Computer Ethics

Required:

• Read chapter one from "Artificial Unintelligence: How Computers Misunderstand the World" [pdf] [url] by Meredith Broussard, 2018.

  In her introductory chapter, computer scientist and data journalist Broussard lays out both her love of and skepticism for computing technology. You might find a bit of yourself in her.

• Read "Be Careful What You Code For" [url] by danah boyd, 2016 (2 pages).

  danah boyd, a researcher at Micosoft and at Data and Society, highlights just how few guardrails there are for developers, from the consequences of algorithmic bias to the implications of crazy metaphors. She offers a call to action, solutions, and ample evidence for considering the implications of code. We highly recommend chasing down some of the links provided.

Optional:

• Read "How to read a book" [pdf] by Paul N. Edwards, 2000 (8 pages).

  Here, an adroit scholar walks through some tips and tricks for reading more effectively. He hits the major points and includes some bonus tips, like where to best organize your reading notes. This is an invaluable resource as our course’s weekly reading load begins to increase. Skim now, but revisit throughout the course.

• Read "Discussion Leading" [pdf] [url] by John Rickford et al., 2007.

  This resource from the Stanford Teaching Commons offers an in-depth analysis of how to have better discussions. Their recommendations, from setting an agenda, asking questions, and increasing discussant engagement are all a part of how to create a better climate for discourse. For leading discussions in this class and beyond, it’s worth a read.

Before Class:

• Daily assignment, due at 8pm the night before class:
  • What do you want to get out of our class discussions?
  • Do you feel able to change outcomes of how tech affects society?
• Preview the course project, part zero.

Required:

• Read "Chapter 1: Conceptualising Data" [pdf] by Rob Kitchin, 2014 (25 pages).

  The introduction to the book describes ways of thinking about what data is (ontologies) and goes on to discuss ethical and political considerations of data. It postulates the framework of “data assemblages” and how thoughts about them influence their own conceptions.

• Read "On Being a Data Skeptic" [pdf] by Cathy O'Neil, 2014 (19 pages).

  This rapid-fire, well-articulated article is about the advantages and perils of data science, with ample advice and examples to advocate for why those who use data ought to be this special kind of “skeptical”.

Before Class:

• Daily assignment, due at 8pm the night before class:

  Use the following questions as a resource to guide your reading. Then, respond to a few of them, or otherwise on theme.

  • What is the difference between data, information, and knowledge? How are they related?
  • How can wider or economic concerns “frame” data? That is, in what sense do data act and how? (Answers might include: as an economic resource, as a form of power or knowledge, etc.) Explain why.
  • How have politics or economics influenced how some data have been defined or created?
  • What are reasons or incentives for controlling:
    • The creation of data
    • The access to data
    • The standards of data, such as metrics or units
    • The means of data collection, such as sensors or know-how
  • From “On being a data skeptic” explain “measuring the distortion.”
  • What is the relationship between models and proxies? Why are proxies used? Give an example.
  • Why might O’Neil have singled out “nerds” and “business people” separately? What do the differences in her comments indicate about how those groups view problems differently? Do you agree?
  • Submit a clarifying question which you’d like to discuss in class.
• Finish the course project, part zero by tonight.

Required:

• Read introduction (pg. 1 - 16; 31 - 32; 17 pages total) from "Sorting things out: classification and its consequences" [pdf] by Geoffrey C. Bowker et al., 1999.

  Sorting Things Out is a classic text on classification and standardization. The introduction writes about the importance of considering the ubiquity of classification and the processes that generate them, standardize them, and enforcement. It also looks at how classification has caused harm and how the processes which create standards can at times yield an inferior solution. The authors take an expansive view of classification, so be prepared to think about the American Psychiatric Association’s Diagnostic and Statistical Manual (DSM) or VHS vs. Betamax.

• Read "Do algorithms reveal sexual orientation or just expose our stereotypes?" [url] by Blaise Aguera y Arcas et al., 2018 (5 pages).

  This essay sets out to debunk a scientific study which claimed to have built a “sexual orientation detector” using machine learning. “Do algorithms reveal sexual orientation or just expose our stereotypes?” presents a thorough analysis of the offending paper and shows that one move to debunk “junk science” is validate the study’s results against some other baseline. In this case, the authors use Amazon’s Mechanical Turk. As you’re reading this, think about what one can learn from a face?

