AI is making big strides but should we be wary of bringing AI into our games?

Saying there are other people scared of AI is not the same thing as understanding the topic. Those links do not back up your initial post in this thread in any way.

 

You do realise you are using the same tactics as climate deniers, you push for finer and finer grained absolute truth without backing up your own argument in any way shape or form.

 

No, no they are not. You are doing it again. You are changing the subject and ignoring the MOUNTAIN of evidence you don't know what you're talking about. I have a working knowledge of pattern based AI and neural networking, you have no idea how wrong you are, this entire thread is a monument to it. You're willfully ignoring the people that know what they're talking about, calling out your vague, almost off-topic, ramblings. You do this in every single topic you start. You pick a buzzword that you have very little knowledge in, opine on the topic, and paddle right along as people rightfully correct you for talking out your ass.

Seriously, pop open Google Scholar, there are thousand of free research docs on the subject, look here's a good (albeit more math heavy) one I found in 10 seconds:

http://citeseerx.ist.psu.edu/viewdo...74F723A7?doi=10.1.1.86.3414&rep=rep1&type=pdf

Until then, it's fine to muse on a subject, but don't pretend you have more knowledge in the subject than you actually do.

 

What's the point of trying to back up our arguments? Some people are simply incapable of listening to reason regardless of how much proof is shoved in front of their faces. They will continue believing the climate is not changing, they will continue believing that the moon landings were faked, and they will continue believing that AIs from Doom are out to get us.

 

These threads remind me WAY to much of my edgy teenage conspiracy theory years back on sites like Above Top Secret.

 

Arowx, that is completely false. You don't even understand the topic well enough to have a conversation about it. Please read a book about fundamental AI concepts.

 

Was just checking the Unreal forums and I discovered he created a thread over there too.

https://forums.unrealengine.com/sho...ould-we-be-wary-of-bringing-AI-into-our-games

Let's see what their conclusions were...

Yeah that's kinda what I thought.

 

Arowx, I will briefly explain how modern AI works, in hopes that this will inspire you to read a book about fundamental AI concepts. Modern AI is nothing like a working AGI. Modern AI is basically just a computer program that tries to write a math formula that will fit a given data set. In most cases, modern AI is very simple, very narrow, and requires a lot of help from humans to determine what data is relevant and how to apply that.

By narrow, we are talking about an AI being trained on a very narrow set of information. For example, if you took detailed measurements of all of the parts of all of the flowers in a flower bed and then categorized those by name or type, then you could use that data to train an AI to recognize flowers based on those same inputs. The training portion of the AI would create a math formula that satisfied the training data.

Think of it like a loop randomly picking coefficients for each variable in a long algebra equation. Each time through the loop, a scoring function will score that equation using the data from the training set and compare its score to the best score from previous randomly generated equations. This is repeated a bunch of times until a reasonably good equation has been found. This is the simplest form of AI training.

Even the deepest neural networks are similar to this. Neural networks resemble functions that feed into each other. So instead of having a single long algebra equation, you end up with a lot of smaller ones that feed outputs into the inputs in the next layer. But in the end, it functions very similarly. There will be a bunch of inputs and one or more outputs for the complete system.

As you can tell from reading this post, modern AI as I described is definitely not sentient. It cannot look at new situations and automatically generate new solutions based on previous experiences. Humans still need to manually do that for the algorithms. Sometimes even the best humans require several tries to set up a useful modern AI system to serve a meaningful role in some type of modern automation.

Anyway, has this inspired you to read a good book about fundamental AI concepts, or are you still worried a modern AI will wipe out humanity after being trained to play an Atari 2600 game?

 

That is awesome. Different forum yet same response to Arowx's misunderstanding of AI topics.

 

Spoil sport I wanted to compare the responses between both communities!

Can you see a difference in feedback, style and level of topic discussion?

 

Yes. They're significantly less sick of your clickbait article titles and misunderstanding of topics because they haven't gotten to know you yet.

 

You should rate a thread on the topics and issues/discussions it raises not bias every thread just because you don't like the creator.

Kudos to the unreal community for being more open minded and willing to discuss a topic on it's merits.

I won't mention the topic(s) I've started with a pseudonym!

