process
Qualified transparency: external inspection / review
Responsibility: accept social / ethical responsibility for discrimination

Enter the algorithm audit . . .

(It's like carrying a torch in the wind.)

Types of Algorithm Audits

Question: not good or evil, but what are the mechanisms for observation?

Code: researchers study code (problems: IP and abuse of public code)

Noninvasive user study: study selection of user behaviour
(problems: sampling and validity of self-reporting)

Scraping: researcher queries with script and observes results
(problems: terms of service and legality)

Sock puppet: computers impersonate users
(problems: false data injection and legality)

Collaborative / crowdsourced: the sock puppets are humans
(problems: false data injection and incentive)

Google Search = Our Example

Google search crawls, indexes, ranks, and contextualizes

Oh Whoa, The Sourcery

The act of reducing computational processes to code (as source)
Code operates like magic (people may be wowed by whiz-bang effects
without knowing how to explain its causes)
Code manages to be highly rational yet deeply mysterious
But code is a component, not a source, of the combinatorial process
Components are "re-sources" (rationalized after the fact)
We can scrutinize algorithms and processes without focusing solely on code

bit.ly/heyjentery