[MVf4l]

1. Please ask BEFORE you collect.

2. You can't expect every user to know they are logged, or how it's affecting the user, or know how to disable/delete it, can you?

3. How can I verify that you did delete the data about me instead of just hiding from me for viewing it? Alphabet is not belong to public sector. So the simple answer is I can't. If you want me to trust you, don't use opt-out as default.

4. I'm sure you can tell the differences between those alternatives and Google products.

5. It's not that hard to respect some one's data. First, do not collect it! Second, if you have to collect it, tell the owner why! Third, delete it completely while requested.

6. Aggregated data collection and use without permissions adds potential risks to the society. (Cambridge Analytica)

Edit: And you guys are doing deep learning, that's gonna consume lot's of data. Duplex for example, you use anonymous phone call data to train it. The question is, where does that data even come from? I'd blacklist whoever collected the data, even it's collected anonymously.

[antpls]

> 6. Aggregated data collection and use without permissions adds potential risks to the society. (Cambridge Analytica)

Everything adds "potential risks". When you talk about risk, you have to give estimates of both the frequency and the criticity, and then compare to the potential benefits. Only then you have all the pieces to take an informed decision, according to your preferences.

[4684499]

How do you define benefits? Sacrifice one's privacy without his permission to make ten of others' life easier, would you call it beneficial? If so, let's rob the wealthy to aid the poor.

They can reduce the risks to a certain level if users were told how they are going to use the data and why before using it. Are they going to do that? No, because that increases the cost, which means less profit, which means shareholders won't agree.

So there comes law.

[antpls]

My point was : it is easy to throw a general sentence to make things look obvious and simple, but it doesn't really help the conversation. At some point, claims must be backed by data and methods.