Optional:

• Read chapter one (pg. 46 - 50 from "Infrastructure" on) from "Sorting things out: classification and its consequences" [pdf] by Geoffrey C. Bowker et al., 1999 (377 pages).

• Explore "Datasets Have Worldviews" [url] by Dylan Baker, 2022.

  Check out this interactive demo about classification.

• Read "Deep neural networks are more accurate than humans at detecting sexual orientation from facial images." [url] by Yilun Wang et al., 2017.

  This is the article critiqued by “Do algorithms…”

Before Class:

• Daily assignment, due at 8pm the night before class:
  • “Do algorithms…” claims to focus on the underlying “science.” Why do you think the authors did so? Why was this distinction important?
  • What strategies did “Do algorithms…” use to make its argument?
  • The authors of “Do algorithms…” conclude that the paper they examined was misguided. Drawing on the discussion of classification in “Sorting Things Out,” think of another misguided method (any method, scientific or otherwise, and not necessarily one related to the “Do Algorithms” paper). Now, describe that misguided method which you have identified.

Required:

• Read "Introduction: Why Data Science Needs Feminism" [pdf] [url] by Catherine D'Ignazio et al., 2020.

  Through a historical examination of women in technology, D’Ignazio and Klein, both leading scholars in the field, introduce us to feminism and its role in shaping technologies. This is just a taste and the whole book is worth a read.

• Read "Gender Equality Paradox Monkey Business: Or, How to Tell Spurious Causal Stories about Nation-Level Achievement by Women in STEM" [url] , 2020 (5 pages).

  A widely known study argued that countries with more gender equity in society have fewer women studying STEM, but this article accompanies a peer-reviewed publication casting doubt on the study’s analysis – a scholarly back-and-forth also playing out in the blogosphere.

Optional:

• Read "Patriarchy, technology, and conceptions of skill" [pdf] by Judy Wajcman, 1991 (16 pages).

  When considering the future of work, one question that’s often raised is how technology negatively impacts the amount of “skill” required to complete a task, aka “deskilling”. In “Patriarchy, Technology, and Conceptions of Skill”, Judy Wajcman questions the underlying assumption that skill is entirely technically derived. Instead, she considers how men’s historical control over technology in the workplace has extensively influenced the ideological and material conceptions of skill, thus concluding in part that “definitions of skill, then, can have more to do with ideological and social constructions than with technical competencies which are possessed by men and not by women”.

• Read "Technically Female: Women, Machines, and Hyperemployment" [url] by Helen Hester, 2016 (10 pages).

  This essay surveys a history of “electronic secretaries” to frame relevant questions of today’s tech, such as: Why are AI assistants so often feminized? We question what it means for technology to “do gender” and in service of which “imagined technology user”? Yet we can turn that question around and ask who “does technology”? and how does labor gets redistributed with the introduction of new software and AI assistants? Ultimately, Hester asks us to confront questions concerning lived experiences of gender and how its programmed, productive vs. reproductive labor, and the (dis)advantages of automation.

• Read "Testosterone rex: unmaking the myths of our gendered minds" [url] by Cordelia Fine, 2017.

  If you were looking to throw the book at someone who continues to insist that sex differences are sufficient to explain gender differences, this would be that book. Take her word for it: “Every influence is modest, made up of countless small instances of its kind. That’s why everything—a doll packaged in pink, a sexist joke, a male-only expert panel—can seem trivial, of intangible effect. But that’s exactly why calling out even seemingly minor points of sexism matters. It all adds up, and if no one sweats the small stuff, the big stuff will never change.” And it’s funny, too: “If we stop believing that boys and men are emotional cripples and fly-by-night Casanovas who just want sex, and start believing that they’re full, complete human beings who have emotional and relational needs, imagine what might happen.”

Before Class:

• Daily assignment, due at 8pm the night before class:
  • Why is feminism relevant to data science?
  • What considerations does data feminism require us to make?
  • “Gender Equality Paradox” is an example of data feminism at work. How so?
• Finish the course project, part one by tonight.

