Hopefully this isn’t too much of theoretical question…
Is the upcoming AI world going to make our attempts at Privacy null and void? Given AI monitoring, listening, data collection and aggregation of personal data seems like it would be difficult if not impossible to prevent.
Is there any discussion going on about that coming risk and if any countermeasures actually exist?
Again, I hope this question isn’t too speculative or obscure paranoia.
AI doesn’t necessarily mean any more data collection than was already going on. This stuff has been around forever already it’s mainly the generative AI that’s recently had a big surge in popularity.
there’s at least the beginning stages of tools cropping up to help mitigate ai analysis of your internet traffic. what comes to mind is how mullvad is working on DAITA for their VPN. the essence of it is “padding all packets to the same size” and sending random dummy packets. time will tell if it’s effective. i personally believe as long as there’s privacy invasion going on, there will always be a group or groups of people who do what they can to fight it, mitigate it, etc.
Link to the mullvad site discussing DAITA:
Excellent. I was hoping for this type of analysis and discussion. Thank you.
Is the upcoming AI world going to make our attempts at Privacy null and void?
Kind of, it’ll start becoming more and more unavoidable to run into people who don’t care about their privacy and in turn will be hampering yours. Ring Cameras, Meta’s smart glass and shit like that. As I have been told on this forum that
“Anything that is not private, is Public”
I am assuming some people are really okay with it. With companies trying to make their products look as normal as possible, you can’t probably always notice that someone is wearing a smart glass or something like that.
Is there any discussion going on about that coming risk and if any countermeasures actually exist?
Not sure what discussion should take place, it’s not like these companies are releasing the data for anyone to audit on what they used to train their models. It’s becoming harder to do a reverse attack/jailbreak the models to get a glimpse of it.
AI doesn’t necessarily mean any more data collection than was already going on.
It does. The general purpose public facing models are shitty at best, even now. Now that it has “seen” all possible knowledge from humanity, it need very specific data to mimic things like humans do. The data collection needs to be more supervised, more precise, so yes, there is a high probability of “increasing” data collection. It’s not just for that, these tools “theoretically” can be served as attack vectors to diminish the effects of “privacy by obfuscation”.
Lets hope things like photostatic veil become real for improving privacy.
"There is not much discussion to be had on something as vague as this. "
I sure agree that it is vague at this point (at least to me). What I think I see coming is all of us being constantly surrounded by “passive” data collection devices that would end up making our attempts at reducing our exposure (Privacy steps), mute.
I was thinking that there would be some smart set of folks discussing this threat exposure and what they see as possible prevention steps, if at all possible.
I am so old in Technology as to have been in the industry before Internet/email began and we faced this weird thing we now call Spam. “How to stop this?” “What are viruses” “Why would someone do this?”, were all discussions of the day.
With Smart cars, Ring type cameras, Smart TV’s, AI Search engines all heading into the great data collection/learning mode, I was wondering if our current efforts to reduce our Privacy footprint need to modified.
I really like that the Mullvad team is at least discussing one component (AI Traffic Analysis).
I feel that actually this is the norm from the start and the Internet/Globalization introduced another level that humanity was not prepared to. I wonder if in some dystopian, not so distant future, we will have a huge division between those that are actively fighting for everyone’s privacy, no matter the nation that they come from, in a globe scale with some kind of cybernetic war trying to protect their beloveds that are against their principles.
New research from Amnesty International suggests OP is right,
AI will be shittier than the bespoke algorithms the intelligence agencies were using to monitor the internet firehose. Those algorithms will spit out actual groups with network connections to [evil country], vs an AI hallucinating groups.
The big problem is false positives. False negatives are already unlikely given all the information being vaccumed up, and all the law enforcement and intelligence agents on standby. False positive is expensive monitoring on some rich college kids larping as revolutionaries before they graduate and join Morgan Stanley. Or a woman trying to send money back home overseas to pay for her kids school.
Probably going to edit this it might not make sense, but i’ll post it here maybe i can get feedback if it’s stupid lmk:
China has had worse privacy than the US ever will long before the advent of LLMs. The problem is the corporations and governments, hell, governments could kill and oppress people long before guns. Now we have more guns then ever, but not the same level of oppression.
Technology is a tool, what matters is who and how its used.
The end of privacy is not when we have brainchips capable of reading our thoughts, it’s when we are forced to have brainchips reading our thoughts or everyone does it because it’s trendy.
Our immediate problem is government wanting too much power and control over population, like 1984 technology not Terminator.
that’s it.
so they create problems like poverty/immigration/terrorism/finding-lost-puppies/war/national-security/not-protecting-children etc. to divide people into supporting losing liberty in exchange which will lead to AI controlling aspects of lives, like checking what kids what on their phones instead of their parents
for me its more of political issue than science
I’m all-natural and completely reject technology anywhere on my body.
The thing is, they want you to see pretty things, but people don’t bother to look into it deeply.
That’s why I don’t follow “trends” and the like.
The problem is that someone will always abuse power and there are no effective regulations in place that are threatening or have any significant enough consequences to deter shitty things from happening.
AI has been analyzing our data for many years, but they’ve gotten a lot better lately. LLMs are newer and just another source of data for the AI. But the AIs still need our data.
I believe that the only thing that can save privacy, and the most important thing required to save it, is having the majority of people actually wanting it. Not just wanting it in concept but in practice. That means refusing to use services that violate our privacy, which is a sacrifice. It means publicly objecting to things like Flock cameras and demanding to get them taken down.
I know many people who disagree with mass surveillance in principal, but very few who are willing to take steps to prevent it. A typical conversation for me is like this (greatly simplified):
Me: FB gets you addicted to it, scrapes all of your data, and tries to predict what you’ll do next and tries to manipulate you. Done on a mass scale, this allows them to control society.
Other: Yes, you’re right. That’s terrible. They’re totally evil and should be stopped!
Me: Are you going to get off of FB?
Other: Well, no. If I did that I’d completely lose touch with a few of my friends that I haven’t seen in years.
It’s very difficult to convince others to take meaningful steps, perhaps because the objections seem mostly theoretical, or perhaps because they’re too addicted to the programs that they’re using.
That is in the ballpark.
No, you can run your own AI models locally. Granted, the kind of models you can run on consumer hardware are a step below SOTA cloud models, but they are surprisingly capable. And for things such as replacing an Alexa in the living room, they are fully there.
Not only this; you can also use cloud hosted models that execute in a sealed enclave, a la confer.to or tinfoil.sh.
I have Lots of Big Thoughts about the relative privacy/security tradeoffs of sealed computing vs local execution but too lazy to write them up. But assuming a reliable implementation, I think those are pretty good options (albeit open weight models and, as you say, a bit behind the frontier labs’).