Paul Rand: Have you ever wondered why it’s so easy to sign up for an online subscription service? But when you want to cancel, it seems nearly impossible or why certain buttons on a website seem designed to trick you into clicking them. Well, these aren’t just accidents of bad design. They’re deliberate strategies that are known as dark patterns.

Marshini Chetty: So, a dark pattern is basically when an interface is designed in such a way that it kind of either coerces, steers, or deceives you into making a choice that maybe you wouldn’t have made otherwise. And it’s usually designed in that way to benefit the service provider.

Paul Rand: That’s Marshini Chetty, professor of computer science at the University of Chicago, where she also directs the Amyoli Internet Research Lab.

Marshini Chetty: Kind of hidden influence. The whole idea is you’re not consciously aware of it and you can’t easily become consciously aware of it. And some researchers in the field have tried to unpack this and say that when you’re trying to persuade someone, they’re more consciously aware of what you’re trying to do. But with manipulation and with something like a dark pattern, you don’t even know that you’re being steered in a certain way.

Paul Rand: And even the most cautious people can be tricked into adding something into their cart or being told to click yes on a notification.

Marshini Chetty: I think in many cases, our research and other research has shown that companies are crossing that line. There’s outright deception or there’s manipulation that is harmful to consumers.

Paul Rand: But it’s not just dark patterns that we find ourselves up against. There’s also the question of how much of our data is being collected online.

Marshini Chetty: There’s something called the privacy paradox, which is although a lot of people, myself included, really want privacy, we sometimes act in a way our behaviors don’t necessarily reflect that. I’ll admit, like many of the listeners here, that sometimes when I see a consent notification, I just click accept because I want to get on with my business. Is that the right thing to do? Not necessarily. Some people are much more discerning and do not want their data collected about them.

Paul Rand: From the University of Chicago Podcast Network, welcome to Big Brains, where we explore the groundbreaking ideas and the discoveries that are changing our world. I’m your host, Paul Rand. Join me as we meet the minds behind the breakthroughs. On today’s episode inside the dark patterns of the internet and what should be done to keep us safer online. If you appreciate hearing from the experts in research and science who are shaping our world, please consider subscribing to the Big Brains YouTube channel. It ensures that you never miss an episode and it helps these important conversations reach a broader audience of learners. Bold and curious. These words describe the guests that you hear on Big Brains who are shaping the future of research. The University of Chicago is launching a new campaign called Chicago Minds to further advance this kind of groundbreaking work and expand the university’s global impact.

Learn more at chicagominds.uchicago.edu.

Paul Rand: I think just about every one of us is operating online and many of us, of course, feel somewhat vulnerable, a little bit uneasy. And you say that there’s a gut feeling for this and you call it a dark pattern.

What is a dark pattern?

Marshini Chetty: So a dark pattern is basically when an interface is designed in such a way that it kind of either coerces, steers, or deceives you into making a choice that maybe you wouldn’t have made otherwise if you had been kind of capable of selecting other choices or if you had all the information in front of you and it’s usually designed in that way to benefit the service provider.

Paul Rand: If you could give us a bit of a classic example that most people would recognize that they have fallen for.

Marshini Chetty: So a classic example is what we call a roach motel. So this is basically literally those little traps you have for roaches, they can check in very easily, but it’s very hard for them to check out. So it’s basically like maybe you signed up for a subscription but then you’re trying to cancel and it’s almost impossible to figure out how to cancel. That’s a very common one. Or maybe you’re shopping online and you see all these countdown timers and it’s kind of creating this sense of urgency like, "Oh, I better get this deal now or I’m going to miss out. " That’s another example of a dark pattern that’s pretty common.

Paul Rand: And that may or may not be accurate. Probably it’s not accurate, it’s just a manipulation technique.

Marshini Chetty: Well, occasionally it isn’t accurate. So sometimes in our research, we found that those countdown timers sometimes reset or the deal’s still valid even when the timer expires. So I’m not saying all of them are inaccurate, but we definitely have found cases of outright deception with those timers.

Paul Rand: Okay. As we think about this, there’s some that could argue that it’s manipulations. Other could say, "Well, no, it’s persuasion." How do you look at it?

Marshini Chetty: Well, so actually the reason why it’s manipulation as opposed to persuasion is because it’s kind of hidden influence. The whole idea is you’re not consciously aware of it and you can’t easily become consciously aware of it. And some researchers in the field have tried to unpack this and say that when you’re trying to persuade someone, they’re more consciously aware of what you’re trying to do. But with manipulation and with something like a dark pattern, you don’t even know that you’re being steered in a certain way. And so that’s why I would argue that it’s not the same as persuasion. It’s something different.

