Ethics in Techniques for large-scale data Graham J.L. Kemp - - PowerPoint PPT Presentation

Graham J.L. Kemp

Ethics in “Techniques for large-scale data”

2019-01-09 Chalmers 2

TECHNIQUES FOR LARGE-SCALE DATA

Master’s level course:

Chalmers MPALG and MPSOF programmes (DAT345) GU Applied Data Science programme (DIT871)

Learning outcomes include: “On successful completion of the course the student will be able to discuss large-scale data processing from an ethical point

• f view”

2019-01-09 Chalmers 3

“JUSTICE: WHAT’S THE RIGHT THING TO DO?”

If you want a moral philosophy perspective on ethical decision making: Professor Michael Sandel’s lectures from a course at Harvard introducing moral and political philosophy First lecture includes some “trolley car” scenarios

https://www.youtube.com/watch?v=kBdfcR-8hEY&list=PL30C13C91CFFEFEA6

2019-01-09 Chalmers 4

DIFFERENT ETHICAL FRAMEWORKS EXIST

and can lead to different recommendations e.g. Kantian ethics vs. Utilitarianism https://www.youtube.com/watch?v=7FR-FuhN2HM But it is assumed that students have a sense of right and wrong, and are capable of discussing a scenario and taking a view on whether an action is ethical, even if they haven’t taken a course in moral philosophy and are not familiar with the theories and principles that define different approaches to ethics.

2019-01-09 Chalmers 5

DISCUSSING ETHICAL ISSUES

Identify stakeholders Identify benefits and possible harm for each stakeholder Weigh benefits against possible harm Get input from others Would you want an internal ethical assessment to be seen

• utside the organisation?

Would you publish an internal ethical assessment on the

• rganisation’s web site?

2019-01-09 Chalmers 6

Determining user personality characteristics from social networking system communications and characteristics

“A social networking system obtains linguistic data from a user's text communications on the social networking system. For example,

• ccurrences of words in various types of communications by the user in

the social networking system are determined. The linguistic data and non-linguistic data associated with the user are used in a trained model to predict one or more personality characteristics for the user. The inferred personality characteristics are stored in connection with the user's profile, and may be used for targeting, ranking, selecting versions

• f products, and various other purposes.”

US patent US8825764B2

2019-01-09 Chalmers 7

Car insurance prices based on Facebook posts

• “Admiral Insurance will analyse the Facebook accounts of first-time car
• wners to look for personality traits that are linked to safe driving. For

example, individuals who are identified as conscientious and well-

• rganised will score well.”
• “These [traits] include writing in short concrete sentences, using lists,

and arranging to meet friends at a set time and place, rather than just “tonight”.”

• “In contrast, evidence that the Facebook user might be overconfident –

such as the use of exclamation marks and the frequent use of “always”

• r “never” rather than “maybe” – will count against them.”
• “The scheme is voluntary, and will only offer discounts rather than price

increases”

https://www.theguardian.com/technology/2016/nov/02/admiral-to-price-car-insurance-based-on-facebook-posts

2019-01-09 Chalmers 8

OTHER SCENARIOS

• In-store analytics using data from video cameras, Wi-Fi and

Bluetooth devices, guest Wi-Fi, point-of-sales systems, payment cards, etc.

• Pedestrian analytics using cameras at feet level

2019-01-09 Chalmers 9

THE VEIL OF IGNORANCE (RAWLS 1971)

“The customer …” “The user …” “The company …” “The developer …” “We …”

https://www.huffingtonpost.com/linch-zhang/behind-the-absurd-popular_b_10247650.html

2019-01-09 Chalmers 10

TEN SIMPLE RULES FOR RESPONSIBLE BIG DATA RESEARCH

Matthew Zook, Solon Barocas, danah boyd, Kate Crawford, Emily Keller, Seeta Peña Gangadharan, Alyssa Goodman, Rachelle Hollander, Barbara A. Koenig, Jacob Metcalf, Arvind Narayanan, Alondra Nelson, Frank Pasquale PLOS Computational Biology 13(3): e1005399 (March 2017). https://doi.org/10.1371/journal.pcbi.1005399 ”we structure the first five rules around how to reduce the chance of harm resulting from big data research practices; the second five rules focus on ways researchers can contribute to building best practices that fit their disciplinary and methodological approaches.”

2019-01-09 Chalmers 11

• 2. Recognize that privacy is more than a binary value

TEN SIMPLE RULES …

“creepy”: when social values and technical capabilities are not aligned ”Understand that your attitude towards acceptable use and privacy may not correspond with those whose data you are using, as privacy preferences differ across and within societies.”

2019-01-09 Chalmers 12

• 6. Debate the tough, ethical choices

TEN SIMPLE RULES …

“Discussion and debate of ethical issues is an essential part of professional development—both within and between disciplines—as it can establish a mature community of responsible practitioners.” ”Why might one set of scholars see [some ethical case] as a relatively benign approach while other groups see significant ethical shortcomings? Where do researchers differ in drawing the line between responsible and irresponsible research and why?”

2019-01-09 Chalmers 13

• 10. Know when to break these rules

TEN SIMPLE RULES …

“For example, in times of natural disaster or a public health emergency, it may be important to temporarily put aside questions of individual privacy in order to serve a larger public

• good. Likewise, the use of genetic or other biological data collected

without informed consent might be vital in managing an emerging disease epidemic.” ”Ethics is often about finding a good or better, but not perfect, answer, and it is important to ask (and try to answer) the challenging questions.”

2019-01-09 Chalmers 14

TOOLS FOR INVESITIGATING POSSIBLE ETHICAL ISSUES

Fraud detection, e.g.

Fraud Detection in Real Time with Graphs https://www.youtube.com/watch?v=AeNufTq1W5I First party fraud [13:48-18:33] Insurance fraud [26:45-27:48] eCommerce fraud [27:50-29:15]

Panama papers

https://www.youtube.com/watch?v=8VJ4njrA0TQ [from 2:42; Neo4j from about 14:00]

2019-01-09 Chalmers 15

EXAM QUESTION (JUNE 2018)

Suppose you are part of a university's IT team and have been asked to help improve campus safety by implementing automatic number plate recognition technology to monitor vehicles on campus. Discuss whether there are ethical issues that need to be considered.

2019-01-09 Chalmers 16

EXAM QUESTION (AUGUST 2018)

Consider the following newspaper report (from The Independent, 12 June 2018) that discusses a patent application filed by Uber: The "Predicting User State Using Machine Learning" patent aims to identify people in an "abnormal state" from factors including number of typos made, walking speed, and how precisely a customer clicks on buttons and links on their phones. "The user behaviour may be compared against the user's prior behaviour to determine differences in the user behaviour for this request and normal behaviour of prior requests," the patent states. "The system can alter the parameters of a service based on the prediction about the state of the user requesting the service.” This means that Uber drivers could theoretically refuse to accept passengers deemed to be in an inebriated state. Discuss whether there are ethical issues that need to be considered if predicted user state is used in determining whether a potential user is provided with a service.

2019-01-09 Chalmers 17

ASSESSING EXAM QUESTION SOLUTIONS

ü Main stakeholders identified ü Main benefits and harm for each stakeholder described ü Weigh benefits and harm, offering an opinion for or against implementation ü Less obvious stakeholders identified ü Less obvious benefits or harm described