Forum adverts like this one are shown to any user who is not logged in. Join us by filling out a tiny 3 field form and you will get your own, free, dakka user account which gives a good range of benefits to you: No adverts like this in the forums anymore. Times and dates in your local timezone. Full tracking of what you have read so you can skip to your first unread post, easily see what has changed since you last logged in, and easily see what is new at a glance. Email notifications for threads you want to watch closely. Being a part of the oldest wargaming community on the net. If you are already a member then feel free to login now.

 Kovnik Obama wrote:
The first half-second a true AI gains access to the IoT, it'll become as much unkillable as the IoT itself. You'll have to hunt it through everything ever networked, down to every single wifi-enabled corporate thermostat system.

This is pretty overstated. General-purpose AI is likely going to require vast amounts of processing power, something that your wifi thermostat doesn't have. And that's assuming the AI could even meaningfully interact with a single-purpose device like that in the first place.

An electronic, digital AI could distribute its processes across many separate networked devices. Its problem would be latency. The more widespread the network of processors, the slower the AI would be able to operate.

Somewhat related poker AI has now evolved into stage where it can beat human pro's in heads up NO LIMIT hold'em and beat them bad(winrate is pretty ridiculously high...). This has been true for limit hold'em for quite while but no-limit is much more complex so that was going to be bit more work. But less than I thought...

Games at least are being solved at frightening rate. Few years ago I was beating crap out of couple bots online but now I would get murdered.

Chess, Go, poker. Can't soon play anything for anything but pure fun over internet.

It probably will be possible to create an AI or bot that will play any one game perfectly within the constraints of luck.

Chess, Go and Poker are amenable to the technique because they have very large, complex probability spaces leading to solutions, which computers can calculate much more easily than humans.

A conventional wargame will be harder to solve, due to random effects.

However, if playing a game against a bot becomes like running a race against a motorcycle, why would people bother?

 Kilkrazy wrote:
It probably will be possible to create an AI or bot that will play any one game perfectly within the constraints of luck.

Chess, Go and Poker are amenable to the technique because they have very large, complex probability spaces leading to solutions, which computers can calculate much more easily than humans.

A conventional wargame will be harder to solve, due to random effects.

However, if playing a game against a bot becomes like running a race against a motorcycle, why would people bother?

Don't see luck being really issue. Poker has element of random but that's still game of skill. If one can create bot for poker one can write bot for say warhammer/warmachine. Sure it doesn't quarantee win on single game just like best poker bot won't quarantee win on single hand(and by definition most of right solutions are "give up") but in the long run you would beat most games.

Poker still doesn't require the abstractive and on-the-spot thinking that a 'true' AI would need to master to be fully aware by our standards. Being awesome at poker doesn't necessarily mean a computer can suddenly think about solutions to say, a car suddenly barrelling at the poker table and what to do.

Or a person asking it, "do you like butterflies, or ladybugs better?"

I would classify most AI computers in scifi (and what most people really think of as "AI") as extremely interactive computers, with responses designed to anthropomorphize them to their users. They are still only awesome at what they are built to do.

 sebster wrote:


Sure, and all I'm saying is that it is testing for a thing that isn't useful. It's like testing for a new metal alloy that's indistinguishable from concrete - we don't need to replicate what we already have in vast quantities

How would you even being to theorize non-human intelligence when we can't even conceptualize definitively human intelligence? Even worse, how would you ever test for it?

Again. I think your subscribing to the Turning Test something it doesn't posit or set as a goal. The actual goal is useful because turning was right. Who the feth cares if the machine can think? If it can carry a 6 hour conversation without acting wonky we can't really tell the difference anyway. The Turning Test persists because its goals are conceptually simple and pragmatic (though the mechanics are very complex). No machine has ever managed to convince more than half the judges it was human. Turning himself posited his test as being passed when the machine managed to fool 70% of the judges.

I think we can attack the test for positing a machine that simply imitates behavior, but that in itself is not a meaningless goal. At best its a low ball goal, because if we can't even make a a machine that can act like its thinking, I can't fathom how we'd ever make a machine that can actually think. The former is a infinitely simpler goal than the later, and both are a lot simpler than positing some kind of intelligence we ourselves are incapable of comprehending (how the would we ever code for that? It seems more like something that would happen as a result of glitches or calculating errors than any deliberate human action).

Hell I'll be happy when those stupid machines that answer all the customer support calls can actually register what I'm saying. I don't really care if they can think. I just want them to listen XD

 LordofHats wrote:
How would you even being to theorize non-human intelligence when we can't even conceptualize definitively human intelligence? Even worse, how would you ever test for it?

You're assuming we need to test. The actual real test of any new bit of software is whether it is useful. It does something a human cannot do, or it does it quicker, cheaper, better or more conveniently. As long as computers increasingly perform those functions, then they will march towards greater and greater intelligence as a matter of inevitability. Exactly what point these computers achieve some test we've set out for them doesn't seem to matter all that much.

It's like a person in the middle of a vast journey, wandering exactly when he will have travelled a really long way. Is it the 1 millionth step, the 10 millionth step, the billionth step? Dunno, just keep taking each step.

I think we actually agree on a lot of this. Your point about it not mattering whether the computer is actually 'thinking' provided it appears as if it were is a good one. I just take that a step further, and say it doesn't even matter if it appears to be thinking.

