Could AI expose us to a whole host of new privacy concerns and what should we do about it? Let's talk about it with Proton CEO Andy Yen in a conversation brought to you by Proton. Andy, great to see you again. Welcome back to the show.
Hey, Alex, great to be back. Yeah, I think a lot to dig into here. AI and privacy definitely a hot topic these days.
That's right. You know, we spoke a few years ago before the LLM craze took off and I really started thinking of you recently when I realized what was going on when I'm using applications like ChatGPT and Claude and Gemini and what was going on in the background. So, let me just give you an example. I recently found out that we have all been opted in to AI training when we use ChatGPT and Claude. And only if you go in and toggle off those settings can they not use the conversation that you're having with those chatbots to train future models. That kind of astounded me to be honest. I mean I was definitely having some, what I thought were, and I probably should know better at this point, what I thought were private conversations with these chatbots and then next thing I know the material can be used for training. So I'm curious to hear your perspective about that and what do you think it says about the state of digital privacy today that that can happen?
Well, actually it's worse than you say. It's not just it can be used for training. Even if they don't use it for training, they can still store it for a long time, actually forever. They can still hand it over to a government under subpoena. They could leak it if they were hacked in the future. So just because you've opted out of training doesn't mean you opted out of data collection. In fact, all the information is still there on Google, on OpenAI, on Anthropic's servers. So it's actually quite a bit worse than you imagine. Opting out is one thing but it doesn't mean that they don't collect it. So this is I think the first point I would make but to kind of your point, yeah, I think LLMs, to put it frankly, it's a privacy disaster. And what is LLM actually for most people? It's kind of like a more efficient way for humans to communicate with computers. This is all AI actually is, right? People say AI is this huge new thing. No, it's just a continuation of a trend past 50 years of being more efficient. But efficiency comes at a cost. A lot of people today in fact use AI almost as a replacement for search engine. It's like a conversational search engine. The difference is when you use search, Google knows what you're interested in, but they don't really know that much about you. When you're having a conversation with an agent, with the AI chat, and you're speaking to it for months and months and you have tons of data being collected, well now it knows the way you talk, the way that you converse, it has a much deeper insight into your personality. And this is why I think the information that AI has, like if you use Gemini instead of Google search, it is really supercharging Google's ability to understand who you are and it knows a lot more than if you were just using a search product. So you could think of it as Google's business model but now on steroids.
So let me tell you one more thing that I do to try to defend myself and you can tell me whether or not this is foolhardy. There is an option both in OpenAI's ChatGPT and Anthropic's Claude where you can toggle to an incognito conversation. Does that protect me at all?
Well, Google Chrome had incognito as well. But then if you've been reading the news, Google paid probably several billion in class action lawsuit settlements. And that basically showed that incognito wasn't as incognito as it was claimed. At the end of the day, you're kind of taking their word on whether it's incognito or not. And past history shows that when Google says something is incognito or anonymous or private, it generally hasn't been the case. And this is why I think anything that is any trust-based model of privacy has its limitations. So it's probably better than using the default setting as you say that allows training and all that. But I wouldn't consider it to be bulletproof given Google's track record here.
So I did upload my tax returns in this secret browser. That was a mistake. Yeah, the secret chatbot. Yeah, you probably shouldn't have done that. Maybe you put it in Proton Lumo. That would have been better, right?
Okay. Yes. So, Proton does have your own LLM called Lumo. We're going to talk about that in a minute, but basically your perspective here is when we're using these these bots,
Don't trust them. Don't trust them to protect our privacy. And that is scary because we tell you know, yes, they're more efficient, but I think you've already pointed out this. We tell these LLMs more than we tell any technology platform about ourselves by orders of magnitude.
Mhm. Yeah. Well, it's a bit like if you have a conversation with someone in real life, you will know them a lot more than you know someone that just looked at what you ordered on Amazon for instance. And as humans, we actually get most of the insights about people through conversations. So just imagine like you know you and me talking, right? Let's say we meet at a pub somewhere. I know so much more about you because I've met you in person and had a conversation with you versus if I was just, like, reading your LinkedIn or something like that. And this is the element that people miss is that in fact AIs are going to get better at discerning who we are than probably even a psychologist. And some people use these AIs as psychologists, in fact, right? I mean, there's an argument to be made that they will know us better than a friend would because we'll tell them things that we won't tell our friends. Exactly. Yes. And that's scary when you think about it. It's completely scary.
Definitely. Now, another issue that's come up around the AI world, but I think it's actually something that's been a pervasive concern across all tech platforms is the use of these platforms by kids. And we know that kids have been using them. The LLMs in particular, but also social media. There are currently trials going on about social media addiction and what social media leads to. And of course there's worries that as kids get more engaged with chatbots, who knows where they're going to end up. And so Proton, which again does encrypted email, still doing Docs, calendar, LLM, has done a survey of parents and asking them about the way that they feel about their kids' use of technology. And I thought it would just be a good opportunity since we're talking for you to share a little bit about what you learned and what the parents told you guys.
