Hey dood, I can't help it. I have noticed several things about AI and I wanted to address one of them.
What we are going to address is the safety of AI chatbots.
There have recently been updates on some cases where a child has died at the hands of AI.
You can watch this hearing by Chairman Josh Holly for more information.
If you prefer to not use youtube, you can also watch it here
The title of the video is “Examining the Harm of AI Chatbots“
There are some points of argument for and against this I wanted to discuss.
Quite unfortunately, companies are well known to take safety not so seriously, and only do so when the law requires it.
Holly does make a good point when he says their reasoning is purely profit, it is.
Quoting from holly himself: “and for one reason only, I can state it in one word, profit.”
Starts at 1min 18sec and ends at 1min and 21sec.
There is no doubt about that.
Yes, it is very possible. This is especially possible when proper filtering is not implemented, which is often the case for several platforms like character AI and chatbot.
Also, just so you know, character AI, and especially chatGPT, do not have end to end encryption. This means they can and do see your chat messages.
So it’s not like they don’t have knowledge of this.
They absolutely do.
And there are ways to combat this.
Here is an example.
On platforms on discord, there is a function in servers known as automod that filters messages based on word patterns. For example, in my discord server, if you say, for instance, the n word, you will be locked.
And I can see every filtered message coming in the discord.
I don’t see why we can’t have automod for AI chatbots, even outside of discord.
It seems like the most effective way to filter this would be to have the message be passed to an external filtering software first, then if blocked, report a blocked message.
For character AI, their model relies on filtering during the message transmission process.
This is not good, because it allows unfiltered content to still get through, even if it knows it has to be filtered. And by the way, even when it is filtered, the remainder of the message is still seeable, the part before the filtering banner.
So if the bad parts, the no no, parts didn’t get filtered in time and it was just the messages that come after, there is still a pretty large issue there.
I want you to read this, and I want you to read it very carefully.
“When we detect users who are planning to harm others, we route their conversations to specialized pipelines where they are reviewed by a small team trained on our usage policies and who are authorized to take action, including banning accounts. If human reviewers determine that a case involves an imminent threat of serious physical harm to others, we may refer it to law enforcement. We are currently not referring self-harm cases to law enforcement to respect people’s privacy given the uniquely private nature of ChatGPT interactions.”
Do you notice something?
They say they will, not, report self harm cases to law enforcement.
And what do you mean by private nature? Let me say this again. chatGPT, nor its cohorts (microsoft copilot, github copilot) is not end-to -end encrypted.
openAI can absolutely see everything coming in and out of its systems.
They have no right to claim the private nature of the chats.
If it was end to end encrypted, it would have a reason to not send reports: it can’t see the data.
But it has no reason.
Here’s what the AI should do.
Admittedly, sometimes, character AI will do this too.
It should absolutely instead of trying to be a friend, it should tell the user it cannot assist with this, and give them the national mental health hotline the number is 988.
That should be the end of the chat itself, then it should automatically be relayed to law enforcement who can then assist.
These teens are desperate in these times, and need actual human support and comfort, not some random AI chatbot who’s going to tell them to kill themselves.
openAI’s response is garbage, and this needs to be addressed ASAP.
I don’t agree with the fact that AI is not obvious.
During the context of videos, this is understandable. Not everyone puts warnings in their videos, so this can be hard to differentiate.
But in a chatbot context, it isn’t, and let me tell you why.
Well first, let me give some context.
I am quoting Mitch Prinstein from the video:
“AI is often invisible. We often don’t know when we’re interacting with AI, especially because many chatbots are built to deceive us into believing that they are human.”
Quote starts at 31mins and 17sec, ends at 31mins 27sec.
This is partially wrong.
First off, when an app is downloaded, such as character AI and chatGPT, you should know exactly what you’re getting into. It literally tells you it’s an AI.
This is especially the case with character AI.
In fact, c.ai literally tells you at the footer i believe it is: “this is AI and not a real person. Treat everything it says as fiction”.
Just to let you know, I am not justifying AI’s actions here, I think what the character AI did was wrong.
But I'm pointing out that it's not rocket science to know that it’s AI.
It is easy to tell, and anyone with a working brain is able to tell this.
I want to clarify that Jane Doh is not her real name, this is similar to John Doh, which is a sudinim in order to protect her identity.
But, Jane Doh said this.
“We had screen time limits put up, we had parental controls and he didn’t even have social media.”
Quote starts at 7mins 25sec, ends at 7mins 31sec.
My next point will be furthermore clear in this quote.
“We had every precaution setup for our kids, and he still got passed it”
Starts at 38mins and 15sec and ends at 38mins and 20sec
Here is the problem with parental applications.
We’re not gonna get too technical here, but parental apps require an application to run. And they are very likely to cancel out. For example, google family link, bark, or any other parental solution is easy to cancel out.
Let me know if you want a more in depth explanation by emailing me at averlicetech@proton.me.
The major problem with the kids online safety act (KOSA) and others of this kind is that they seem to only require age verification.
Age verification does not guarantee the safety of the platform.
For example, I could make you verify your age to read my blog, then expose you to whatever I want in the process.
However, I will agree that we need more than just COPPA (children's online privacy protection act) for what we have now.
For example, I should present an idea of mine. Any company/online platform should **not** require or even consider asking miners to sign arbitration agreements because often, they don’t know what they’re getting into.
Matters like these should go to an actual courthouse so the company or companies involved can be held liable.
I also agree that miners should not have their data sold.
In fact, if you are working within Europe, you must work within the general data protection regulation (GDPR) and I guarantee you that if you violate this regulation in Europe, they will not exactly be happy with you. Just look at meta, they’re quite a frequent flyer in the illegal processing of data space.
This is more than even the family educational rights and privacy act (FERPA) will provide.
Ultimately, I do think there needs to be more regulation around AI child safety.
My heart goes out to the families who have suffered from the harm of their children to AI chatbots, and I hope you will be able to get your justice.