Required:

• Read Chapter one, just "STORIES CHANGED BY AN ALGORITHM" (12 pages), Interlude (4 pages) and Chapter two, from "WHICH ALGORITHMS ARE HIGH STAKES" on (20 pages) (36 pages total) from "Voices in the code" [pdf] by David Robinson, 2022.

  (The text is too dense. Feel free to skim in parts.) Robinson delivers a poignant and approachable introduction to the question of how to govern AI and other algorithmic decision systems by both surveying the field and opening a case study on kidney transplants.

• Read the introduction from "Automating Inequality" [pdf] by Virginia Eubanks, 2017 (13 pages).

  In the introduction to her acclaimed book, Eubanks captures the personal consequences of poorly designed decision systems, particularly on poor and working class people.

Optional:

• Read "Fairness and Machine Learning" [url] by Solon Barocas et al., 2019.

  Scholars Barocas, Narayanan, and Hardt collaborate on one of the most readable and insightful texts to address questions of fairness in AI, ML, and related decision systems.

Before Class:

• Daily assignment, due at 8pm the night before class:

  Chose one of the four governance strategies Robinson lists in his second chapter. Identify a technology or system (one not mentioned in the text) that you think could benefit from this kind of governance. Why?

Required:

• Read "It's the (Democracy-Poisoning) Golden Age of Free Speech" [pdf] [url] by Zeynep Tufekci, 2018 (4 pages).

  Zeynep Tufekci considers how power to censor functions on our oversaturated social networks, and the role of misinformation and the attention economy in this. The article provides striking clarity to issues we collectively face on this platform.

• "How social media could teach us to be better citizens" [pdf] by Ethan Zuckerman, 2022 (5 pages).

  Zuckerman comments on historical aspects of social media, connecting to the mid-20th century, asking what social media does for us.

• Read chapter four (pg. 74 - 110; 36 pages total) from "Custodians of the internet: platforms, content moderation, and the hidden decisions that shape social media" [pdf] by Tarleton Gillespie, 2018.

  Read the whole chapter, but, if you’re short on time, focus on the “automatic detection” section (pg. 97 to 110). Tarleton Gillespie, a social media scholar, establishes the groundwork for understanding social media platforms. In chapter 4, “Three Imperfect Solutions to the Problem of Scale”, Gillespie considers several models for content moderation, one of which is algorithmic “automatic detection” techniques.

Optional:

• Read chapter 17 from "Genius Makers : The Mavericks Who Brought AI to Google, Facebook, and the World" [pdf] [url] , 2021.

  In this chapter of his book on the rise of neural networks, Metz, a veteran journalist of Silicon Valley, concisely describes the claims which some make about those tools. Will AI “solve” content moderation for us? Read on.

• Read "Hey Elon: Let Me Help You Speed Run The Content Moderation Learning Curve" [url] by Mike Masnick, 2022.

  Will Elon Musk solve free speech? Read on.

• Read "The Problem of Free Speech in an Age of Disinformation" [url] by Emily Bazelon, 2020 (14 pages).

  Read this article for an up-to-date account of free speech and its history in the United States. Notice the similarities between analog and social media. It questions how different governmental approaches to speech may be making us more or less free.

• Read "How Facebook Hides How Terrible It Is With Hate Speech" [url] by Noah Giansiracusa.

  Cut through the claims of AI content moderation and listen to Facebook’s own assessment that they ‘may action as little as 3-5% of hate … on Facebook.’

• Read "The Risk of Racial Bias in Hate Speech Detection" [pdf] [url] by Maarten Sap et al., 2019 (9 pages).

Before Class:

• Daily assignment, due at 8pm the night before class:

  There’s been a whole mess of issues at Twitter recently. Blue check marks. Layoffs. Re-platforming. What’s going on with content moderation at Twitter? What do you think should be done?

  Try to be specific in exactly what problem you are addressing and the trade-offs involved in your proposal.

  Your answer might include:

  • De-platforming
  • More automated moderation
  • Hire more moderators
  • Regulation…
    • e.g. changing liability of publisher
    • or otherwise

Required:

• Read "ChatGPT Is Dumber Than You Think" [pdf] [url] by Ian Bogost, 2022 (5 pages).

