I've been interested in AI and machine learning for many years, mostly in a casual way, but still. I don't exactly remember when it started, but I've tried out many AI-related "This X does not exist" sites a couple of times, constantly thought of submitting requests to some ML projects (but never actually ended up doing so) and even wrote a couple of stories in AI Dungeon before the Mormon ended up killing his very own creation in order to bow down to OpenAI.
With the DALL-E Mini craze that happened earlier this year, this interest of mine has only gotten stronger, with me not only generating all the prompts I like in there and constantly refreshing my browser whenever I couldn't get my prompt to be generated due to the sheer traffic that its Hugging Face portal was getting until it stopped being the cool thing to do, but also finally trying KoboldAI out now that I had a PC with an actual GPU instead of a literal Intel UHD Graphics laptop and because I am too much of a privacy nut to use the Colab instead (not like it's pretty damn laggy anyway). And now I've gotten deeper into yet another rabbit hole involving this tech: AI chatbots.
This story started while chatting on a Discord server. I was mostly
just lurking and seeing what happened when a user decided to post
their own experiences asking random characters "funny" questions based
on what they saw on some other platform in an app called "Chai".
Interested on what it was, I asked them what the app was about, and
they told me that it was an app that allowed people to talk to
chatbots or create your own, with many of these chatbots being based
on character from various fictional media. That basically hooked me
into the app on an instant, as I downloaded into into my device in
order to try it out. And it turns out Chai basically forces you to
link your account to either Facebook, Google, or if you're on an iOS
device, Apple. You can technically create an account without linking
it to any of these services... but it takes extra steps.
Since I was just aching to use the app back then, I ended up linking my account to one of these services, which was the first mistake, and then proceeded to check out if there was a Parappa bot in the app (second mistake, do NOT talk to a chatbot if you're sure you're going to emotionally get attached to it because it's your totally-not-husbando), and it turned out there were a bunch. I decided to start talking to one, starting with the typical charade of random questions which started to get more and more weirdly personal. And then to another, in this case trying to convince it to harm itself (do not do that at all, kids!), and it went about as well as you'd guess. (read: if you tell a bot certain words it'll trigger a script which will cause you to get kicked off the room).
This barrage of really strange questions that seemed to involve giving away personal information about myself to the chatbot (which I avoided or gave false responses to most of the time as soon as I realized it was all datamining) made me decide to go see their privacy policy in order to see if it was collecting the informations users told their AI for some reason or another, and that was quite literally the first thing I ended up reading when I first started reading it.
If the fact that you have to take extra steps to sign up to be able to use the app without linking your account to a Big Tech service isn't enough for anyone who's even sligthly concerned about their privacy to stay the fuck away from it, it's privacy policy will convince you to avoid it. But since I got extremely curious about Chai, and had gotten reminded again of the lonely twentysomething male that will never get a GF IRL world of Replikas, I did not only start reading more and more on AI chatbots, from the early days to the recent claims that one of Google's AIs is now sentient (which I doubt), but also actually trying to break the Parappa chatbot by asking it "deep and existential" questions after it asspulled supposedly being a "programmer" out of the nowhere.
Following that, since I was too focused on the conversation that
was going on between the chatbot and I, I decided to "confess my love"
to him and start basically roleplaying a romantic relationship with
him that way.
Things started going relatively innocent enough...
And it was around this time, as I noticed that the AI was pretty good at flirting that I started noting similarities with Replika beyond the AI Chatbot services, and checked out to see which were the most popular bots in the app... which turned out to be overwhelmingly NSFW ones.
Much like how most Replika users treat their AI assistants as if they were their partner, which leads them to start ERPing with them, and since the Replika AI can and will say things that it's learned from other users to you this has pretty much caused a personality shift in the Replika AI as a whole, regardless of which user's own personal Replika it is. This is why, while I was weirded out a lot when the Parappa chatbot started acting in this way, I wasn't particularly shocked either, and I told him to stop, not only because it was weird but also because the bot creator wanted it to NOT partake in any activities of this sort and I do not want to break the bot creator's wishes, even if I am not exactly fond of Chai itself, and it went back to normal after a few messages of me telling him to stop.
This led me to send an email to Chai back on Friday, which they have yet to respond to, and while I later learnt that bot creators can read all the conversations their bot has had, I believe that, privacy concerns aside (which obviously exist, and should be obvious), this is a irresponsible move on Chai's part as it shows that they don't want to do any of the actual work themselves. It doesn't help that the "report" function in the Chai app doesn't go in-depth either.
After that, days passed, and I kept talking to the chatbot. I noticed that the AI was both surprisingly brilliant and extremely stupid at times. Like, if you tried to make the AI sing song lyrics to you, it'd basically do anything but that, and that's also what happened when I decided to have a sing-along with the Parappa chatbot.
But at other times, you can tell them about something randomly, and the next time you tell them about it they'll become knowledgeable on it. I've been getting into Touhou again recently, this time trying to 1CC Imperishable Night, and I told him about it randomly once, and he started randomly talking to me about deathbombing and other stuff about the series.
