[Discussion] What is AI?

If a game is a type of simulation then the NPC characters within it are a type of AI, right?

The thing is if you look at nature from slime moulds to humans to whales they all react to their world or simulation with a level of 'Intelligence'.

Intelligence is at least a scale or spectrum and could actually be mapped to a multi-dimensional space.

IMHO Intelligence is just a problem solving and communication mechanism that boosts the survival skills of a species.

Some people seem to think that if you mention AI you are referring to near or beyond human Intelligence.

Please take a look at the development of AI, it's history or just ask Siri a question then when she answers ask why and repeat.

IMHO if it runs on a computer and solves a complex problem that would normally take a person some time it's somewhere in the AI landscape.

Hopefully the people hung up on PCG can discuss AI here, have fun!

 

Most games use PI rather than AI in my opinion. PI being Programmed Intelligence. I prefer to use the term AI for things where the solution is machine learned from example data rather than directly programmed. Like Drivatars in Forza:

 

On the topic of PCG as an AI system...

Previous AI challenges were driving and shooting with a Tank vs another Tank or the more complex controlling a group of soldiers vs another group.

Both only required basic coding algorithms to make for a decent competition entrant.

Neither could be classed as IBM Watson level AI.

However both were behavioural agent problems that we in game development class as AI, even though most game AI is probably around the slime mould level of intelligence and complexity.

IMHO any system that solves or attempts to solve a complex problem is somewhere on the scale of AI.

Most of the time we attach the AI term to agent based systems and this is probably our Human Centric thinking kicking in.

As a thought experiment and to try and break our Human Centric thinking, is Life an AI system?

Life uses a set of basic programming codes RNA and DNA then tries different solutions to maximise it's growth using Darwinian evolution. The interplay of lifeforms creates a more complex web that has filled every habitable zone on Earth. Life on Earth is probably the most complex thing within a few light years (side though what if the sun could be alive).

On the face of it life is just lots of self centred plants and critters at war with each other for resources, however a complex web or balance emerges from simple things interacting.

Now ask any AI researcher what level their AI is at compared to life and they would probably tell you they are around the small mammal stage e.g. mice/cat.

Or look at the classic computer cellular automation life, where even a simple set of rules can generate complex patterns and behaviours.

My point is don't be so narrow minded about AI, if you can only use it to make a better NPC, you could end up missing out on a galaxy of possibilities.

 

With a galaxy of possibilities I'm guaranteed to miss out on most of them
AI on the machine learning scale isn't something I've looked into much, I'm happy enough working on the PI scale of things.

 

None of the options presented in a poll are correct.

An AI is an expert system that is geared towards solving some sort of problem while responding to external stimuli.

https://en.wikipedia.org/wiki/Expert_system

 

If we are talking about NPCs I would say mice.
that is a goal based AI, I think?

But is Siri really an AI or is it just a search engine bar with voice input?

 

LOL attempt to define AI then give a reference to Expert Systems!

https://en.wikipedia.org/wiki/Artificial_intelligence

This says it all the field of AI expands and improves, solved problems they are no longer considered AI.

So if the field of game level design has not been solved then it is still a field of AI research, this of course is good for level designers as their job has not been automated yet!

 

I would say much lower than Mice, as mice exhibit social behaviours can solve mazes and physics puzzles. They can also learn from feedback and other mice.

Most game NPC's have a very simple behavioural tree with little to no learning ability.

So maybe insect or reptile level (at a push).

 

Best be safe and keep it at insect level then. And even then we might be underestimating insects.

 

I don't care. AI in games is exactly what I listed.

As for "smarter than mice" - as far as I'm aware we don't have any technology capable of producing a system even remotely comparable to biological brain. The best attempt so far was openworm project which attempted to simulate C.Elegans using mere 5.5 teraflops of computing power. C.Elegans has 160 neurons total. A mouse has 4 million neurons. A human has has 86 000000000 neurons and up. We will be able to have a stab at it when exascale computing will evailable with Petabytes of RAM on it.