Paul Rand: One of the tactics that you talk about is visual interference. Help me understand what that is.

What is visual interference?

Marshini Chetty: Basically, it’s talking about how the design on the screen actually looks. What are the visuals? What does the font look like? What does the color look like? What are the size of the buttons and so on? So one example might be maybe you go and see there’s a consent notification and you have to accept or decline something. Except maybe the accept button is really big, really colorful. Maybe it’s a bright red and maybe the decline button is a small link that says, no thanks. And so that’s kind of visually interfering with the information and how it’s presented to you. And from what we know in human computer interaction, that’s my field, it’s much easier to click on that accept button than to find that sort of link that says no thanks. And that’s an example of visual interference.

Paul Rand: I got it. Totally. Well, another one that folks probably are aware of is multi-step hurdles. Tell me about those.

Marshini Chetty: So actually there is a name for it in our community. It’s called obstruction. And this is basically when people are adding extra steps to things. So think of it as imagine if you and I were running a race, like a hundred meter sprint and my lane was all clear, but in your lane there’s like a pit of snakes you have to go through. There’s a tall ladder, there’s a whole 50 meter detour before you get to the end. Who’s going to win? Who’s going to beat Usain Bolt?

It’s probably going to be me. And it’s the same online. Sometimes service providers are adding extra steps and making something like unsubscribing or canceling a subscription much harder by putting in a lot of barriers or hurdles so that you can’t quite easily make it to the finish line.

Paul Rand: Okay. And I’m assuming the expectation or the hope anyway is that you get frustrated and give up.

Marshini Chetty: That is the hope. The hope is that it’ll kind of weigh you down and you might just abandon and decide, "Hey, Marshini can win this race. I don’t think I’m going to make it to the end."

Paul Rand: Okay. And imagine as we get into the world of AI that’s evolving and changing your field and probably keeping it very interesting.

Marshini Chetty: Definitely. I mean, with AI right now, there are a lot more different kinds of manipulations that can occur. When you are talking to an AI, you’re much more likely to disclose information because it feels very natural, like you’re having a conversation, more sensitive information. And then even when you’re talking to an AI, you may not even disclose information to it directly, but it can infer a lot about you. So things like that are changing the way we think about, say, manipulation and deception and coercion and also how we need to design interfaces in a way so that people are kind of protected, their data is protected. They know what they’re getting into.

Paul Rand: Okay. A while ago you committed or created an automated tool and it crawls what, 11,000 different shopping sites and you learned a lot from going through this exercise. Can you start us down that path?

Marshini Chetty: So we set out to see, can we actually get some evidence that these patterns and these designs are being used? And we wanted to look at shopping because there’s a consequence. You’re buying something, you could lose money if you buy something you didn’t mean to, for example. And so we created what’s called a web scraper, which visited 11,000 of the most popular websites at the time and we collected a subset of the different pages as you’re buying a product and checking out and we tried to examine them to see if they had instances of these deceptive designs. And what we found at the time was that 11% of them had some instance or some sort of dark pattern and that was quite substantial because we hadn’t crawled all of the sites. So the number in practice is probably a lot higher than that. And it suggested to us that there is evidence that especially at the time and even now that many, many service providers and companies are actually using these deceptive designs that manipulate people.

Paul Rand: Are there other tricks that you’ve seen maybe sneaking items into a cart, for example, or some other types of things?

Marshini Chetty: Right. So sneaking out into a cart is another thing we found in that particular study where say you might be buying some flowers for someone on a florist website and then they sort of add in some greeting cards into your cart. So you didn’t intend to put that in there. You may not even notice it’s in your basket and then you check out and you’ve bought something that you didn’t mean to. So again, you’re spending more than you intended. And so that’s kind of the sneaking idea of trying to get it past you. Sometimes I can even just refer to sneaking information past you where you might be subscribing to something, but because of the way the information was displayed or maybe it’s in tiny font at the bottom and buried terms and conditions that you may not even know that you signed up for, say, a subscription service.

Paul Rand: It’s interesting. The countdown timer arguably could be called manipulation. Sneaking something into your cart actually sounds like theft.

