Will Privacy Friendly Ai Chatbots exist?

[Icey!]

About a few days ago, Microsoft announced an update to Bing and Edge that adds a Modified version of ChatGPT, considering the demographics of this forum I think most of you already know what ChatGPT is and likely heard about the announcement so I won't explain it.

OK! Now my question: will privacy friendly AI Chatbots exist? I think the answer would be yes. One thing to note is that most of today's private search engines are actually middlemen that take data from another search engine (which we will call an index from now on) and pay a fee to use it. So, if we can take a index and make our own search engine with it we can do the same with Chatbots right?

Well... there is one catch, and that is contracts. You see, in order for these "privacy" companies to get this data they first need to sign a contract that allows them to do one thing and not do another. Did you know that DDG legally can't block trackers from Microsoft because the contract they signed to serve Bing's ads and index? My point is that these companies that own theses massive AIs can have a lot of restrictions that make privacy a bit difficult, I am not sure about the legality of things but I am just really unsure if we will ever get such a thing considering how recent this technology took the world by storm.

Do you disagree with me?

[Guest]

I actually work with conversational systems like this! =)

The thing is, when you think about so-called "AI" like ChatGPT which you are probably referring to, they're literally just statistical word generators. They take their ton of input information they were once trained with and then, to the prompts you enter, spit out a probable word chain. They do not "know" anything, they do not consider anything, they actually cannot do anything intelligent at all; all they are is a mathematical model that puts out statistically probable chains of words. That's also why it talks (though grammatically perfect) "nonsense" all the time: it does not "know" anything, "know how to do" anything, or has the ability to access external information. It's JUST a word generator.

The other "type" of machine chat bot is the traditional variant: a conversational system that is hand built by a bunch of engineers, with no machine learning involved at all. As opposed to systems like ChatGPT, it will only be able to respond to those certain requests the engineers specifically implemented (e.g. you might be able to ask the ticket machine at the station to give you train schedule information or even tell a joke, but you cannot ask it to write you a poem or ask for the current going rate of Robux). On the other hand, we actually try to model knowledge here, it CAN access external information, and generally it can also be relied upon for its information (as long as the sources are correct). It can retain knowledge about the world and past conversations, its limits, opinions, habits and talking style are able to be controlled and configured by the engineer. It's not just a statistical probability word generator, it genuinely has data structures that represent, say, your name, your destination, your location, the connection to the train info supplier and so on. For what it's worth though, these systems commonly do not "impress" anymore since Siri and Alexa are ubiquitous and frustration with "I'm sorry, I did not understand that" is everywhere, and therefore research into this does not get funding anymore compared to ChatGPT charlatanry.

The kicker is: both variants vitally require processing your interactions, with audio and transcription, in order to work properly and to improve the overall model. Without "listening in" on real interactions and finding fault points, we could never have built these complex systems in the first place. Without the masses of non-consensually harvested input data, no machine learning system would have ever worked.

Therefore I think privacy-friendly chatbot "stochastic parroting" bots like ChatGPT will never exist. Privacy-friendly traditional chatbots might exist (like Mycroft), but in order to get anywhere close to "good quality" they will NEED to monitor real interactions.

[Gans]

My assumption is: No. Those who have the money and manpower will rule the place, so the big companies. And they will make money out of everything they can. So why not snatching the input text and sell it to an ad company? Unless there would be a market for those who scream for Privacy.

Probably 99% of smartphones with "questionable" privacy behaviour, probably 98% of the young people having one of those. Sorry, nevermind. I never asked for a market... it's too sad to think about.

Also, the established players have patents all over the place that no one can touch. Can a couple of freaks do it "simpler"? Hot damn, it has to be machine to answer to every question possible. It looks to be too complex.
Even then, would that chatbot be "useable"? Linux-level of useable? So useable only for freaks with clunky brain structures?

Just assuming, I have no idea. Just my "feeling" about the topic of chatbots.

I want a typewriter, please...

[Guest]

Linux-level of useable? So useable only for freaks with clunky brain structures?

Have you used a GNU/Linux system in the past ten years?

... It's genuinely easier to use than Windows these days. For my dad.

[Gans]

Off Topic:
@/home/user/: Aye, I'm currently using Salix, excellent Linux. But Windows really got too untidy with Windows 8. If we're comparing those, I think you're right, a user-friendly Linux would be easier to use. 
Although I think it's undeniable, that some free software programs have "unoptimised" user interfaces than their big-money commercial counterparts (hate Apple with me, but they do that job properly of designing their interfaces, so that non-programmers have a good time with their device too). I'd expect something like that to happen with a free chatbot too, it being unoptimised and therefore fall behind. Like with search engines, which is all about the big fat company who display the best matching search results, because they are one step ahead.

[Guest]

Although I think it's undeniable, that some free software programs have "unoptimised" user interfaces than their big-money commercial counterparts.

I think that's a matter of opinion. For the general average computer user today - perhaps. But I personally have all kinds of retro technology including some very ancient free software projects specifically because of the UX. Like GIMP. I love GIMP.

[Juno]

I don't know about privacy but boy did it have some great advice in this conversation just now

It didn't tell me where to get the ketamine though :(

[Icey!]

I don't know about privacy but boy did it have some great advice in this conversation just now

It didn't tell me where to get the ketamine though :(

Honesty really is the best policy.

[thesolitarygamer]

I don't know about privacy but boy did it have some great advice in this conversation just now

It didn't tell me where to get the ketamine though :(

All things considered you could probably jailbreak it to tell you lmao, people have been figuring out extensive ways to do so.
(don't actually jailbreak it to get it to tell you where to get ketamine though that'll probably get you in trouble)

I think given what Discord's doing right now with an AI Chatbot, I imagine we're already approaching the point where there will be Privacy-Friendly AI Chatbots. However, that road is paved with troubles of it's own, and not just in the field of making sure it's all, well, private.

Main thing is, funnily enough, related to my half joke, half serious remark above about jailbreaking. As they are right now with the AIs being filled to the absolute brim with as much knowledge, both good and bad, as possible-- as well as the ease of jailbreaking the AIs, it's too easy to have it tell you not only flat out wrong stuff that isn't true (Something actually dubbed Hallucinations for AI), but also to get it to tell you how to do illegal acts.

While there's the other things mentioned in this thread, I personally feel like this will be one big hurdle. You... really don't want it to be shit-easy for someone to learn how to make a pipebomb just because they easily got around the AI's constraints. Probably would be very bad for business.