Fastbook Lesson 3

1. Does ethics provide a list of "right answers"?

No. Ethics is dealing with good and bad; a set of moral principles. It deals with what questions to ask and how to weigh each decision. Ethics provides us with different perspectives of the problem and solution, whether it's utilitarian or hedonistic or consquential and then it depends on the situation and the principles you hold. Ethics are just a well-founded standard of what is outght to be done, there's no fixed set of rules and it's not the same as religion, law or social norms.

2. How can working with people of different backgrounds help when considering ethical questions?

Cultural relativism needs to be considered while answering ethical questions, because diversity and differences factor down to the decisions we take. For exapmle in some cultures it's considered auspicious to kill someone before they reach the age of 50, or like in Indian cultures where Sati was practiced out of duty. Placing objectivist views over such practices and ideas doesn't completely portray them.

3. What was the role of IBM in Nazi Germany? Why did the company participate as it did? Why did the workers participate?

IBM supplied the Nazis with machines to facilitate the holocaust. These were used to store the information of a person, whether they were Jewish or not, how they were killed in concentration camps. These machines required a lot of maintenance and an ongoing relationship with vendors, which IBM provided. There's another aspect to the story when we get to know why only IBM was selected for this task. It was bacause of the religious affiliations of IBM and it had declared itself German and free from foreign influences. By 1935, the profits coming in were so much that it became impossible not to report them. It indeed appears that no one at IBM in the United States really knew what was going on at Dehomag, and there is no smoking gun anywhere for Watson’s or any American’s witting involvement in Nazi crimes. To the blind technocrat, the means were more important than the ends. The destruction of the Jewish people became even less important because the invigorating nature of IBM’s technical achievement was only heightened by the fantastical profits to be made at a time when bread lines stretched across the world. We must keep in mind: "Technology is neither good nor bad; nor is it neutral." The fruits of human inventiveness—whether punch-card machines, miracle drugs, atom bombs, or plowshares—contain no morality whatever in themselves, but how they are used is absolutely always a moral matter.

4. What was the role of the first person jailed in the Volkswagen diesel scandal?

James Liang was jailed and he was just following orders.

5. What was the problem with a database of suspected gang members maintained by California law enforcement officials?

It had 42 babies below the age of 1 registered as gang members and about 28 of them had admitted on being gang members. This databse was rarely updated only appended and it remains a question of how many of those entries were actually correct.

6. Why did YouTube's recommendation algorithm recommend videos of partially clothed children to pedophiles, even though no employee at Google had programmed this feature?

7. What are the problems with the centrality of metrics?

Centrality in the metrics often leads to gaming, which means that for example certain YouTube channels can realise the techniques and secrets to show up higher on a search result by using means like fake likes, views and followers etc. There could be a myopic focus on short term goals, like take the concept of Buzzfeed articles. It doesn't matter is the user just clicked because of the title the content no longer has to live up to the hype. There could be widespread manipulation, like how on social media political parties could then spread falsehoods about their opponents, ginning up panics about minority groups, and undermining people’s trust in the independent media. It could also lead to unexpected negative consquences. Much of AI/ML are centered around optimizing a metric and that's where this becomes a boon and a bane.

8. Why did Meetup.com not include gender in its recommendation system for tech meetups?

9. What are the six types of bias in machine learning, according to Suresh and Guttag?

10. Give two examples of historical race bias in the US.

11. Where are most images in ImageNet from?

Two-thirds of the images in ImageNet are from the US and other western countries that is why it performs significantly worse on data that doesn't come from the same region.

12. In the paper "Does Machine Learning Automate Moral Hazard and Error" why is sinusitis found to be predictive of a stroke?

This is a classic example of a measurement bias. Predictors of stroke were mesured as: Prior stroke, Prior accidental injury, abnormal breat feeding, cardiovascular disease history, colon cancer screening and acture sinusitis. The first, second and fourth are fine. The other four predictors—accidental injury, benign breast lump, colonoscopy, and sinusitis—are somewhat more mysterious. Could we have discovered novel biomedical risk factors for stroke?

These measures are far removed from the prediction of a stroke. Here we see the presence of a billing code and notes recorded by a doctor. Moreover, unless the stroke is discovered during the ED visit, the patient must have either decided to return, or mentioned new symptoms to her doctors during hospital admission that provoked new testing. In other words, measured stroke is (at least) the compound of: having stroke-like symptoms, deciding to seek medical care, and being tested and diagnosed by a doctor. Medical data are as much behavioral as biological; whether a person decides to seek care can be as pivotal as actual stroke in determining whether they are diagnosed with stroke. Many decisions and judgments intervene in the assignment of a diagnosis code—or indeed, any piece of medical data, including obtaining and interpreting test results, admission and re-admission of a patient to the hospital, etc. Here we predicting both heavy utilization in healthcare as well as a stroke. The problem is we haven't measured stroke separately, what we have measured is the symptoms, tests, and the diagnosis of strokes. This seems like a reasonable proxy for a stroke but a proxy is not wanted when you want to exactly know which disease it is. It's because these proxies can often determine any medical condition.

13. What is representation bias?

14. How are machines and people different, in terms of their use for making decisions?

• People assume they are objective and error-free.
• Algorithms are more likely to be implemented with a no apeals system in place.
• They are used at scale(they may be replicating bias at scale) and are cheaper.
• ML can create feedback loops

15. Is disinformation the same as "fake news"?

Fake news refers to those news stories that are false: the story itself is fabricated, with no verifiable facts, sources or quotes. Disinformation is when that false story is fabricated to cause harm to a person or a groups of people and is used to influence public opinion and obscure the truth. Whereas misinformation is false or inaccurate information that is mistakenly or inadvertently created or spread; the intent is not to deceive.

16. Why is disinformation through auto-generated text a particularly significant issue?

17. What are the five ethical lenses described by the Markkula Center?

18. Where is policy an appropriate tool for addressing data ethics issues?