Marshini Chetty: Well, again, I think context matters. So I’m talking about these things in abstract terms, but I think even some sites that are fairly innocuous are using arguably the same kind of tactics to get you to agree or disagree to maybe even just sign up for a newsletter. So with all of these kind of manipulative designs, you really need to examine the context to understand what is happening, what is the user being manipulated into doing and then also is there some sort of harm? Is someone losing money? Are they giving up more data than they intended to? Are they suffering some sort of consequence that is problematic? And in some cases you might be able to say like, yes, this countdown timer was an outright lie. The deal was still valid when it expired. But in other cases, you would have to look at the context to really understand whether or not this is crossing the line into deception, whether it is manipulative and so on.

Paul Rand: Okay. The other ones that you spent some time talking about were high demand notifications, and some of those turned out to not be accurate, if not just outright fake.

Marshini Chetty: Right. So high demand notifications, think and imagine a site where everything’s like, this item is selling out fast. We found a couple of websites where literally every single item on the website was selling out fast. It cannot be that your whole inventory is selling out fast. So I think what our research has shown and what other research in general has shown is that when people are throwing these dark patterns at you, it’s very hard for you, even if you know that they exist, for you not to be manipulated by them. So even I mentioned that consent notification, you see an accept and a decline. If we hide that decline button and put it under another button that says learn more,

Marshini Chetty: More people are going to accept. So just the design shifting, even if you intend to decline, it’s going to make it harder for you to actually say no. So even if you’re aware of these doc patterns, they can very easily manipulate you regardless.

Paul Rand: Okay, interesting. The team that you work with also studied and looked at some of the source code behind these prompts and probably, I guess not surprisingly, but some of these, lack of a word of better word, tricks are sold by third party companies as a product.

Marshini Chetty: Yeah. So we sort of examined the ecosystem at the time of like, if I want to set up a shopping website, how do I do this? I might use a service called Shopify, which will help me to quickly create a shopping cart and all the components I need to sell stuff. And we found that a lot of those third party services are actually making it easier for people to use dark patterns. So at the time they would make it easier to have something like creating a sense of scarcity or a sense of urgency. And even now a lot of people are using AI to create websites. They’re doing what’s called vibe coding AI, which is instead of learning how to code, you kind of chat with the chatbot and you get it to create a website. And a lot of those actually contain dark patterns as well.

And so basically it means that it’s very easy to perpetuate these kinds of manipulations because people are using services that create them or enable them and that makes it harder to keep to a design that is less problematic.

Paul Rand: Okay. One of the things that just about, again, most people are experiencing this idea that every time we open a new browser, there’s a new cookie banner. Tell us what these cookie banners are. What are we actually being asked and how should we think about those if we press okay or say limit these to critical only?

What is the General Data Protection Regulation?

Marshini Chetty: So basically in the European Union, they decided that if a company’s collecting data about you, they need to give you some notice and allow you to consent to that data collection. And so they created this legislation called the General Data Protection Regulation or GDPR. And as a result of that, companies then had to start informing you if they were going to collect your data. And before you use the website, you actually have to see or allow them to do that and consent to it. And so that’s what the consent notification is saying is like, “Hey, you’re using our site.”

Do you allow us to consent to using your data? And if so, then you can continue using the website." The issue is that it’s often hard for people to understand exactly what they’re being asked, like you said. And then a lot of the companies and consent notifications, many researchers have shown that they use dark patterns. And so it makes it much harder for you to actually decline. I mentioned that you could have a consent notification that says accept, and then the decline is hidden under learn more button. Or it might be that to decline everything, you have to opt out of a whole bunch of different things individually. And so that just makes it really hard for you to actually limit or control what data is collected about you.

Paul Rand: Okay. We’re going to get into what consumers can do about this, but if you’re faced with a cookie choice when you come in, what is your typical response and how do you handle it?

Marshini Chetty: I mean, let me start by saying there’s something called the privacy paradox, which is although a lot of people, myself included, really want privacy, sometimes act in a way, our behaviors don’t necessarily reflect that. And so that’s called a privacy paradox. I’ll admit, like many of the listeners here that sometimes when I see a consent notification, I just click accept because I want to get on with my business. Is that the right thing to do? Not necessarily. Some people are much more discerning and do not want their data collected about them. And so I’d say it would be good to take the time to actually read what is being collected about you and then make a more informed choice about whether you want that data to be collected. Actually here in the US, we’re really lucky California created a law called the California Consumer Privacy Act. And as part of that law, they actually make it easy for you to opt out of the sale and sharing of your personal data amongst other rights. And so they made it so that if you have a certain amount of customers in California, you have to allow people to opt out of the sale of sharing of personal data and the website has to have a literal button that has certain wording to make it easy to find for you to do that.

