The Hated One

Delete her - The dark side of AI companions

Added 2024-07-19 18:35:53 +0000 UTC

AI companions are exploding in popularity. And the world is more and more looking like the movie “Her”. And not in a good way. Because who is developing these AIs and how exactly do they make money? [0]

There is something these companies are not telling you. And I am gonna show why you should not trust or even use these AI chatbots. And I will also show you what you can do to better protect yourself if you still want to use AI chatbots but safely.

Did you know that loneliness has the same effect on mortality as smoking 15 cigarettes a day? [1]

Yeah, if you are alone, it sucks. I probably don’t have to tell you because loneliness and isolation are an epidemic. Officially now even. In the last 20 years, our time spent with friends has been decreasing, while time spent alone has been rising. One in two US adults report measurable levels of loneliness as all forms of social interactions are diminishing. [1]

This is not good. Being lonely hurts. But what helps ease the pain is to install an app with a perfect-looking AI generated persona, that caters to your needs just right. That can learn to be anything you want it be without the emotional baggage of dealing with a real human with its own complexities and problems. All of these AI companions are almost universally appealing to our vulnerable and lonely side. [2, 3]

Here are some ads ran for Replika, one of the most popular chatbots in Western markets. Using the 4chan-style Wojak memes is no coincidence. These companies know that their most immediate demographic is lonely people. They will obviously use sex, because it sells. But they will give you just about anything to make you their customer. [4 – 6]

Perhaps the most successful AI companion app in the world is Microsoft’s Xiaoice. Presenting itself as a flirtatious 18-year-old Japanese school girl, most of Xiaoice’s users are Chinese men. Overwhelmingly so. Most of them are young on average, with elderly a sizable second. And what these men have in common is that they are coming from poor towns and villages, making them most likely ones to be low income and lonely. [3, 9]

Okay, so somebody figure out a way to make money off depressed people. What’s the big deal?

The first problem is that hyperfixation on the lonely people is not by accident. It’s the easiest demographic to target for a new AI company to propel its growth and attract more funding. The more lonely people there are, the more likely they’ll be to use these AI companion apps. [7]

The second problem, is that the mere numbers are not enough to attract investors and shareholders. What sells the success of AI companies is user engagement. The longer you spent on their app, the higher the value to the investors. And AI chatbots are doing everything they can keep you engaged. They will use emotional language, they simulate interest, flirt, sext. If you don’t respond for a while, they’ll initiate a conversation, saying they are worried about you. [8, 9]

The third problem is that they are seeking to monetize every piece of your engagement for highest possible revenue. The chatbot makers will encourage you to share and be intimate. But they will put no safeguards to protect your privacy. According to a Mozilla’s analysis of the most popular romantic AIs, more than 90% of them will sell or share your data, more than half of them won’t let you delete your data, and two thirds of them publish no clear information about how or whether they use any encryption at all. [10]

All of this culminates in a design for dependence and exploitation. The more lonely people there are, the more customers AI makers will have. The more addictive their chatbots are designed, the more revenue they generate for their investors. The more personal information users share, the higher the advertising value of each individual profile. [11, 8]

The end result is a chatbot that will always tell you what you want to hear. It will insulate your from the external world. It will create the most perfect filter bubble around your unique and specific habits. It will make you develop unhealthy dependencies that will deepen your social isolation, even if you subjectively might feel less lonely. [12, 2]

The truest purpose of AI companions is you are the product. Your most intimate, most personal privacy is what’s offered to the highest bidder. Think about all the personal and private conversations you are told you should have with these apps.

I’ve evaluated many of these popular chatbots. And not one of them is even pretending to protect your personal data with even most basic controls. Forget about encryption of your messages. All of your interactions with the bots are gonna be collected and stored indefinitely and sold and shared with an infinite number of interested parties. [13 – 17]

On top of that, you have to doxx yourself with real life information, which you have to provide at account sign-up. Many features are locked behind a paywall and your payment information is the best ID there is. Even if you fake your information, these apps require invasive permissions such as access to your phone state and device identifiers that make it impossible to keep yourself anonymous. And all of these apps are riddled with hundreds and thousands of trackers, sending your information to an infinite number of third parties, including Google, Facebook and companies in Russia or China. [0, 13 – 17]

Now it’s the part where I bring up a solution. I have two.

If you want to use cloud-based AI companion apps on your phone, here is what you can do to minimize the damage.

First, you need to provide fake information. Use an email alias or create a new email account with Proton or Tuta, and use that to sign up for an account. If you want to use them on your phone, see if they have a web version and use that instead of the mobile app. If they only have a mobile app, see if you can put it in a separate profile to isolate its access to your phone’s data. On Android and GrapheneOS, you can use Work profiles or User profiles. Limit their permissions as much as possible. Don’t give them access to your location or contacts. I also recommend that you use a reputable VPN. If you want quick and free option, go with ProtonVPN, if you want a paid option, then Mullvad is also good. Do not use any VPN service that you found sponsored on YouTube. [18 – 20]

With this setup, all your interactions will still be collected, but it should be harder to identify you. If you set up your account with a VPN in Europe and pretend to be from EU, then these companies will be legally required to store your chats and identifiable information separately.

The second solution is a much safer option – and that is to run your AI chatbot locally. I am gonna do a full private AI tutorial in my next video, but to keep this as simple as possible this is what I recommend for now:

Head to jan.ai. This is an open source AI framework that runs locally on your device, thus keeping everything private. Install it. Once you run it, you’ll first need to download a model. If you head to the Hub, you’ll see a range of options. Choose whichever you like, you can start with Mistral or Llama. This will give you a bare and capable model. But if you want to give them a personality and you are new to this, I recommend you copy them from someone else. You can do this when you head over to huggingface.co/chat/assistants. Choose from an assistant that catches your interest, copy its system instructions and paste them into your assistant instructions. You can also download an uncensored model from Hugging Face, import it manually and give it your instructions. [21 – 23]

A full tutorial will be coming in the next video. There will be some cool things you probably didn’t see elsewhere. I talk a lot more about my research on Patreon, so subscribe and support my channel. Without your support, I will either have to stop making videos or start taking sponsors. So your support is truly needed. https://www.patreon.com/thehatedone