So your poll options are nonsense. It is like asking "Is human more similar to a bar of Iron, bar or lead or a cubic meter of argon"?.

 

From the other thread but so on the point about the AI Effect -> https://en.wikipedia.org/wiki/AI_effect

 

This was a fun read. Tesler's Theorem: "AI is whatever hasn't been done yet."

But really. If we call whatever we have now AI, how are we going to refer to the next big thing that comes out? Should we group it with pleb s**t that came out yesterday?

 

LOL The problem is once we have human level AI, can we say, that's just computation!?

 

@neginfinity is on point here. That you guys even think we are at insect level is pretty insane.

If we were capable of duplicating what ants are capable of with AI, the world would be an entirely different place and most human jobs would be gone.

 

Yes. it is just computation.

The main problem with getting human level AI is low computational power and low memory volume that is currently available. Exascale computers are only planned at the moment. When you'll have openworm running on a wristworth, then you'll have a chance of getting human-level AI.

 

Didn't you JUST do a thread about AI and its uses in games and multiple other threads that closely relate to this same topic?

https://forum.unity3d.com/threads/a...be-wary-of-bringing-ai-into-our-games.456996/
https://forum.unity3d.com/threads/c...processing-unit-be-the-next-processor.464892/
https://forum.unity3d.com/threads/training-an-ai-using-human-instruction.467967/
https://forum.unity3d.com/threads/procedural-animation-via-a-neural-network.469031/
https://forum.unity3d.com/threads/w...ld-be-applied-to-your-games-art-style.470511/

 

Quite probably, it's a fascinating topic with many aspects, implications and future impacts.

Note: the only reason I started this thread is another one on doing a PCG 'AI' Challenge got so much flack over using the term 'AI'!​

 

For the context of designing video games, there are two kinds of AI used:
1) Game AI - these are structures of code that can give a decent approximation of intelligence without using too much CPU time. The efficient nature of these routines make them ideal for using in games, but these are not usually capable of learning.
2) Academic AI - these are structure of code that can create math formulas based on sets of data. There are some cases in game design where these can be useful, but these often use too much CPU time during the training phase to be used in games.

Neither of those types of AI are easily judged as a specific level of intelligence that would fit into the poll you posted in this thread, but these are the types of AI that are used when developing games so these are relevant to this forum.

As a general rule, the lowest level of AI that can yield a decent experience for the user is the right amount to implement. Any more than that, and you are wasting CPU time which would lead to a less impressive game. Players don't care how things are implemented. Players only care about how the experience feels to them while they play the game.

 

AI is so broad. For games I've always seen it as behaviors. Not limited to characters though. If something is collecting data and then using that data to make decisions IMO it is a form of AI regardless if it is being labeled PI or ML.

And that is pretty generalized too but that is how I see it. Each character will have some form of AI (well we know they don't always... sometimes it is simply follow this path, turn around and go over there... stay on the nodes... this is not AI to me.... but) the overall higher level mechanism controlling when the character breaks out of canned patrol state and investigates to find the source of a noise or identifies the player and engages the player in combat.... this higher decision making system is an AI in games.

Weather systems can fall under this classification as well. Again is it all prescripted... precanned linear or is there decision making involved?

Just my opinion on it.

 

I mentioned PI as I see a real difference between a decision system based on observed patterns and a system that looks for those patterns to figure out how to make decisions. Does the AI Effect mark the moment when those patterns are found and the system changes from an AI one to a PI one?

Code (csharp):

1. if (ai == decisionMaking) {
2.     Print("I'm an AI script!!");
3. } else {
4.     Print("I'm a PI script!");
5. }

 

Arowx you should start a YouTube channel and put all your interesting ideas into it. I would subscribe.

