Right, but the context of the article is the Signal messaging app. I agree with everything you’re saying, I just don’t find it relevant to this particular scenario.
I don’t see anything in the quoted text that says they exclusively consider harm contained within customers’ messages. It’s vague enough that it could be actually thorough. Vendor lock-in in particular is very relevant for Signal’s main product, as they don’t do interoperability.
I feel like we keep missing each other.
The quoted text is from an article co-written by the new president of Signal expressing her views on what a given platform’s responsibilities are. She then became the president of Signal, so given the view quoted from the article she co-wrote, how would one “monitor and mitigate the harm done by the product” if that product is Signal and if the integrity of e2e encryption, and therefore privacy, is to be maintained?
By investigating possible harms done and fixing them. And vendor lock-in is a harm done by Signal IMO, and in no way is related to messaging, nor does fixing it require inspecting messages. Unless you disagree, then I wonder fixing what harm you see that Signal could be doing requires undermining e2e.
How do you attribute harms to Signal if you can’t see the message contents?
I don’t care for vendor lock-in myself, but the new Signal president hasn’t spoken about that that I’ve heard. I’m only talking about what was stated in the article; I haven’t read anything else about her.
Unless you give an example of a “harm” you’re thinking, I don’t think I’m going to understand what you’re talking about. Vendor lock-in is clearly a harm because it makes people unable to choose a better solution, and Signal is clearly doing that by disallowing third-party clients. Here, I attributed a harm to Signal without seeing any message contents. If that’s not an example of attributed harm, then I’m afraid I have no idea what you’re talking about.
2 Likes
I don’t have an example. I’m just going off what the article says. Given the context of that article, I assume “harm” would be some sort of harassment or “hate” or bullying or the like since the title of the article is “An open letter to trans* Netflix employees.” I think its a reasonable assumption, so let’s go with that as my example of “harm” that needs to be “monitor[ed] and mitigate[d]”.
Thanks for clarifying your concept. I’ll bite. You don’t need message content to evaluate harm. Take their “no account without phone number” policy: that one is harmful to privacy (especially that of trans people) in places where you can only get a SIM card by disclosing your personal data. There you go, a harm attributed to Signal, related specifically to the implied context of the article, without reading any messages.
There are ways to evaluate harm without reading messages. Metadata is what matters, after all. If you have the social graph, you can correlate public drama with changes in your network (not sure how much metadata Signal has access to): hate towards group X correlates with some nodes being spammed? Your platform is probably facilitating some brigading. It might be a good time to introduce UI or protocol changes.
Or, like I mentioned, have a group of volunteers reporting how often they encounter bullying, under what circumstances, and for example find ways to give them tools to defend, or whatever. Still no reading messages.
Or you could ask users to report nasties and ban those nasties from connecting to your servers. No encryption gets subverted in the process 
By assuming e2e will be undermined, you’re taking an uncharitable interpretation of what was written.
2 Likes
Ah good. I appreciate the responses.
However (heh),
Those examples all depend on either people being completely truthful in their reporting (which cannot be enforced) or making educated guesses. I’m actually OK with the latter (like with your nodes being spammed example), but, at the risk of being cynical, not the former. Besides being unable to enforce truth (without verifying the claim by looking at messages), there can’t be any consequence for false reporting. For example, I could report you for bullying me. They would either take me at my word and punish you in whatever way they’ve decided to punish such offenders, or… well, there’s nothing else they can do. You can’t plead your case without giving up your messages, and, thanks to MeToo, I won’t have to prove my assertion. Without compromising e2e, the only two outcomes are either you get punished or nothing happens, and “nothing happens” would be unacceptable.
Now I say all that, but it sort of goes aside the point you were making. So let me finish by saying you’re right, I hadn’t thought of those. I appreciate your insight (and your patience in letting me fully explain myself).
I don’t claim that my examples of intervention are any good, just that there’s place for different approaches,
Being perfectly truthful isn’t necessary. We invented statistics to suss out information from noisy data – inaccurate or otherwise. I think this gives plenty of space to maneuver for a person in power, given noble intentions.
2 Likes
I gotcha. I was just offering my viewpoint, not trying to assert it.
Yeah, I’m sure there are metrics or other data that could be useful in that regard that I haven’t thought of, instead of just taking every complaint on a case-by-case basis. But then again, how much bullying would be required to get a good dataset?
I share your concern, thanks for sharing it.
I wonder what on earth happened to free speech? I prefer to call it free listening, because shutting out 14 billion ears is a much more precise description that shutting out 1 mouth. Way too many people have been lured into accepting all kinds of harm to free speech on reasons of ethics - ethics that always gets defined, not by you, not by me, not by world pop, but by a little elitist group in which we have no say.
Well, and the “AI Ethics” is just crazy. Me being a highly trained philosopher I do see your concerns. This does not bode very well for Signal.
1 Like
Agreed. I am more than a little disappointed that this isn’t the obvious response coming from more than just you and me. You would think people who have certain priorities that motivate them to buy a L5 would also be motivated to not use centralized honeypots. Even if their intentions are pure in the beginning, the centralized nature makes it obvious that the eventual result will be the same as what reddit or twitter turned into. The central matrix.org server has similar risks, but if pushed, the protocol based design makes it easy to switch to a different server.
Any communication method that is central server based rather than protocol based is an obvious problem as many of us have been whining about since the 90’s. Its about time people wake up and step out of the mainstream.
5 Likes
Some of them, but not all. For example I’m sure there are those who consider having only free software running on the phone itself as the main thing they want to achieve, and Signal is okay from that point of view.
3 Likes