Armitage_Shankburn wrote:

Unlikely wrote:

Yeah, like I say, it's a decent chatbot.

We'll see about that . Summon Escape and Dyno!

:'P

Like any AI project, it can get pretty (very) good at tightly defined problems.  The hype/horror train for AI is to be approached with a healthy regard for that.

Never say never but for the time being that's where it is.

The fact that neural nets work without us being able to analyse diagnostics in a meaningful way (they are kind of like a black box and we can only really evaluate them based on input/output) means they should not be used for anything of real importance.

It's also worth noting that any AI solution is a product of the environment it was "trained in". It will be significantly affected by the scope of the data used to train it. AI needs to be shown what it's looking for by example - or by telling it that its decision is good or bad (usually by some sort of score). Humans need to set that up.

There are clever ways of making one neural net compete with another. For example say a bot is identifying pictures of a boat. You can build one net that tries to identify a boat and one that tries to fool it by making fake boat images that look like boats but aren't. This can very quickly create a real expert at identifying boats, and expert at creating fake boat images.  But - both those need to be set up by us. And we still need to give initial training to both nets to get them going.

And. Any new data that doesn't fit the pattern of the AI bot's/net's  "experience" may not produce a desired outcome.

We need a net to be able to know it doesn't fit and use some other method of evaluation - or own up to know it doesn't know what to do (so a human can take over) - instead of taking a possibly catastrophic decision.


I'm sure there are people trying to create nets that train other nets and set up environments to create other nets (meta nets). Head spinning stuff.

Also it's the reason why I'm always polite to Alexa

Look how well that worked out for John Connor and his mates

Why? Is it the threat of it doing secret stuff to ensure it can’t ever be deleted and shut down? If so, I’m not really sure how they get over that problem?

Fuck me. Evil corps eventually rule the world then eh? Sheesh.

Imagine AI in the hands of the Tories

Wookienopants wrote:

If you haven't already, have a read of Scary/smart by Mo Gawdat. Very very interesting read on AI by one of the engineers working on Google X

Will give that a look. Cheers.

yourfavouriteuncle wrote:

Why? Is it the threat of it doing secret stuff to ensure it can’t ever be deleted and shut down? If so, I’m not really sure how they get over that problem?

hunk wrote:

Sex and war robots

Seems like an odd combination of functions to me.

I’m not fucking a nuke.

Yossarian wrote:

hunk wrote:

Sex and war robots

Seems like an odd combination of functions to me.

Now now, let's not kink shame. It's ok if Hunk likes to fuck and then invade a foreign nation.

Have you people never watched blade runner (sigh).
My post was tongue in cheek but sadly I am totally convinced people are capable of utilising (abusing) AI in such a depraved manner.

hunk wrote:

Have you people never watched blade runner (sigh).
My post was tongue in cheek but sadly I am totally convinced people are capable of utilising (abusing) AI in such a depraved manner.

Of course they are.
Human beings can abuse and weaponise everything.
It's why the issue of AI is such a frightening prospect