  Bogost playfully unpacks the implications of large language models such as GPT-3 from OpenAI. Do they work as our enthusiasm for them might project?

• Read "AI's Jurassic Park moment" [url] by Gary Marcus, 2022 (2 pages).

  Marcus, in his very abbreviated style, lays out why current large language models built on deep learning fail and why we might need another approach.

• Read (stop or start skimming at "Language as the key to general AI") "Can machines learn how to behave?" [url] by Blaise Agüera y Arcas, 2022 (21 pages).

  Agüera y Arcas, whom we’ve seen elsewhere in class, makes the argument that there is no clear line separating what AI models can understand from what people do. The issue, in his view, is only that we need to teach those models what is and is not ok.

Optional:

• Read "Language Models Understand Us, Poorly" [pdf] [url] by Jared Moore, 2022 (5 pages).

  Jared lays out a couple of views on language understanding. You can also review the slides.

• Read "Language Models as Agent Models" [url] by Jacob Andreas, 2022 (10 pages).

  Andreas makes the case that language models can rightfully be thought of as models of humans, of agents. The philosophical and historical context is a bit lacking here. Take cse490a1, the philosophy of AI, to really get into the details.

• "Inside the Heart of ChatGPT’s Darkness" [url] by Gary Marcus, 2023.

  Read this for some quite toxic examples of how ChatGPT fails to moderate its language.

• "ChatGPT Is a Blurry JPEG of the Web" [pdf] [url] by Ted Chiang, 2023 (8 pages).

  Chiang, in an analogy which tugs at computer science fundamentals, compares language models like ChatGPT to lossy photographs.

• Read "Semantics derived automatically from language corpora contain human-like biases" [pdf] [url] by Aylin Caliskan et al., 2017 (4 pages).

  This is the classic paper which set off the debate about bias in langauge models. You can also check out Caliskan’s similar 2022 paper finding bias in image models.

Before Class:

• Daily assignment, due at 8pm the night before class:

  What kind of issues are there in using large language models? Why might they be or not be a very good idea? In a sense: do they work?

• Finish the course project, part two by tonight.

Required:

• Read "Three Expensive Milliseconds" [pdf] [url] by Paul Krugman, 2014 (1 page).

  While reading this article, consider: who is today’s infrastructure for? What are some metrics being optimized for which have led to some (perhaps) unexpected consequences? How are these metrics and systems shaping our world?

• Read chapter two from "Geek Heresy" [pdf] by Kentaro Toyama, 2015 (21 pages).

  The classically trained computer scientist Toyama, in an excerpt from his book about realizing how Utopian tech fails the most marginalized, describes the role of tech as amplifying—whether for good or for bad.

• Read "Survival of the Richest" [url] by Douglas Rushkoff, 2019 (3 pages).

  From Douglas Rushkoff, a noted media theorist read of, “Apocalypto – the intolerance for presentism leads us to fantasize a grand finale. “Preppers” stock their underground shelters while the mainstream ponders a zombie apocalypse, all yearning for a simpler life devoid of pings, by any means necessary. Leading scientists – even outspoken atheists – prove they are not immune to the same apocalyptic religiosity in their depictions of “the singularity” and “emergence”, through which human evolution will surrender to that of pure information.” This idea is investigated more deeply in his book “Present Shock”.

Optional:

• Listen to "Speed" [url] by RadioLab, 2013.

  An enlivening radio show which goes over many of the same concerns about speed as the readings.

• Read "Tech Leaders Justify Project To Create Army Of AI-Controlled Bulletproof Grizzly Bears As Inevitable Part Of Progress" [url] , 2022.

  The title explains it all

• Read "The Microsoft Provocateur" [pdf] [url] by Ken Auletta, 1997 (14 pages).

  This piece shows us the thinking of tech folks at a pivotal time when the nature of the internet was not yet decided. Skim the sections which are just about Mhyrvold’s life—we mean to focus on founders and their utopian visions.

• Read "Origin Stories of Tech Companies if Their Founders Had Been Women" [url] by Ginny Hogan, 2019 (1 page).

Before Class:

• Daily assignment, due at 8pm the night before class:

  Oftentimes in computer science, we consider only the upshot, only the Utopian worlds we might create.

