I'm not sure I understand the poll options, but I think the key point is that artificial "intelligence" is actually not "intelligent". It's just very very good at copying/mimicking, fast and at a large scale.
NB: edited for typo (as => is)

Odd question. If there is no human actively involved in guiding an AI through each of its tasks, it's 100% artificial. That AI is the product of humans and learns from humans is irrelevant for that definition.

I don't get your question(s).

But in general i can say i am interested in and suprised from many AI generated pictures. How good they are in a lot of things, fingers/hands, hairs, eyes excluded.

Most it's 99% AI with the final 1% being the prompt made by a human and the cherry pick of witch actually fit what the user was looking for, but there are some more complex stuff that still require some hand editing. (As far as I know).

Edit: I'm referring principally to Image AI, now I realize the question was more general about AI.

It sounds very interesting, Now with having AI capable of so many things, I'm fascinated by the fact that we can have an AI NPC or side character in games that could learn your patterns/gameplay style and act/speak accordingly. Especially now with AI being able to mimic all sorts of people and characters it would be possible for it to be applied to the NPC/side character's voice actor to say different things as it learns more. IDK this might be too much out there or game breaking or maybe even just bad design, but the idea of a evolving personalized character just seems so interesting to me.

That's a strange question. Artificial is not opposite to "human". Artificial is opposite to natural. As oxford dictionary defines this word - "made or produced to copy something natural; not real" And it's obvious that Artificial intellect, is something that was "made or produced to copy natural intellect". And, to be honest, everything that now called AI is technically NOT even an intellect, artificial or natural. It's just a generative language model. It's just a neural network "learned" to predict what words should go after some other words. Nothing more. So, you ask it a question - a series of words, and it "knows", that in learning dataset after those words most probably will go some other words - which is, since dataset is collected from real human interaction, usually is the answer to that question. Or not. If some question were answered many times in a dataset - AI predicts the answer pretty well. If the question was never answered before - usually AI will answer with some bullshit, that just looks like an answer, but actually is not. And that's it, nothing much more.

if you mean the AI's brain, then it is made mostly from a collection of human-made "training data", if you mean how the AI was made, it sort of makes itself, in a manner of speaking. https://www.youtube.com/watch?v=R9OHn5ZF4Uo explains it nicely

here is an answer by an AI

I'd say it's firmly in the "mostly artificial" category. While AI is designed and created by humans, it's not entirely divorced from our own intelligence and capabilities. We're the architects, but AI systems learn and operate in ways that mimic human cognition to some extent.

Everything from humans simply pretending to be AI, like the Mechanical Turk, in 1770, to entirely artificial, with no human input at all. The simpler it is, the easier it is to completely automate.

Current generative AI, which is what I'm assuming you're talking about, is trained on data sets, and gives what ever response seems the most consistent with what it has seen in that data. Of course when it sees something new, it just makes things up, and is often completely wrong, hallucinations. In that case, it is working entirely on its own, but is often guided when answering certain questions.

TLDR: It depends on the AI.

It would be helpful if AI were called by an accurate name, like Adapting Predictive Application. It's useful to recall, at all times, that AI does not have any idea what it's talking about, nor that it's talking at all. It contains nothing that would be able to house that awareness.

What I always turn to is this helpful repositioning of the question:

"The question of whether computers can think is like the question of whether submarines can swim." - Edsger Dijkstra

AI inspiration is 100% human. the only intelligence we know is human, and that is the one we try to replicate

AI at our current level is ever more based off of human activity, written word, and art.

It is easier to train an AI with masses of information across the internet, then just build an artificial intelligence from the ground up.

There's also quite a spectrum to AI that most dont understand.

Like when i say "computer," what do you think of? A calculator? No? Well technically it is. Also the oldest computers were analog and using punch cards and such, which is where the term "bug" comes from. They had to literally check through the machine for literal bugs.

So in some ways we've had AI since before the internet. But now they're trying to create AOI - Artificial Organic Intelligence, in order to replace humans and maximize profits.

Robotics have already been able to make human sized robots for quite some time, but without a decent interface, they're not much more than mannequins with servo motors.

So while they work on perfecting the ability to manufacture cheap robots en-masse, they still need a good human/robot interface, and that's why you've heard about chatGPT so much recently. Which in turn brings back up the ol' Terminator argument of SkyNet and the Singularity. When we finally make an AI so well, it can make more of itself, in a similar way to how we produce, and then takes over the manufacturing capabilities to expand upon itself.

Right now, I'd say I'm optimistic about AI as a supplemental tool but not as anything more than that. A couple games have come out recently that are based around the use of artificial intelligence, my favorite of which is called Suck Up! The premise is pretty simple - you're playing a vampire and using your mic, you have to try to convince the people in town to let you into their houses. The game uses AI to interpret what you say & formulate a response and in the cases I've seen, it actually does it really well.

Further into the future, I'd love to see how AI might let us to explore possibilities we've never had access to before. It could be used to train smarter enemies maybe, it could help flesh out open world environments, etc.

However, with that all said, we should be careful not to get too reliant on AI. The bulk of coding, art, writing, and such should continue to be done by humans in my opinion. Especially in AI's current state, it's a horrendous programmer on its own. A lot of the code it spits out is meant to be more of a guideline so it gets broken really easily.

