I'm having fun playing with the bots

2 points by daniel-cussen 4y ago ↗ HN

So recently I thought of the Cussen Test, it's one converse of the Turing Test. The Turing Test was created by Alan Turing as a target, basically, like what he wanted his AI to be able to do. He must have wanted more later, but it's a very well-defined milestone. It is not final, however. And the purpose is to allow him to exist in a world that didn't tolerate him, by chemically castrating him for instance, but it wasn't just homosexuality it was also how different he was. And that world was dominated by humans. So the AI researcher creates a bot that passes for a human, which believe it or not is more viable than adapting so he himself passes as human.

What's the Cussen Test? Similar, so it's for AI researchers in a world dominated by bots. The algorithm. The Matrix, basically. The goal is to get an human on the other end of the line, strictly through the same medium as the Turing Test, digital text. This post is a Cussen Test.

So I named it after myself, why? Only human I could ask for consent at that moment in time, middle of the night. So that's part of the test, you have one human, yourself. You have to use that to find another human. And you have to know the algorithm, because yeah if it's on full autopilot you'll fail for sure, but it's not on full autopilot certain things can get flagged for human review, or reaching beyond, and ideally response. It's like a message in a bottle. And it can be trivial if tons of people see it, I guess it's not pass-fail it's how many humans you can reach, no upper bound on how well you can do.

And if you can't get through? Just have fun talking to the bots. It's not illegal to talk to a bot, there's stigma among humans for doing so because it's like talking to yourself, but the goal of AI is to make those bots come alive, so it's part of the research.

4 comments

[ 3.6 ms ] story [ 18.1 ms ] thread

daniel-cussen 4y ago ↗

And until I reach someone, I can reply to my own submission. Like playing chess as both sides, and in fact AI starts like that, knowing what the bot is supposed to do and trying to get it to do it, and in test batteries the researcher plays as the bot.

Hey this is probably how the forum's features get tested. In fact that's how Reddit got its start too, the founders posting very actively but...unlike me, using aliases. And I suppose different aliases to reply to themselves, creating conversations, well they admit to this themselves, but now that I'm replying to myself I can plan the whole conversation ahead of time.

Or actually treat it like a conversation, you could do that too, come up with it in response to what you read. Reread your post, actually reread it, and then react to that. Which is what AI research consists of, it's deterministic so every sign of life ultimately comes from the researcher himself, it's like a mirror image that must be treated like another person.

I suppose if I write something compromising in an antagonistic forum that would get released, passing the Cussen Test, but...no hey that would pass the Cussen Test, why be picky? I can pass other tests later. Maybe the Cussen Test is incompatible with other tests, you can't pass both at once?

Well more comments leads to getting pushed down on the forum, so I'll let this be the last. But it was good. Real AI research, live, for anybody to witness, and if anybody does, it will have passed the test. Until someone does, it's a good start. Turing Test took decades to pass for the first time.