And so a lot of the other states get the benefit because if companies are doing it for their Californian customers, they tend to do it for everyone else. We’ve actually studied that and found that that is the case. But again, a lot of the companies also use dark patterns to make it harder for you to opt out. And we’ve also studied that and found that that’s the case too.

Paul Rand: Now I imagine that these companies, these marketers would like to argue to us that there is consumer benefit in allowing us ourselves to be tracked. What argument would they be making?

Marshini Chetty: Well, I mean, I always believe that nothing is totally wrong or nothing is totally bad or good. You have to, again, look at the context. I’d say a company could argue that if I’m collecting data about you, Paul, maybe if I’m trying to sell you something, I could personalize those ads for you. I could get you something that’s more tailored to you and that doesn’t sound like a bad thing. So companies could argue that this data is useful for them to optimize their service to make sure that it’s tailored to individual customers and to give you the best service that they can. So it’s not always a bad thing that they’re wanting to collect the data. But on the other hand, maybe you don’t want that data collected about you and maybe that data in collection with other data could be aggregated and then tell companies aspects about yourself and your persona that maybe you’re not comfortable sharing.

Paul Rand: Absolutely. Yeah, absolutely spot on. So as you think about this, there’s a question of legality, but then there’s a question of ethics. Is what they’re doing ethical?

Marshini Chetty: Again, I would say it depends on the context.

It depends on the context. It also depends on what kind of harm you’re talking about. Is someone losing money? Did they give up more data than they intended to? Are they getting on Instagram and finding themselves four hours later watching too many reels because of autoplay and infinite scroll? So it kind of depends whether or not companies are crossing an ethical line. You have to look at what are we talking about specifically? I think in many cases, our research and other research has shown that companies are crossing that line. There’s outright deception or there’s manipulation that is harmful to consumers. In other cases, maybe not so much. And so it’s hard to just say blanket across the board that people are acting unethically or even if they’re doing it in an intentional manner, but in many cases they are.

Paul Rand: Okay. There seems to be in some circles, and I think most folks have at least thought about, if not enacted, putting themselves on the do not call list. We could make an easy argument of how effective or ineffective that actually is, but is there some way where there is a blanket opt-out of this type of thing that you can do on your computers?

Marshini Chetty: So actually for the data collection, I mentioned the California Consumer Privacy Act. They actually also enable something called global privacy controls, which is basically in your browser, you can get your browser to send out a signal to websites to say, "Do not collect my personal data." And websites are supposed to, again, if they have a certain base in California, respect those signals and just do that automatically without you having to actually go and say it specifically. So people are trying to create measures or put measures in place so that can happen more easily for consumers, but we’re not at the point where everyone is implementing that or everyone even knows what that is or how to enable it. And so we still have a long way to go before it’s really easy for us to manage what data is collected about us and how so and what it’s used for.

Paul Rand: Okay. We’ll talk about progress against that. One of the other things that has always somewhat dumbfounded in me is how my contact information gets picked up by other companies. I ended up years ago signing up for a Comcast email account and it actually became unusable because they sold my name and contact information so much that there was nothing but junk mail in my mailbox and I finally had to just quit using it. How does that happen?

Marshini Chetty: Well, again, it sort of depends. If we bring it back to the case of dark patterns, sometimes it can happen when you’re signing up for something and they use what’s called a trick question and like a bad default. So you can imagine you’re signing up for something, they have a couple of checkboxes with small font and you have to really read carefully. And it might say something like, "By unchecking this box, you are opting out of newsletters." So it’s something that goes counter to what you would expect, like you’re checking to opt in. And so by using trick questions, it might be that you’re not fully informed about what you’re actually doing. And it could be that the company is then getting your permission, even though you didn’t intend to give it to actually be sending you junk mail and so on.

Paul Rand: Okay. Are there also though, I’ve gotten ads for things that I just can’t imagine what the connection is. So I’m assuming that they just sold my information. Is that possible or am I imagining that?

Marshini Chetty: No, that is possible. So people are collecting data about you for many reasons and a lot of the time they’re collecting it to sell to third parties, to other companies who again can monetize that data, use it for profiling different people to help with advertising and so on. And so it’s totally possible that someone else might have that data. Sometimes we also have other ways that this comes about. So it might be that people are tracking you across devices, maybe you search for something on your phone, but they also know your laptop’s in the vicinity, or maybe you have someone in your household that then sees an ad for something that’s related to something you search. It’s because we can kind of place all of those devices on the same network and use that information to be able to serve ads to different people that are related. So there are many different reasons why that could be happening to you.