You might be interested in this article, which breaks down the extent to which human help is still needed behind the scenes for AI (which is really just "machine learning," not intelligence) to parse any of its data.

It is an exploration AI with countless records of probes coming from outside the earth to study the reactions of earthlings and to investigate the intelligence and nature of life forms. It's always investigating humanity by sneaking onto the Internet.

I thought for a moment about making a joke about this, but decided against it because it might be considered fact. Cliche, right?🙂👽🛸🛰
Well, we live in an age when people are pretending to be human to create news and novels are being sold under the false pretense that they were written by humans....

I think we are decades way from AI having any significant impact on our jobs or on us. Currently these are simply tools, which most people aren't even aware of. I'm looking forward to the part when AI enabled robotics will do away with jobs of high risks or manual labour.
In gaming I suppose it will be very useful for developers, where they don't need to spend time doing mundane work bts. And can focus on aspects that need more attention.

Even with an AI who can make another AI in the possible future, since AI actually required at least a prompt, I would say AI is mostly artificial - there is always a prime mover or prima causa.

Well, I think...

Wait. Wait just one second.

i c A I o

?!

Frankenstein monster of the 21st century.

View attached image.

I think the problem is calling it artificial intelligence to begin with, it is anthropomorphism, us projecting human traits on non-human things. It really is a marketing buzzword. Better terms would be machine learning, pattern recognition, data mining, natural language processing, etc.

Virtual assistants and chatbots are all designed to mimic human-like behavior when you interact with them. These language models are basically very good at generating realistic-sounding language but ultimately they don't "think" or "understand" or "feel" because that's not how their work. The models were trained to predict the next token given all previous text (subject to context length limits), it is really is that simple.

A researcher coined the term stochastic parrots to describe them: https://en.wikipedia.org/wiki/Stochastic_parrot

The training process follows that idea, you take a piece of known text, withhold a part of it, and tell the model to predict the missing part. Then you make small "adjustments" to the weights of the model according to the actual output vs. the expected output. You repeat this process over and over until the model has seen all the training data, while taking into account how much it's improving as you go on, so you can adjust certain things like learning rate and momentum (i.e you start by making big adjustments at the beginning of the training, but you slow down these changes down the line). This is a well known algorithm in mathematical optimization known as https://en.wikipedia.org/wiki/Gradient_descent where it's used to search an unknown space for the "optimal" solution given some constraints and an objective function to optimize (i.e less error)

The next phase is called RLHF (reinforcement learning from human feedback) where the model is finetuned to match what we expects it to answer. So if you ask it how to do inappropriate or illegal things, it will refuse to answer. Basically a way to make the model more "politically correct". This is also done by showing the model many examples of prompts+outputs and telling it which ones are good and desirable and which ones are not, adjusting model weights accordingly.

The secret ingredient that makes LLMs this good is the scale of the data they are trained on, we are talking about models that contain billions of weights trained on trillions of text tokens taken from all over the internet. This includes books, research papers, encyclopedias, newspapers, social media posts, video subtitles, chat logs, source codes, law texts, medical texts, and on and on and on...

The technology itself is not exactly new, it's been around for decades, but the sheer size of the model and data is what made the recent breakthrough possible, something that was not tractable a few years ago given the available computational resources of the time. Computers got bigger and faster, which allowed us to feed these models with more and more data, and the models output suddenly got much better.

In fact researchers have somewhat hit a stop because they are running out of data! Yes, there's very little open data left to scrape and feed these models anymore, and they are looking for other sources to get massive amounts of additional data (think licensed data, private data, data behind paywalls, etc). Plus the internet have caught on to what these big AI companies are doing, and have been putting more restrictions and limits to stop them from just swallowing this data for free (think reddit/twitter new api limits, lawsuits from major news outlets, writers and artists, etc.). You see all of our discussions here (SG forum is open access) will too one day be scraped and fed to train the next batch of LLM models ;)

What we nowadays call AI are mostly just generative algorithms capable of "learning" human patterns by going over thousands of pieces of sample material and then replicating the style of the things they were fed by using a prompt as guide to turn literal noise into something that looks coherent at first glance. So it's not so much intelligent, more of a digital facsimile of what human brains do when starved of input.
Now, while there's zero risk of this disembodied little chunk of the ghost of a human mind birthed by our collective intrusive thoughts of ever developing sentience on its own, it is really weird that we as a species managed to simulate some of the processes our brains do despite the fact that we don't fully understand how the hell it works. We just randomly puked a small piece of an incomplete homunculus and decided that turning it into a tool was the obvious thing to do.

I assume the more an AI system learns, the more it's able to operate independently. When it's aware of its independence, we'll be at a crossroads. No idea what that looks like though.

AI is just maths.Lots and lots of calculations and computations."AI is scary" depends if AI is being supervised by human or not.Just like a Police Dog.Under care and supervision:not sentient and scary but very Dumb.On its Own:screaming intensifies.
This is what 4 yrs of studying AI has taught me.You wont believe how difficult it is to train a machine(Neural Networks).They are just Too Dumb(until they get TONs of DATA)yikes.

i don't know how many times 've read each and every comment up here. and really don't know how to say a good thank you. maybe doing generating gibs.

that's what's all about, our lil world. and i do like it.

hey, AI-Steam... generate me Factorio, pretty please?