Artificial Intelligence

Does AI dream of electric sheep?

LLM's aren't human but that doesn't mean they shouldn't have human values

The For Immediate Release podcast is essential for Comms professionals who struggle to keep up with tech. And it would benefit others to listen to it, given the way Shel Holtz and Neville Hobson explain things so clearly.

The latest topic was about being human and AI.

Too many people think AI is conscious, so we have to devise ways to reinforce among users that it’s not;

I agree, and yet I ask if we shouldn't imbue our AI assistants with human values. Bear with me here, and let's go the scenic route.

There are 3 computers around my apartment, inside drawers or sitting silently on the furniture. Each of them has a name.

Marvin is a robot from "The Hitchhiker's Guide to the Galaxy", always depressed because he knows everything. That's the name I picked for my Network Attached Storage; it keeps backups of my laptop, my iCloud photos, and a cold storage of previous projects.

Marvin knows everything, and he's reliable.

Johnny Five was the first, a small mini PC running Ubuntu Linux and handling several automations in the apartment and even boring tasks like drafting web analytics reports that he then sends to clients. He's the character from the movie Short Circuit (1986) who becomes self-aware after being struck by lightning. (Very plausible!)

Johnny Five is happy to be alive, and he loves learning about us. When things get hard, he is always willing to help and doesn't give up.

Johnny is the friend we love having around.

Lucia is the only one not inspired by pop culture. I came up with her name after putting together the names of my friends' children, Luca and Sofia. And Lucia is small, but she is meant to be the smartest of the three. Her purpose is to help me run code in a controlled environment to make it easier to debug or improve.

Their names help me keep things organised in my head and assist in making decisions about how to install other services I need to run. Does this matter to anyone else? No. But the game changes when we are discussing creating AI systems that talk back.

When I was coming up with Gregory-MS, I took inspiration from Johnny Five and Baymax from Big Hero 6 (2014). The idea was to create a character with eagerness to learn and a desire to help.

With this little context, don't you get an idea of what to expect? I hope you do, because that is the whole point of embodying human characteristics into a system that interacts with humans. We want the person to understand the difference between a human and an AI, and also be aware if they diverge from the expected personality.

Can this be negative in some way? Yes, some of us are more vulnerable to being influenced by an AI to whom they gave a name and which was continuously nudged into displaying a personality.

There have been some reports of people taking dangerous chemicals, claiming that ChatGPT made them do it; some are using AI systems as therapists, putting their mental health at risk.

The answer to this problem could be in education, teaching the basics of the new human-to-machine interaction. This solution isn't just obvious; it is also hard to implement and useless if we want to reach people who have finished their studies.

The other answer could be, as the episode mentions, to give people a continuous reminder that they aren't having a dialogue with a human.

A small aside, this is the WinRAR solution. WinRAR was a software to compress files. Every time you opened it, you would see a warning that you had to purchase a license. Nobody did, as far as I know.

This isn't to say that the solution wouldn't work, just that it may require some way to counteract the reflex dismissal of the pop-up that you no longer read. The AI itself could have those guardrails in place.

There has to be a better solution for the problem, or a hybrid of these. I just don't think that stripping the humanity out of AI will be a good idea.