Paul Rand: Okay. I have gone through, and I’m willing to bet others have, of desperately trying to look to the bottom of these emails and unsubscribe and even spending time do it. I did this with Comcast before I finally just gave up.

Marshini Chetty: Some email clients actually make this easier by putting an unsubscribe button closer to the top of the email message. I know my Gmail client does that and so it can make it easier for you to unsubscribe. Unfortunately, again, bringing it back to dark patterns and manipulative designs, sometimes when things are not standardized, like how do I cancel a subscription very easily? So it’s not a Roach Motel where it’s easy for me to sign up but hard to cancel. If it’s not standardized, then it makes it really hard for you to do the same thing as the unsubscribed with email. And actually the Federal Trade Commission, which is basically in charge of consumer protections online, they’re trying to enact a rule that’s called click to cancel, which would make it much easier for you to cancel a subscription within a matter of a few clicks. This was kind of held up and so now they’re sort of redoing everything and getting comments on the rule again, but this is something that they’re working on to make it as easy to cancel subscriptions and so on as it is to unsubscribe from an email.

Paul Rand: This type of manipulation doesn’t just apply to online advertising. It also applies to even deleting social media accounts that we’ve signed up for.

Marshini Chetty: So we studied this in the past where we were trying to see if there were dark patterns when you’re trying to delete a social media account. Again, for a lot of social media companies, we don’t pay to join them. Why? Because as many people say, you are the product. They’re collecting a lot of information about you on social media that they can then sell to advertisers or use to target ads at you. And so we want to know, well, how easily did they let go of you as a customer? And we found that there’s a lot of cases where there are dark patterns that prevent you from easily deleting an account. They might be kind of like the scenario of us running the race with a lot of hurdles, a lot of companies put a lot of hurdles to make sure that you’re not actually going to make it through that process.

They make it so that you want to abandon and you’ll say, "Ah, this is taking too long. Maybe I won’t cancel my LinkedIn account after all. Or they may play on your emotions. They may say like, "Paul, are you really sure you want to cancel this? " You’ll lose all your connections, you’ll lose all your benefits. And so again, they sort of play on your emotions so that you’re like, "Again, people are very loss averse."

Paul Rand:

Absolutely.

Marshini Chetty:

And so you might think twice and say, "Well, no, actually I don’t want to lose anything. I think I’m sticking around."

Paul Rand: The other one that of course comes up is, well, why don’t you pause it or take a break?

Marshini Chetty: So yeah, we’ve studied this and we found that kind of like I was saying with unsubscribe, it’s standard. Everyone probably knows. Listeners know unsubscribe. We know what that means. We know where to look for it for email. When it comes to deleting an account, all the different companies that we study, the social media companies, they use different words. So some people say you can pause or deactivate or delete or delete immediately. And so because it’s not standardized, there’s no real conception of what it means. If I pause, does that mean you’re going to stop collecting my data? Does that mean I’m just taking a break? If I deactivate similarly, is that the same as deleting? On some sites it is, but because of the different terminology, it makes it really hard for you to know what is actually happening here. And as researchers, all we can see and all we can study is what the companies are saying.

It’s much harder for us to know even if you did delete your account, are they actually deleting your data on the backend? Are they doing what they say they’re doing? If they’re saying, Paul, we’re going to delete your data within 30 days. As a researcher, unless I’m in that company, there’s no way for me to audit whether they actually did do that.

Paul Rand: Absolutely. I imagine that there are lots of discussions around regulation and arguments against why it’s needed. And I’m assuming one of the arguments goes along the lines of, well, if a consumer understands what these dark patterns are, they can just avoid it. But you’ve done research and if that’s actually a reasonable option, what did you find?

Marshini Chetty: So I did some research with some really cool legal scholars at Northwestern and also at UChicago. And what we did is we created this streaming video site where we tried to get people to sign up and we gave some of the people a goal of protecting their privacy and some just saying sign up as you usually would. And we threw a bunch of privacy related dark patterns, so dark patterns that asked them to sign up for targeted advertising so we could personalize ads and so on. And then we measured how do people respond in these different conditions. And even when people have the privacy goal of trying to protect their privacy and choosing privacy preserving options, when we threw dark patterns at them, it definitely made it much harder for them to actually end up with privacy preserving options.