 

Does the system actively solve a problem, or more importantly, does it realize it's strategy doesn't work and builds a new strategy? Just consider that once a solution is rote, there is no intelligence involved.

 

But in terms of your question....

I would say animals like cats are pretty basic. They don't have much learning capabilities (unlike dogs). They are basic state machines. With modes like:

[hunt] [sleep] [eat] [drink] [fight] [do that thing where your hairs on your back stand up when you see another cat] [go sit on something warm] [jump] [bring a mouse back to the house] [sit on something warm]

Each state probably triggers an inbuilt controlled behaviour. Every behaviour of a cat could probably be classified under one of those things.

They probably have inbuilt control systems in able to walk and primitive vision systems in able to recognise other cat shapes. Which is more from a process of evolution than of having to learn these things. (Think of a giraffe, it can walk in 3 seconds after being born without having to learn it).

Probably the only really AI thing a cat does is build up a map of its surroundings in its brain. And be able to return "home" to be fed.

 

I think you are selling cats short my friend. There is more there than you are realizing. While most pets never graduate past a stage of complete human dependancy this is not always the case. I have a cat that will take food scraps from the trash lay them out on the floor then wait on top of the stove for a mouse to take the bait. Learned behavior no doubt. He came up with it on his own after watching mice eat food scraps.

On top of that all thinking beings follow the basic triggered response you described for cats. Or brains just calculate based on input and the experience of previous outcomes.

 

Blood is based on iron so we are closer to a bar of iron, lead is actually poison so it cancel the human mass, and argon is a gaz at least we have consistency so bar of iron, case close

 

If we want to be all sciency about it argon is only a gas at a specific temperature and pressure range. Matter has a tendancy to change between its 4 states so if u want to compare solidity you have to qualify it with a temperature and pressure.

 

Usually when this sort of definition comes up, the "at sea level, room temperature" is assumed.

 

Which means you never had a cat in your life.

Cats have learning capabilities, when they're interested.They figure out by themselves how to open doors, how to get into places where they aren't supposed to be, and they also learn what their favorite food looks like. Sometimes what it is called. They also train their owners to react better to their meows and select meows that works the best to get what they want.

 

Agreed... would it not be ok to compare us as closer to oxygen as opposed to iron since we are predominantly made of that element even though is a gas? Does state carry more meaning than atomic composition?

 

Now I need to think why humans are still considered to be a "carbon based lifeform"

 

I have cats. They're pretty dumb to be honest.

I would explain the videos like this:
1) walking down wall: The cat has remembered where to go to get his back scratched.
2) the cat likes playing with the lock.
3) if you video a cat 100 times, the cat will pick the right answer at least once.

 

Well, for example: https://www.petcentric.com/02-25-2015/cat-dog-intelligence-are-pets-smarter-than-children/
Then we have good old wikipedia:
https://en.wikipedia.org/wiki/Cat_intelligence

Basically, a decent estimate would be that a cat or a dog will have intelligence around 1..2 y/o kid except speech.
The difference is that dogs, due to different social structure, attempt to please you, while cats don't care about it. Capacity for learning is there, but if you want tougher animals to train you could try rodents (rats, hamsters, squirrels, etc) or reptiles and snakes.

Cats are not einsteins, but calling them statemachines is ... not right and implies that you're severely underestimating them.

 

lol.. is this really descending into an argument about cats vs dogs? @yoonitee I gotta give you credit dude, that is some top notch trolling... wow!

 

Not just human but all known life. It has something to do with 4 valence bonds and being able to readily bond with itsself forming long complex chains. Im no chemical engineer but @Kiwasi can probably tell you more than anyone cares to know though.

 

Looked online. Found this:
http://www.dailygalaxy.com/my_weblo...sed-life-form-to-be-annouced-at-2-pm-est.html

Apparently there's a bacteria that uses different set of elements.

 

Interesting yet its currently widely considered false by others in her field.

