I'm impatient -- the answer is this: ask permission every. single. time.

[russellhanna]

You guys covered good stuff, and data science can proceed on encrypted data (link at bottom) it looks like. So I'm in. (You had me at hello, right!)

I think your vote controls example is spot on, and also shows some narrowness of where encryption adds value...

Differential privacy might make it easier for me to consent securely. But my consent is only risky to me because:

• (a) it's still casual usage to say entities "own" person(al) data (as your course does about hospitals owning medical data) when in fact entities only have open-ended consent,
• and (b) it's hard to ask for consent for each and every use of person-data, so consent contracts are written with these open ends.

I really like the idea of differential privacy. While I may sound like, "security is a Band-aid" I'll take it! Band-aids are good.

In the longer view, I want to lobby you for 2 things right away:

1. Shift your focus downstream to the "perception is reality" problem, which is only sort of related to privacy. To enhance our discussion, I would say "information" also requires relevance and logic. Encryption does not address either. Some people only sort of care about evidence. Many people aren't even aware of the fallacies of reversing a rule or proving a negative. For example, "the 9-11 terrorists were Saudi muslims." is mostly true. But the reverse is NOT AT ALL TRUE: "All Saudi muslims are terrorists." Yet it's not uncommon for people to conflate these -- either when reading them or when making travel regulations. To support my assertion about negatives, I bet you have seen plenty of journalism that assumes "absence of evidence" is equal to "evidence of absence." Clear thinking and evidence are key, yet they seem to be ignored pretty easily -- encrypted or not. And when we get these things sideways, they provoke emotions that can push us even farther off the target.

2. Help make sure that individuals own their data artifacts. This one is key, but I want to introduce it properly, below.

Yes, you've identified the system of "attention" in social platforms as one of the roots of evil. It's sad that it works so well, too. That's going to be hard to address. Maybe I'm going to have to break my own rule, for that, and seek government prohibition as "better than the alternative." Yeesh! I don't want to get somebody into the business of telling others what is good for them. That's like leaving a loaded handgun in dad's bedroom closet. The kids are guaranteed to find it! I'm so glad, lately, that the founders of the US had the sense to build a crippled government. I'm not entirely comfortable with thinking of "attention" as similar to opioids, but if the shoe fits....?

The first step is making sure that I own every piece of information about me. THERE IS NO PUBLIC INFORMATION. (Sorry about the caps. Bold-italics just didn't cut it.) The problem begins when "information" is captured/retained. I say: artifacts made from me belong to me, irreversibly. Let me go emotional to be clear: We once debated ownership of "my person." It took 5,000 years or so to make "ownership" of "the person" an inalienable right of every individual, rather than splitting hairs about it. Any artifact of "my person" ("personal," you might say) must also be considered inalienably mine. Once ownership of personal data is properly assigned, there are quite a number of existing laws to protect me and my data from indentured servitude, theft, or "taking," to say nothing of "ownership by another." I'm not even sure this is controversial. Only kinda new. These artifacts have only been fungible since actuaries got computers, so the topic simply hasn't come up, much, before. (To be precise, most of history's heinous abuses of personal data were perpetrated by governments. Peer-to-peer abuse is new.)

I do think the [differential privacy] and the [sorting out the ownership] problems are compatible. I agree with you that we should not let "perfect" become the enemy of "better than before." I just want to get the story straight, right from the jump.

ICYMI -- a bit of a presentation on using encrypted data (successfully) directly within ML modeling: https://www.youtube.com/watch?v=culuNbMPP0k&feature=youtu.be&t=3397