Videogame Turing Test of Future AI

[MrPopo]

Well, imagine an AI that can understand written language and respond to anything you type. It also stores its experiences to memory in a way similar to humans, and those stored memories shape its personality. If you kill the AI, that memory is deleted. It is so advanced and realistic that it behaves entirely like a human behind a computer screen. Would it be unethical to kill such a game character?

So in that case "kill the AI" would be defined as uninstalling the game. From a technical perspective the AI would be permanent and would control its in-game entities in a similar fashion to how you control yours. The AI would be put into hibernation when you close out of the game, but it wouldn't "die" until you removed the data files that make up the AI (uninstall the game).

I think a more interesting question is the intentional lobotomizing AIs would need to go through in order to make them fun opponents. Take a fighting game. Even without doing things like reading your inputs an AI can tell what attack you're doing at frame 1 and immediately execute the proper higher-priority counter. Today in order to avoid SNK Boss Syndrome the attack behavior for CPU opponents includes things like occasionally choosing the wrong move or a delay before you can react to a move to simulate human reaction time. Now imagine an AI that has more self awareness. It'd be like the scene in Robocop 3 where Robocop is conflicted between "Protect civilians" and "Obey the company".

[MrPopo]

I can program a computer to tell you it hurts. Characters in games even act hurt quite frequently these days.

That somehow seems different though, right? Like it doesn't respond intelligently to input. It just knows if x happens, behave like y. But what if it was a more advanced AI that took in multiple streams of information (auditory, visual, tactile) and then integrated that information in a way that was adaptive over time in the way that pain is, so that it learned from its mistakes. Like if it had a sense of what its physical breaking point was, and could use information from its senses to adapt to pressure, temperature, and other forms of damage that approached that breaking point. Over time it would learn to be cautious as it moved around its virtual world. Is that sort of programming stil different enough from ours that it would not be considered aware of its pain? When does something become aware? And how do we distinguish what is aware from what is not? Can will kill replicants Bladerunner style, or do they eventually, for all intents and purposes, deserve to be treated with the same level of dignity as a more traditional biological organism?

To get more existential on you, how do I know that you are aware? But to your point about replicants, I believe that core reason we don't indiscriminately kill other people is the social contract that says "I'll let you live if you let me live". Until such a time as human-like AIs can make good on the threat of "if you try to kill me I'll kill you first" I suspect we'll treat them as disposable.

[pankakes123]

First off, if that ever happens it will be in no less than 30 years. We're talking REALLY complex AI here. Also, I don't think any of us will find it to be bad killing them, but I'm sure some artificial intelligence rights group will act out for attention.

[DinnerX]

Computers simply do what they are programmed to do. If they learn in any since, it is only in the ways they are specifically programmed to learn. They have no feelings, emotions, or irrationality. We can attempt to program an appearance of these things, but an appearance is not the same as actually having them.

But I suppose we're not talking about computer's here. We're talking about video game characters. The world video game characters exist in is pretend, so how could it ever experience real pain?

[MrPopo]

Computers simply do what they are programmed to do. If they learn in any since, it is only in the ways they are specifically programmed to learn. They have no feelings, emotions, or irrationality. We can attempt to program an appearance of these things, but an appearance is not the same as actually having them.

But I suppose we're not talking about computer's here. We're talking about video game characters. The world video game characters exist in is pretend, so how could it ever experience real pain?

I'd argue that people only do what they're programmed to do. It just seems like free will because the inputs we can process are so complex.

[DinnerX]

Computers simply do what they are programmed to do. If they learn in any since, it is only in the ways they are specifically programmed to learn. They have no feelings, emotions, or irrationality. We can attempt to program an appearance of these things, but an appearance is not the same as actually having them.

I'd argue that people only do what they're programmed to do. It just seems like free will because the inputs we can process are so complex.

Could be. But we can't prove that either way. Computers we know for certain are programmed. Until it is shown that all human actions are as programmed as a computer's, I don't think computers will ever have a claim at being equal to us.

But I suppose we're not talking about computer's here. We're talking about video game characters. The world video game characters exist in is pretend, so how could it ever experience real pain?

I'm going to retract this. I could easily connect the computer's "consciousness" with its real-life experiences into a virtual world and it wouldn't be able to tell the difference. The pain it felt would be as real in the virtual world as it was in the real world.

If an AI was considered equal to a human, I don't think we'd be playing games with it at all. It'd be like enslaving someone.

[MrPopo]

That's the tricky thing about consciousness. I think at some point we'll develop a sufficiently advanced set of hardware and software that can reprogram itself in response to external stimulus and after enough iterations it will "wake up". I believe the key is to ensure as much stimulus as possible; today computers take in an extremely limited set of inputs compared to what humans do.

[Cronozilla]

Let's just be clear for a second here ... there's BARELY AI in games. What most people call "AI" in games ... is just a moniker. In reality those systems are deterministic finite state machines AT BEST. Which is NOT AI.

If we DID have better AI in games, the only thing that would happen would be that, basically you wouldn't be able to win.

AI involves breaking down the mechanics of a system and abusing tricks to win. You use clever searching practices to generate a better answer iteratively, and you do this over time to a degree that a human can't.

AI software does not feel.
AI software does not learn (that's machine learning, and that generally doesn't work very well ... or in real-time).

The core of almost all AI is an evaluator function. It's a mathematical formula developed to weigh potential choices in a realm of possibilities.

That's it. "Smart" AI ... wouldn't even give you a chance to win. (Ever played a computer at Chess that had all the tricks of the trade? You literally can't win)

What we want is iterative decision making that looks human ... but even that is a simulation and any sort of "feeling" that appears from that is purely for the sake of the human participant.

We have no technology, hardware or software, anything near the level of cognitive ability. Even if we combined every piece of computational technology we've ever made as a species ... it would not be enough. Brains are very complex things, and trying to make something conscious would require something in a similar realm of complexity (or at the very least, that's the first place you'd start looking)

Even when talking about those research groups who have ML robotics projects where they're teaching their software how to communicate and handle and recognize objects ... it's all pre-determined in a construct they've developed. And, any sort of "I feel sad" junk is because they were told to be sad in specific situations. It's not actual emotion.

For you to have an actually cognitive, feeling, thinking, species of machine, they'd either need to be thousands of times more complex than anything we're currently capable of (also consider AI and ML are relative new fields, they've only been around for about 40 years) and, it's likely something that would develop with out being specifically designed. (This idea is called emergent behavior)

Anyway. No. I wouldn't feel bad. It's not like you're going to leave the computer running forever, or that program. And to be perfectly frank, if you were really worried about some consciousness coming into existence when you found an NPC, you'd never move from that spot. The engine would only allocate memory if you were around, the moment you get out of some range, and it'd be deleted ... unless it's made by Microsoft, then it would have several memory leaks and an entire family of rogue AI NPCs would be living throughout your computer system :3