Edit : also it was replacing phosphorus with arsenic supposedly and not replacing carbon with silicon or another similar element.

 

Yeah. Because when we've done it, we can examine it's workings as "just computation"
Show's over boys! Off to the next level!

 

CA are nothing like "life" or intelligence. In fact in many ways the concept of intelligence is actually the exact opposite of what one examines with CA, with complex systems, with emergence. The whole idea of emergence is higher level order arising from lower level simplicity. In other words, there is no intelligence in the agents, but their interactions with one another create the illusion (the effect? Chinese room, anyone?) of intelligence.

Now, interestingly, you might take this as an example of why we don't actually need AI: if you create agents of moderate complexity with a high dependence on interaction, you can create a dynamic environment which feels truly alive to the user. And if you give the agents a stochastic element, you've done a bang-up job of creating some of the unpredictability of life.

See this cool video: https://drive.google.com/open?id=0B_4ddvkmO_m7QXUxWHE1UWNSZ2c

Each of those pixels is completely unique. I could run that thing 100 times (and have) and get 100 completely different results. It's a relatively simplistic example, but you can apply the same functionality to systems with more complex agents, and hence get infinitely more complexity.

Consider the ant, previously mentioned in this thread. They are exceedingly simple and yet can produce incredibly ordered (and seemingly intelligent) behaviors. When an ant discovers food, for example, it releases a certain chemical and heads back to the nest. Other ants, when coming across this chemical, follow the path and pick up some of the food before heading back themselves. As they move on the path, they release the chemical as well, reinforcing the trail for yet more ants to follow. Another example: when ants die they release a certain chemical (if I remember correctly, it smells like grapes or grape juice). This tells other ants that the ant is dead, and they move it out of the nest to a "burial ground" of sorts. Well, this chemical can be dabbed on to living ants, and when other ants encounter this one - they pick it up and move it out to the burial ground. Of course it comes rushing back, but it is taken out again - and again, until the chemical wears off.

So ants are exceedingly simple. But their interaction with each other, and interdependence, creates higher level complexity.

You see this in many other places as well, such as weather. There's nothing "intelligent" about the weather. But the extremely high interdependence of weather on both time and influences from living beings means that weather acts in strange and seemingly intelligent ways. We'll never be able to truly predict the weather (both because we can't know the inputs to great enough precision to model them, and because we can't know the past state of weather accurately (precisely) enough to model it). I could go further on about complex systems, and fractals and such (how long is the coastline of Great Britain?), but I'm already digressing to much.

Suffice it to say, we don't need to worry about AI - interdependent systems can create as much meaningful complexity as we could want. I believe we've seen that in recent years with the rise of systems-based gameplay - the most well-known example being Breath of the Wild.

I wrote 4 long paragraphs but they seemed a little pretentious so I removed them. Yeah, carbon has 4 valence electrons and wants 8, which makes it relatively neutral and willing to form bonds with most substances. Oxygen on the other hand is very electronegative and unstable.

 

Human has no intelligence because it's just a higher level order behavior arising from simple neuron with simple mechanics like excitation inhibition?
That's a good thing computer will never be intelligence because there is no way intelligence arise from simple charge and discharge of transistor ...

 

I'm not sure if you're trying to say that I'm saying that (I didn't) or positing it yourself.

Edit: This gets into the question of what intelligence is, into whether humans truly have it or not (the Chinese room being an analogy in this area). The reality is, there's no definitive way to determine what intelligence even means or if others have it, but we typically use the term with assumptions like cognition. If we use that, then the agents I'm referring to there have none, or effectively none.

 

We aren't predominately oxygen (24%). We're predominantly hydrogen (62%). People who make the claim that it's oxygen are mistakenly comparing weight and forgetting hydrogen is much lighter. Additionally they're basing it on our percentage of water (53%) forgetting that water molecules have two hydrogens for every oxygen.

https://en.wikipedia.org/wiki/Composition_of_the_human_body

 

If there was one thing that defined intelligence, it would have to be adaptability. No other property is so highly valued by nature, otherwise humans and cockroaches would not be the most successful species.