  • For one of the technologies mentioned in the readings or another of your choosing, comment on the future worlds we imagine for it (e.g. a Utopia) and complications to that future which might arise in practice (e.g. a Dystopia).
  • What are other utopias you have heard described when talking about emerging technologies? Do you find those utopias compelling? Why or why not?

Required:

• Read "Critical Race Theory for HCI" [pdf] [url] by Ihudiya Finda Ogbonnaya-Ogburu et al., 2020 (1–16 page).

  This article calls human-computer interaction research, and computing research more generally, to explicitly attend to race, namely through critical race theory. Through this theory, analysis of the field, and storytelling, the authors show that despite (some) efforts in computing, racism persists and bares redressal. Pay particular attention to the stories.

• Read "Roles for computing in social change" [pdf] [url] by Rediet Abebe et al., 2020 (9 pages).

  This reading offers concrete suggestions specific to computing research. It advances four roles such research can play in discussions around fairness, bias, and accountability.

• Read introduction through page 16 from "Race after technology: Abolitionist tools for the new jim code" [pdf] by Ruha Benjamin, 2019.

  Benjamin, in a recent book, offers a “race conscious orientation to emerging technology not only as a mode of critique but as a prerequisite for designing technology differently.” Affiliated with Princeton’s Center on Information Technology and Policy, she brings a fresh perspective to many of the foundations of computing.

Optional:

• Read introduction (16 pages) from "Stuck in the shallow end: education, race, and computing" [pdf] by Jane Margolis, 2008.

  This introductory chapter details an early 2000s study which attempted to figure out why so few black and latinx students were enrolling in computer science courses. It, “shows how segregation and inequality along racial lines operate on a daily basis in our schools, despite our best intentions” (16).

• Listen to particularly minutes 15 through 30 from "Episode 12: Confronting Our Reality: Racial Representation and Systemic Transformation with Dr. Timnit Gebru" [url] by Dylan Doyle-Burke et al..

  Gebru, who received her Ph.D. from Stanford and then worked for Google before very publicly not working for Google, discusses the necessity of focusing on your values (hers being racial justice) in conjunction with your work.

• Read "Combating Anti-Blackness in the AI Community" [pdf] [url] by Devin Guillory, 2020.

  This reading is more about a computing research community than about the external impact of our technology, but each influences the other.

Before Class:

• Daily assignment, due at 8pm the night before class:

  Answer at least two of the following.

  On “Critical Race Theory for HCI”:

  • What is interest convergence?
  • Why do you think the authors include information about their background? What does this achieve?
  • Why do the authors contend that theory is an important focus?
  • What does “true recognition of the pervasiveness of racism” (pg. 9) require?
  • Do you have any stories along the lines of those shared? Would you be interested with sharing any with the group?

  Choose one of the roles from “Roles for Computing in Social Change.”

  • How might your work as a computer scientist engage with your chosen role?
  • What social problem would you like to work on?

Required:

• Read chapter 12 from "Artificial Unintelligence: How Computers Misunderstand the World" [pdf] [url] by Meredith Broussard, 2018.

  Having read Broussard’s commentaries on the first day and as an introduction to the data unit, we now finish with her conclusion: a renewed plea for computing technology to serve the people who made it—humans.

• Read foreward and introduction (pg. xi - xxvi; pg. 1 - 5; 23 pages total) from "Hope in the dark: Untold histories, wild possibilities" [pdf] by Rebecca Solnit, 2016.

  Solnit, a writer and activist, reflects on our desire for social, cultural, or political change given the appearance that we have not arrived there (considering issues from global warming, to human rights abuse). Originally responding to the war in Iraq, she explores how news cycles and our personal narratives frame these issues and makes the case for hope nonetheless–“tiny and temporary victories”

Optional:

• Read From 'SAN DIEGO' on from "The Code: Silicon Valley and the Remaking of America" [pdf] by Margaret O'Mara, 2019.

  From the UW historian O’Mara hear this story of how two UW CSE alumni have navigated their careers as computer scientists.

Before Class:

• Daily assignment, due at 8pm the night before class:
  • Do you feel able to change outcomes of how tech affects society?
  • What’s an idea from this course that every UW CSE student ought to understand?
• Finish the course project, part three by tonight.