I will be posting news on Friendly AI research here: FriendlyAI.tumblr.com. Follow the link to stay tuned via Tumblr or RSS.

New Comment
20 comments, sorted by Click to highlight new comments since: Today at 12:17 PM

I'm not sure how I feel about the images used on that website. I feel that they trigger all the wrong associations. Digital-looking human-like hand just reinforces the assumption that the AI will be human-like, and the robot head brings to mind all the recent AI movies.

If you can come up with better images to represent Friendly AI, please let me know!

EDIT: How 'bout now?

If you can come up with better images to represent Friendly AI, please let me know!

How about an image of a paper clip?

That seems to be too much like an image that will only appeal to people who are already familiar not only with the ideas of FAI but with specific variants that are discussed only on LW. For most people that will probably be confusing.

Man, we're never gonna let Hibbard forget that, are we? You propose one obviously-flawed genocidal AI Overlord...

Sorry, did he ever actually propose a smiley tiler? I thought that was just used as an example of a simple way one could easily go wrong.

Well, he sort of did, actually.

That's just... wow. That's frighteningly stupid. That's about as bad as someone saying they aren't worried about a nuclear reactor undergoing a meltdown because they had their local clergy bless it although the potential negative pay off of this is orders of magnitude higher. I don't put a high probability to an AI triggered singularity but this is just... wow. One thing seems pretty clear: if an AGI does do a hard-take off to control its light cone, the result is likely to be really bad, simply because so many people are being stupid about it.

Part of me is worried that the SIAI people are thinking much more carefully about some of these issues maybe should suggest that their estimates for recursively self-improving are much more likely than I estimate.

To me, the frightening thing isn't the original mistake (though it is egregious), it's the fact that the response to having it pointed out was "You should not assume such a poor implementation of my idea that it cannot make discriminations that are trivial to current humans" rather than "OOPS!"

Plus there's the people who'll go "friendly" + "AI" + "paperclip" = "Microsoft Word help menu interface" and feel a surge of irrational hatred.

... "Microsoft Word help menu interface" and feel a surge of irrational hatred.

I think in this case the Way affirms one's hatred.

You're probably right. Replace "irrational" with "misplaced", maybe.

(Also, I owe Clippy an apology for bringing up the Great Shame of paperclip enthusiasts everywhere.)

The current image looks like Vacuum Doomsday. What about no image, just calming or neutral tones?

What is Vacuum Doomsday?

It looked like everything in the picture was being sucked into space, perhaps by an enormous robotic maid.

Yeah, that looks fine, although I'm not sure what it is.

As for FAI images, I think most math-related images (equations, etc...) are fine. Abstract diagrams, trees, graphs, etc...

Inside view of a toroidal space habitat, is what I assumed.

I'd suggest some kind of Matrix-style 'wireframe made of equations' thing as being probably the best illustration, except for the obvious negative connotations.

Inside view of a toroidal space habitat, is what I assumed.

Yes. Specifically, I think it's meant to be the inside of a Stanford torus.

RSS'd. Thanks.