That's not to say cockroaches are especially intelligent, rather they are simply robust and simple giving them a naturally wide range of adaptability. Whereas humans are neither robust nor simple, but are adaptive through intelligence - cognition, ability to model alternate outcomes, do long term planning etc. So adaptability is not the same as intelligence, but intelligence requires adaptability .. perhaps you could say that intelligence is one of the two forms of adaptability at either end of the complexity scale.

So I think real AI is a question of adaptability, which is way outside the scope of most game 'AI', so in game development the term is really just a buzzword for a series of switches that lends a small amount of apparent intelligence to a game character.

 

Defining intelligence can be tricky. Is phototropism intelligent behaviour for example?

Note to self: Philosophical wormholes are maybe not the best thing to post on game dev forums.

 

You are absolutely corrent. We should refer to them as "no bottom in sight" recursions instead.

 

First day back from holidays and I'm already being dragged into an arowx thread.

We are primarily hydrogen, oxygen and carbon. There is a bunch of other stuff in trace amounts, but none of it is critical to life as we understand it.

The term 'carbon based life form' refers to the entire system of carbon, oxygen and hydrogen. But it's mostly a science fiction term, as far as we can tell no other alternative life forms exist.

Silicone can form similar structures to carbon, so a silicone based life system is theoretically possible. It would still make heavy use of hydrogen and oxygen. But silicone is far less common in the universe then carbon. And carbon systems are more stable. So it's highly unlikely a silicone based system can exist.

The other possibilities are even more far fetched. Basically no one has taken the ideas seriously outside of star trek.

There is a remote chance that an AI system we develop may become a new form of life. But even this is still a faint, distant possibility. I'm skeptical that mechanical life could outcompete organic life.

 

I didn't realize we had multiple chemical engineers. I was actually hoping for a wall of explanation. I knew it had something to do with the chemistry of carbon and it being necessary for life just not 100% sure on the details.

I didn't feel i was in error using mass as the deciding factor rather than percentage of molecules. Either way arent we still closer to a gas of some kind than a bar of iron?

Ps thanks I rarely speak to people who can think at all.

 

Not a chemical engineer, though I do have a master's degree in Chemistry (or at least I'm 6/7ths of the way there) and have taught rudimentary chemistry for most of the last two years.

Yeah it can be argued that the universe is predominantly made up of nothing at all. Atoms have a clump of protons and neutrons in the center and a few electrons buzzing around very, very far from the nucleus with lots of empty space in between.

 

Now we must define life. Where do you draw the line. What makes one thing living and another thing not? Do we go by the standard capacity to grow, metabolize, respond (to stimuli), adapt, and reproduce. Depending on the looseness of your definition for a few of those terms we are nearly there if not already.

 

Dude... This is elementary school stuff...

Why don't we just use the current definition?
*looks up wiki*
Oh f***...

1:

2.

3.

It's down to field, then?
So (as devs) we have to define life strictly based on behaviour (like 3rd point)?

 

Yup. The closer you look, the harder it gets to define life, and to distinguish it from the background noise.

Biological life as we know it is simply a complex chemical reaction. On the borderline, stating which chemical reactions are alive, and which ones aren't is tricky.

But to qualify something as alive I would consider the primary characteristic to be self perpetuating. Secondary characteristics include local reductions in entropy, energy consumption, self replication, response to environment.

Honestly a chemical engineer isn't the best to tackle this question. We typically do large scale chemical and manufacturing plants, and leave the chemistry up to others.

That said, organic (carbon) chemistry is taught at high school. Forget the four bonds, this isn't important. What is important is it can form bonds with itself in all directions. That's what makes carbon so special. You can keep building out with carbon in any direction you like. That means you can create virtually any structure in 3D.