I don't know, I guess I've always assumed stuff like real machine thinking will be emergent, the end result of ever more complex designs.

Hell I'll be happy when those stupid machines that answer all the customer support calls can actually register what I'm saying. I don't really care if they can think. I just want them to listen XD

You should see how they manage an Australian accent. I don't know a single person here in Australia that can get siri to work.

 sebster wrote:
You're assuming we need to test. The actual real test of any new bit of software is whether it is useful.

Isn't a better interactive interface useful? If nothing else maybe they'll finally figure out how to work accents XD

I just take that a step further, and say it doesn't even matter if it appears to be thinking.

They seem like the same thing to me. Functionally a machine that is programmed to act like a thinking being is a thinking being. Is that useful? IDK. It's not like we've only ever pursued stuff because they seem immediately useful. Humans do lots of stuff just because it's cool or something to do. Like building pyramids! If you had nothing but time, money, and a city of people who worshiped you as a god you'd tell them to start stacking rocks too

What's the difference between a brain and a computer anyway? We're programmed too. We just weren't programmed digitally... or were we?

 LordofHats wrote:
Isn't a better interactive interface useful?

Definitely. But that's where something like this test can be a bit of a trap. Afterall, if a machine listened to every word I said, and responded directly without trying to drag the conversation over towards something it was more interested in, well then I'd probably figure out it was a machine. But it would also be a much more effective source of information than asking a person

If nothing else maybe they'll finally figure out how to work accents XD

Truly the realm of science fiction.

They seem like the same thing to me. Functionally a machine that is programmed to act like a thinking being is a thinking being. Is that useful? IDK.

Yeah, don't get me wrong, I'm not saying we shouldn't create these kinds of interfaces. I'm hanging out for the day I can say 'tea, earl grey, hot', and I fething hate earl grey.

I'm just saying we should assess that technology by how useful it is, not whether it passes an arbitrary test.

It's not like we've only ever pursued stuff because they seem immediately useful. Humans do lots of stuff just because it's cool or something to do. Like building pyramids! If you had nothing but time, money, and a city of people who worshiped you as a god you'd tell them to start stacking rocks too

I'm definitely a big fan of letting researchers do all kinds of awesome stuff just because it's awesome. Sometimes awesome things end up being practical as well as awesome. And even when they don't we often make all kinds of discoveries along the way. which lead to amazing stuff. And even if that doesn't happen, we still get that awesome thing.

But my idea of awesome is more open ended. We should be encouraging and supporting development in to applications that we didn't even realise were possible a decade ago. We certainly shouldn't be focusing too much on a test developed in the very earliest days of electronic computing. Would you rather develop a computer that can trick you in to thinking it is a human, or a smart phone app that can diagnose a melanoma so you don't even have to leave the house?

 Pouncey wrote:
Video showing the robot passing it:

https://www.youtube.com/watch?v=fsF7enQY8uI&feature=youtu.be

A Turing test is when a robot must convince a human in another room that cannot see it, that it is a human and not a robot.

It is considered to have passed that test when the human believes they are seeing things on the screen sent by a human and not a robot.

If a human in another room were to see the information on this Captcha that this robot sent in... They would believe that a human were taking a Captcha and not a robot.

Test. PASSED.

And knowing that you're a human taking part in a Turing test is the most obvious indicator that you are talking to a robot and not a human, since we only give Turing tests to robots and not humans.

That is not a Turing Test. If captcha were the requirements for passing a Turing Test then the test would have been passed long ago. Why do you think there are a wide variety of different types of Captcha, from the click here ones, to the solve the maths problem ones, to the type these words ones? Because each of those individual tests can be passed but requires the coding to do so, which is complex. So by having a variety of different types of captcha makes it more difficult for someone to make a universal bot which can bypass all the varieties in use.

Also, your success at passing a captcha is determined by a robot, not a human. So this would only succeed in showing that robots can trick other robots which you can also do by sticking two chatbots in a conversation with each other and watching as neither will react any differently to a chatbot as they would a human.


Automatically Appended Next Post:

 sebster wrote:
Would you rather develop a computer that can trick you in to thinking it is a human, or a smart phone app that can diagnose a melanoma so you don't even have to leave the house?

The first one because it will result in a lot less people clogging up A&E because their phone told them they had cancer when they tested their mole which they had had for years but thought might possibly have got bigger, maybe.

 A Town Called Malus wrote:
The first one because it will result in a lot less people clogging up A&E because their phone told them they had cancer when they tested their mole which they had had for years but thought might possibly have got bigger, maybe.

You've assumed diagnostic technology will always have the high false positive rate it has right now. In effect you've chosen to judge existing diagnostic tech against a potential future system that could pass a Turing test.

This thread brings back some fond memories of the Ebola thread where somebody was convinced that it meant we were all going to die.

But then it turned out that's not how things really work and we are all going to live. Doomsday threads are real rollercoaster rides.

 sebster wrote:

 A Town Called Malus wrote:
The first one because it will result in a lot less people clogging up A&E because their phone told them they had cancer when they tested their mole which they had had for years but thought might possibly have got bigger, maybe.

You've assumed diagnostic technology will always have the high false positive rate it has right now. In effect you've chosen to judge existing diagnostic tech against a potential future system that could pass a Turing test.

Unless that phone app is capable of performing a biopsy, that will be the case as it can only examine it via observation.