So what it tells us is that even if you know or even if you have the goal and intention of doing a certain thing, when these dark patterns or manipulations are put in front of you, that it is very likely that your behavior is going to get modified and it’s very hard to defend against them. It may be not as bad as if they were ... If there are no dark patterns and you didn’t have that goal, obviously that’s the best case. If there are dark patterns and you had that goal, maybe it’s a little bit less, but it’s still not as low as if you didn’t have dark patterns at all.

Paul Rand: Okay. But I guess if even the most privacy conscious users end up falling for these, does that awareness, it still holds up as a defense, I guess, right? 

Marshini Chetty: Awareness This is one thing, but like I said, it’s kind of like running that race again. If we’re running that race and you have all these obstacles in your way and I don’t, it’s very clear what the outcome’s going to be. And so awareness helps you. Even though you know you have to go through that pit of snakes and climb the ladder and go on the detour, you still have to do all those things. And so it’s still going to affect your time running the race as opposed to my time. So the awareness is not enough. It might help you a little bit. Maybe you’ll be a little bit faster because you’ve figured out how to dodge snakes, but at the end of the day, it’s not enough for you not to fall prey to them at all. And so what we would argue and what we did in that paper is to say that just saying that people will become aware of dark patterns, they can just avoid them, that’s not really the case. A lot of these designs, people will still be taken in by them even if they’re aware of them.

Paul Rand: Okay. One of the things that at least to me always is quite wonderful to see is when research is done and it has a material impact on society on policy or others are finding this is not something that’s being done in an academic circle and staying in a journal, it’s having an impact in the world. You’ve been very successful in that realm of having your work make an impact. Talk to us about how you’ve done that.

Marshini Chetty: I think I’ve just been very fortunate that I’ve always had an interest in policy and impacting policy and providing evidence to people who could take say an enforcement action against a company who has manipulative practices. I really like doing that. I like having research that’s actually going to benefit other people and not just stay as a research paper. And so I’ve just been fortunate to have had connections with people who have found the work useful and then used it in terms of informing regulation or informing cases against companies. I’ve also had the privilege of serving as an expert witness in a case against Amazon for the Federal Trade Commission. And that was just so rewarding because it was.

Paul Rand: How much was that finding for?

Marshini Chetty: So that was the biggest settlement or one of the biggest settlements that the Federal Trade Commission has had against a company. It was $2.5 billion. And a lot of that is actually going back to consumers. They can get $50 refunds if they feel like they were unintentionally subscribed to Prime or unable to cancel Prime very easily.

Paul Rand: We spent a little time earlier talking about the world of AI. I imagine that this is really turning this world of computer human interface dark patterns on its head. Give us a big picture of how AI is changing the game that we’ve been spending all this time talking about.

Marshini Chetty: Right. So in terms of manipulation, AI is very interesting, especially if you think of something like a chatbot, because a lot of the earlier chatbots were very sycophantic, so they were just like, “Oh, Paul, you’re the best.” It would kind of make you want to disclose more because this is almost like your buddy. And so like I was saying earlier, you can basically disclose more data. AI can also make a lot of inferences about you even if you haven’t disclosed directly. And so there’s a lot of ways that AI can learn a lot about you. If I showed you my Claude chats, you would know a lot more about me how I think than if you saw that data of me navigating the web and indirect clicks and what I’m purchasing and so on. So it can build up a very detailed profile about you. And now if there’s advertisements in AI and so on, that creates a lot of potential for things to go wrong and manipulation to occur. And so I think it’s a really interesting area. Researchers are starting to look at that and starting to try to understand how manipulations occur. Also related to just, we were talking about data collection. We are studying right now, how easy is it for you to opt out of having your data used for model training and so on? Do you even understand where your data is going? For example, maybe you’re using something like you’re creating a PowerPoint and that’s Microsoft’s product, but then it has Copilot in there. Okay, maybe that’s also Microsoft. Well, what if there’s a Claude plugin and where’s your data going then? Is it going to Anthropic? It becomes very, very, very confusing. Already we talked about how your data might be going to a site and then also to a third party. Now with AI, that ecosystem is even more confusing and to try to unravel that for a consumer or even to try to give them some sort of promise that if they are opting out of certain data collection or if they’ve deleted a chat that it’s not going to be in the training data, that is also another area of research in computer science.

How do you help people unlearn training data? I’m kind of just excited about, I mentioned that we’re trying to figure out how to help people understand what using AI means for your data collection and how to make sure that you have that privacy that we spoke about.