 

The Unreal General Discussion section feels active with many responses have double digit replies. If you a perform a search for this threads though you'll notice that the bulk of them have single digit responses which leads me to conclude that they're ignoring his clickbait or ignoring him.

https://forums.unrealengine.com/sea...1432&starteronly=1&contenttype=vBForum_Thread

 

Arowx, also keep in mind that there are plenty of us that really do love to hang out and discuss AI topics. The responses in this thread on this forum are telling you (Arowx) that you would benefit from reading a book about fundamental AI concepts. If you actually understood modern AI systems, there are plenty of people in this forum that would enjoy discussing various ideas about futuristic AI scenarios. But you cannot post threads about AI wiping out humanity after being trained to play an Atari 2600 game before you have taken the time to read about fundamental concepts.

 

Arowx, you really don't get it. People are reacting like this to you for reasons stated in this thread and your other active thread: you read the bare minimum about a thing, get all worked up into a lather, and then post about it without understanding it. When confronted with the fact that you have no idea what you're talking about you either try and change the subject or double down and start comparing people to climate change deniers.

Who you are has a massive factor to play with how you're received because you've been doing this for years now. People here know what to expect from you and you're quickly trying peoples patience.

 

Arowx, the comments I posted to this thread are not the result of like or dislike of you. Disagreement is not an automatic sign of dislike. My comments are purely intended to help you, even after you tried to liken me to a climate change denier.

You seem smart enough to understand these concepts and you seem driven enough to learn these things, but so far you have chosen not to read about fundamental AI concepts. This is the second AI thread you created where I encouraged you to do some reading, and I honestly think you could learn fundamental AI concepts if you took the time to read about them.

If you want to post about these topics in the future, please take the time to learn the fundamentals before posting.

 

Additionally many of the Unreal community members have an account with Unity. That last response I quoted shows that some of them are already aware of who he is and threads he creates.

 

No, we're sick of it over there too. Let me compare them for you, "Learn the topic you're arguing about." once again you are trying to insulate yourself from criticism. That is the general response from everyone. You got your answer, it's just not the one you wanted.

You're creating hundreds of nonsensical discussions, often times the same ones over and over again.

 

History Lesson:

What if I told you the RAF cancelled an AI fighter pilot program because they were too good and the technology was based on Norns cute teachable evolvable game AI (Creatures circa 1998)

Norns used a mix of genetic and neural network technology to make trainable evolvable digital pets.

Article about it here https://groups.google.com/forum/#!topic/comp.sys.ibm.pc.games.flight-sim/-NmesQBVHLU

And they learned to fly fighter jets in a simulator, against real pilots.

So these guys



Defeated these guys


Is that in your modern AI textbooks?

 

You have showed us that you cannot remember or were not able to learn from your previous AI threads, so the expectation I have for you now is to continue making new AI threads with the exact same premise while continuing to ignore or belittle the responses you are given.

Here have a quote from the actual person who made it unlike all the paraphrases people like to use.

 

I remember this. The AI isn't bound by constraints like blood flow during actual manoeuvres. It also didn't account for stress on the AI controlled planes. Not only that, but they didn't "learn how to fly jets." They learned how to apply forces that kept the simulated jets from crashing. Eventually they got goo

Basically, this is exactly the sort of thing that IS in modern AI textbooks under neural networks.

 

You need to start separating concepts. The fighter jet program was not a game, it was a simulation.

I will repeat, the fighter jet program was not a game, it was a simulation.

An AI trained in a high fidelity simulation of aircraft combat can potentially produce an AI that is highly dangerous in aircraft combat (if it was ever allowed into a real jet that it could interface with).

An AI trained as an antagonist on modern video games - which basically means to:

• Broadcast their presence all the time with "Who's there?", "What was that?" and "I've got a bad feeling about this..." in order to signal their presence to the player;

• Forget everything in 5 seconds "I must have been seeing things" in order to give the player another chance.

• Not be accurate when shooting the player, so that they give the player maximum chance of survival;

• Have a need to tell the player all about their 'motives' when they have the player cornered, followed by giving the player an easy opportunity to take advantage of them;

• etc.

... will learn EXACTLY THAT! Which (I'm sorry to disappoint you) does not actually make for a dangerous AI!

What you're doing is assuming that a game antagonist actually has the purpose of killing the player. This is wrong. AI programming is all about NOT killing the player when you easily could have, and masking bullet-provoking behaviour as deadly aggression. A game AI has one purpose, which is to die for the player's enjoyment.

Games are not simulations. Neural networks and genetic algorithms can be applied to things that can produce somewhat dangerous AI, such as aircraft combat simulations (which games are not). Mainstream video games != dangerous AI.

 

As a reminder, General Discussion is a place to discuss Unity topics not covered by subforums, and game development in general. Despite sounding similar video game AI is not the same as AGI, this discussion does not have a practical purpose related to either Unity or game development.

This topic is closed.