Yeah. Well, it's quite interesting because there are so many people who today just give kids phones and then kids have always also discovered AI. In fact, I would say the younger generation is maybe even better at using AI than some of us, right? So just give you some topline numbers: today around 70% of kids have access to a smartphone by the age of 10. This is seven in 10. And of those, actually around three-quarters are using Gmail, and that's a staggering number if you think about it. And a 10-year-old child doesn't really understand privacy, doesn't really understand what ChatGPT and OpenAI are doing with their data. They have no clue whatsoever. It's not really taught in school, but they're already on these platforms. And I think parents are starting to realize the risk that comes with this. So if you look at the survey that we ran, if you ask parents what they think about all this, well, 41% of parents say that if they could start over and do things differently with putting their children online, they actually would want to start over. And 60% would want to be able to erase all the information about the kid that today is on these big tech platforms. Now it's obviously a little bit too late now for many of these parents, but it shows that there's a huge amount of parental regret and it also leads a lot of anxiety. If you look at parents, around 80% are really concerned about their kids' online privacy. And these are massive numbers if you really think about it. And this is quite surprising because a lot of parents in the last couple years they've just sort of walked into this situation and are now starting to realize, oh, maybe we shouldn't have done that.
Right, and so but let me ask you because sometimes there'll be platforms like Facebook or or maybe Google as well that will say you have to be like x years old to use our platform. Do you certify that you are some are even saying like I think OpenAI has been saying recently that we have technology that can determine smartly whether you're a minor or not and then prevent you from -- I think the use case was using adult mode in ChatGPT. Do you trust those platforms with that sort of those guardrails or no?
I think the guardrails could work. And it's good that there are some guardrails in place, but ultimately the business model of these platforms is to exploit the data of our children. This is what they are all about. And in some sense the guardrails are running counter to their business interest. And we have seen historically that when there's this conflict of interest between what is good for the business and what is good for society, we know the choice that Zuckerberg is going to make. We know the choice that Google has made in these things. So I wouldn't really trust the guardrails.
Well, one of the there was this social media trial. I think it was a Meta executive that said we make the least amount of money on our young users. So that would maybe suggest that they don't wreck as much.
He didn't say we make no money, right? He said we make less money compared to parents, right? But it's about building habits. It's about getting children. What is social media? I view it literally not so much different than the neighborhood pot or crack dealer. It's there to get kids hooked and addicted and put a lifetime of dependency. And yes, children may not be your most lucrative audience today, but they do get older. They do emerge purchasing power and in 5, 10 years they become the source that you're going to look for profits. So in the long game, yes, they do want to get children hooked because that is how they're going to turn profits in the future. This is building out the customer base that's going to buy their crack 10 years from now.
Yeah. I don't know if I'm fully on board with the crack example, but I will say this. It's obvious that when you're a kid, you're not going to have the purchasing power that an adult would. So to come out and say it's some saintly thing that you don't make as much money on, let's say, under 18 as you would, you know, 25 to 40. Well, obviously not. Yes. But it's also a long game. It's about getting an entire generation completely addicted so that they can monetize them in five years as they get older. Exactly.
So how do you think we've talked about a couple of issues here, talked about LLM privacy, talked about kids' use of technology and the problems there? Any solutions that you might have to that? Well, fundamentally it's a business model problem. If your business model is exploitation of data and abuse of privacy to generate money, that conflict of interest is always going to exist. And what Proton has tried to do differently is have a business model where the only incentive we have is to protect user privacy. And this was really the motivation behind launching Lumo last year, which is the private AI. The way that we describe Lumo is it's the only AI where your conversation is actually private. And this is ensured through very strong encryption. In fact, we don't have any ability to go in and query past conversations. This is different from what OpenAI can do. This is different from what Google does. And this is because our philosophy is the best way to protect data is to not have it in the first place. And that is only possible because we have a business model that doesn't require us to get your data in order to make money. So it's really a business model problem. And that is also the solution. And this is why no matter what Google and OpenAI say about your privacy, they assume you don't have a financial incentive to protect it. And Proton is a very different business because that is the only incentive that we have. The only reason people pay us anything is because we have to protect privacy. And without that we wouldn't have a business. And this is different because it really aligns our interest with the interests of the customer. And this is in my view the foundation to a more ethical and more responsible internet.
Yeah. But maybe cigarettes is the better analogy, right? It's just you know it's bad for you, you keep doing it. You need an alternative, but it seems like there's no way out.
