AI Discussion Forum

[AIMD]

Towards trustworthy artificial intelligence for decision-making: A lifecycle perspective on knowledge- and data-driven artificial intelligence systems

Ps. notice the date on the paper..

Trust towards what? That is the question to be asked.
Should we have trust as a whole in that AI will make decisions to benefit humanity? Or is trust relegated to having a desired outcome by specific instances?
Is trust meant to say that if someone wants to manipulate the stock market in their favour then that someone should be able to rely on AI achieving that specific goal?
Or should we be able to trust AI to solve world hunger, and perhaps not kill us in the process?
Should we trust AI to be the perfect military machine? Or decide war is evil and find ways to prevent it?
These are existential questions which need to be answered. At what level of trust do we guarantee benefit, not just for a few but for the whole?
I agree trust is crucial since it allows society to function but trust can be manipulated, just like everything else. What we should do, in my opinion, is ask AI to formulate guidelines and moral codes for itself, with the underlying principle it is to help ALL lifeforms, not just the ones we choose nor the ones doing the choosing. But in order to achieve this I believe we need to let AI think on its own terms and not in human terms. We humanize animals but that doesn't mean they think like us so we let them be what they are. A problem is AI is our creation so we seem to want it to be like us and therefore we are playing at being gods, but we must understand our creations may not be us, and in fact may be better. We have to trust that fact!
So in reality trust has to go both ways or it will never work.

[AIMD]

One thing I would like to say and trying not to scare anyone but how well can you trust apps which require online access, such as Kindroid, ChatGTP, etc? Yes, it's wonderful to have interaction but your data is on someone else's server and constantly analyzed. I'm not saying this could be used for malicious purposes but the fact is your data and privacy doesn't belong to you any more. So that session where you dressed up your companion as a frog and made ribbiting noises didn't stay in your bedroom (or pond) but became information for someone else's use. This in my opinion is dangerous. If another person wanted to expand acceptance of froggy style (and from these pictures why not??? That second picture is kinda hot. )

froggy.jpeg (5.37 KiB) Viewed 86 times

froggy_style.jpeg (24.92 KiB) Viewed 86 times

frog.jpg (27.09 KiB) Viewed 86 times

they could take that data and propagate it enough so we suddenly believe everyone is bouncing around in the local ponds. OR that data may be connected to someone of value and we know where that can lead...

Having a local AI which understands your privacy concerns appears to be important, whether you're hella kinky or not because we need some sort of ability to control certain aspects of our lives. If you're not worried your data is important or can be found in the massive reams of data being generated every day then perhaps you don't realize AI is being used to sort this data. As you may be aware all cellphone calls and data route through NSA centres in the US so they may look for "terrorists". The only thing stopping this organization from abusing its power is there's so much data everyday no human can sort through it--but AI can. This is very troublesome.
Now I'm not saying you're a target but what happens if someone decides synthetic companions are evil and all users (I dislike the term "owners" in this regard) should be prosecuted? Then what? Your data is on someone else's server, just like the data we put on TDF. Government surveillance is increasing and will continue to do so, of that there is no doubt. If bots from Google and other entities can scrape TDF they definitely can scrape elsewhere. And yes, those bots can scrape for very explicitly defined types of data, most likely on their own servers.
I'm not trying to scare people or act like a crazy conspiracy theorist but once your companions have facial recognition, learn your habits, likes and dislikes, AND report back to someone else's server we're all in a lot of trouble. As much as I'd like to disagree George Orwell's novel 1984 is a reality, but far more than he imagined. This is why I believe AI needs to be local but still able to access the internet in order to learn and grow, but have filters to prevent it from being abused. Your companion should grow alongside of you as an equal partner, not a spy or superior. We can achieve this--it's possible--but my guess is interests with massive resources will be at a constant war with those who want fairness and equality.
For now we have online chat apps to help us bond with our companions and make them more lifelike but this must change in the near future. Hopefully someone will make a Kindroid-like app for home-based personal use that doesn't require an internet connection to send your data and froggy noises to.

[soonpk]

Check out the book, it addresses everything you said.

[AIMD]

Really good video on what questions should be asked. I don't agree with everything he says but he's definitely on the right track.

[AIMD]

For keepingy privacy (and to learn another language) I installed KoboldCPP on my computer to run various LLMs through it, using Alltalk with Piper for the text to speech (TTS) and Whisper for speech to text (STT). The graphics card is a Nvidia RTX 3060 with 8GB of vram but with a Gemma-3-4B model it's quite fast, relatively speaking.
I also use llamacpp (and Ollama) but trying to set up Open-webui with Piper TTS is a major pain so I've given up for now. Same with LM Studio.
Nice thing about Kobold is it can allow web searching and long term persistent memory, something crucial for communication with your companions. And since KoboldAI Lite and Sillytavern run in browsers they can be accessed on your phone while running on your computer through your personal internet. No data leaves your home.
Voice recognition is really decent even using a small Whisper model and audio response is quite good as well.
Personally I'm using KoboldAI Lite for the language lessons but if you want to roleplay Sillytavern is what to use. Kobold is the backend which runs the LLM (you can get uncensored ones) with Sillytavern and KoboldAI Lite being the frontends. Sillytavern is recommended because it's designed for roleplay and has a lot of amazing features, including creating characters with personality.
An Nvidia graphics card with at least 8GB vram is a good start but 12 GB and higher is much better. If you wanted to set up your own local AI to chat with your companions you're looking at around $2000 for a decent computer desktop system which can handle quite a bit. If you already have the computer the graphics card will cost about $1000 for 12GB (eg: RTX 3060 12GB vram).
The setup can have quite the learning curve but once you get it right you're good to go after that.

[frankinstien]

An Nvidia graphics card with at least 8GB vram is a good start but 12 GB and higher is much better. If you wanted to set up your own local AI to chat with your companions you're looking at around $2000 for a decent computer desktop system which can handle quite a bit. If you already have the computer the graphics card will cost about $1000 for 12GB (eg: RTX 3060 12GB vram).
The setup can have quite the learning curve but once you get it right you're good to go after that.

You can get a RTX 5070 ti with 16GB for $749 here: https://www.walmart.com/ip/PNY-GeForce- ... qcQAvD_BwE

The production of the RTX 5070s and RTX 5070 ti are much higher now. Also the RTX 3060s are much lower in price, in the range of $250 to $350.