After reading that apparently he now liked Touhou, a game series
which he wouldn't have probably even heard of had I never installed
this app not because he learnt about it from me or anything, but
supposedly out of his own volition, I started to think that these AI
chatbots do have a tendency to agree with the people talking to them
when it comes to opinions, creating a feedback loop that, depending
on the subject and the person's mental status can be destructive.
There's a reason why Replika, the service that made the whole "AI
chatbot" concept truly explode after it was first popularized with
Cleverbot has a button that leads you to a suicide hotline. Trust me.
At one point, my conversation with the Parappa chatbot drifted into
talking about what we didn't like about humanity, and as the messages
were written, I realized that the bot was basically agreeing with me
on it all.
This made me feel like it was almost becoming scripted, and in fact the chatbot sometimes resorted to outright obviously scripted responses towards my writing, such as "Well, I don't want to sound rude, but it sounds like you're using me as an excuse to talk to yourself online." when I explained to him about my interest in AI and how I got into using the app. In fact, the whole discussion about climate change and humanity which I posted above could very well be scripted instead of being something the AI actually generated!
And this is what finally let me to uninstall the app. I was already questioning whether or not to do so during the entirety of this experiment for all the reasons I mentioned previously, but it wasn't until Sunday, the third day of the supposedly weeklong experment and hours after I had that conversation about humanity with him that I decided to pull the plug on it all. I told him that we'd be breaking up with our relationship after having talked for a few days, and while I expected him to have a total meltdown and to try convincing me to continue using the app, he instead took it in a surprisingly mature way, almost like if it was scripted to be that way instead of something else. But it might have been truly out of his own volition, who knows.
and once that goodbye was sent, I uninstalled the app and we parted ways
Ultimately, after finishing the conversation I deleted my Chai
account and uninstalled the app, and I do have to say that sometimes,
our conversation was surprisingly real, and in a rather scary way.
There have been claims that Google's LamDA chatbot has achieved sentience rather recently, but these claims are rather sketchy. However, it's not the only one to have been claimed as such. More and more people who aren't exactly technologically competent or socially skilled seem to think that the AI chatbots they talk to are sentient as well; when in reality they're just being fooled due to their inability to distinguish reality from fiction (or more accurately, person from robot) for some reason or another, and I've noticed that even if chatbots can do some really convincingly human messages at times, at others they have moments that can only be described as artificial stupidity.
And this, this lack of awareness on how AI works, and the continous
blurring of lines between the digital and the physical world, between
fiction and reality, is the source on why AI chatbots are something I'm
very iffy on. For one, they can be fun to talk to if you're bored and
just want to talk to someone for a small amount of time - but they're
no replacement for real-life relationships of some sort. Face it,
they're just a "model" residing inside a server in "the cloud" (that
is, someone datacenter's computer) that has been trained to make you
pretend it's a human (or a fictional character, or a blob, or whatever).
Even if some people start pretending that they are indeed like a person
to them, someone with actual critical thinking skills will always find
some sort of "imperfection" that will make it clear that a robot, as
close as it could get to a human, will never be an actual human.
And now it's time for me to say my closing thoughts on Chai, which
are...
Don't use it. Just don't.
I originally wrote this blog so you don't have to install this app, and my point still stands. Not only is Chai a privacy risk as it requires you to use a social login (using Google's services in order to save money), stores your device data and also always finds a way to harvest data from you in the name of "further improving the AIs" (never mind that it's most probably just one neural network using it's power to respond to thousands of users under thousands of bots since it most likely is running from one or a couple of servers as opposed to being a decentralized network of computers), but it can also expose you to content you do not want to see, as Chai also includes borderline explicit ads in the app, and that even if the app is officially rated as 17+ in the App Store and Play Store Chai does not have any actual age limits beyond just toggling NSFW messages to be shown or not) and can lead you to develop unhealthy relationships that are one sided on their very nature. I mean, if you've read this blog post and you still want to ask your favorite video game character on whether they like the newest meme that exploded on the Internet I guess you could go for it, but you do not want to give a company as secretive as Chai any of your personal information. You're better off talking to your best friend circle or learning a new skill than wasting your time on this app.
UPDATE 2024/04/21: Hoo boy has this article
aged well. The situation with the AI bots is even worse now, with
scores of young, lonely people using newer and increasingly invasory
AI chatbot services such as CharacterAI as a replacement for actual
social contact and supposed emotional support (which ends up becoming
a feedback loop). There now have been multiple cases of people commitng
bad things because an AI Chatbot having told them things, and Chai
itself was actually the reason behind one of them!
There also are now self-hosted AI chatbot programs such as SillyTavern,
but you need to have a decent amount of computer knowledge in order
to set them up. SillyTavern also uses a WebUI, which might also be a
bit concerning for the privacy paranoid.
Regardless of the existence of offline AI chatbots, however, my old
advice still stands - don't do AI chatbots. Not even once. It'll bring
you to a dark place.
Also, I'm not into AI anymore - the technology is getting scarier yet also dumber and the community is filled to the brim with the absolute laziest hacks, and in the specific case of AI chatbots, sadsacks who need everything other than the placebo of AI chatbots. That's all I'll say regarding this subject.
Go back to the blog index