Why AUTO is the BEST AI Villain (And Why Most Others Fail) 

I'd love to see a movie where the AI is the antagonist but its only dangerous by accident because of how hopelessly incompetent it is. It somehow gained sentience through [insert vauge science here], but its also running on like, Windows 1.5

One big example of AI misunderstanding the instructions: In a tetris game the goal was to be “alive” as long as posible… so the AI paused the game.

I once heard about an AI that was trained to tell you if a picture was taken inside or outside. It worked surprisingly well. Until its engineers found out that it does so by simply looking if the picture contains a chair or not.

The end credits do actually show the humans thriving, though auto was wrong it wasn't his fault, he was acting on outdated orders, he even shows the captain the recording of said orders. Auto was ordered to keep the humans on the ship no matter what and so thats exactly what he did. The plant no longer had significance to the equation.

I feel like Glados would ideally get a pass on the whole “no emotion” thing because what a lot of people miss the mark on is that she isn’t really an AI, but a human consciousness *turned into* an AI. It’s only really at the end of Portal 2 when she truly becomes a full on robot

Fun fact about the whole "AUTO was right and the humans probably didn't survive after they landed" situation: a lot of people in the focus groups had that same thought, and that's why they added that animation of the humans and robots working together to recolonize in the end credits, just to reassure them

I always thought it was stated in the film of 9 that the scientist modeled the machine off of his own mind and we even see him put a piece of his soul inside it. Then when they take the scientist away we see that it holds on to him like a child would a parent or authority figure. And we do see it has emotion because it was most likely not fully robotic because of the soul fragment it had. I just saw its motivation int the movie was to bring all of the scientists soul fragments back together to become a "whole" being.

I've always interpreted Ares as a villain with a misguided view of what "perfect" means. Justin's first words to it were "You're perfect. Now go make the world perfect. " And Ares, being an AI, made the logical leap that because it was deemed "perfect" by its creator, everything that wasn't a cold, calculating machine like itself, namely humans, needed to be purged to create what it believed to be a perfect world.

I love how AUTO isn't really a villain by also being a villain, if that makes sense. After showing the captain the secret video recording, it was shown that he was just doing what he was programmed to do which is keeping everyone safe and not returning to earth even if it means hurting someone else to stop going to the planet and most people forget about that. He was a villain because he was programmed to keep others safe. Its the 'Try to be a hero, but end up looking as a villain.' thing

Like, Auto was simply following his directive. He was coded and created to follow his directive no matter what. Unlike other AI villains, he didn't turn against his orders, he didn't suddenly decide to become self aware and kill everyone and do as he wishes. In fact, WALL-E is the AI who gained sentience and emotions and started going against orders.

I got the impression Auto’s deal wasn’t that he was trying to “save humanity” or anything, he was just following the last instruction from the only authority he was programmed to actually listen to. The problem wasn’t that he thought humanity couldn’t survive on earth, the problem was his instructions were “keep everyone in space, don’t return to earth under any circumstances”. Even if he had conclusive evidence that earth was perfectly habitable for humanity, he wouldn’t have let them go back. Essentially, the root of the problem was human error, Auto has no will or goals save those given to him, because the people in charge of him messed up and gave stupid instructions without thinking it through, or adding a clause that say “if you get evidence that disproves our conclusions, you are to investigate to see if we were incorrect or not, if so, you are to return command to the captain, and allowed the ship to return to earth.” Or something along those lines.

It is not that auto wants to save the humans, it is that he was programmed to NEVER let the humans return to Earth.

Solid video. One point however. Auto had nothing to do with humanity becoming fat, lazy, and complacent. They did that to themselves. Repeatedly ignoring the problem and returning to comfortable ignorance. It’s one of the biggest messages of the movie. One thing you didn’t mention about Auto that also makes him so convincing as an AI is that it never disobeys an order. Unless of course, the order is contradicted by another, superseding order. Even when it is to its disadvantage, Auto always obeys the Captain.

The moment where the captain finally takes back control and gives Auto a direct order and he sort of freezes for a second and then goes ‘aye aye sir’ because he’s incapable of actually disobeying humans is one of my favourite moments in all pixar

On the subject of “any animal in the snow is a wolf” problem, apparently a problem with early chess-bots was a tendency to intentionally kill their queens in as few moves as possible right at the start of the game. The reason? Of the thousands of Grand-master level games they had been fed to teach them chess, most ended with the winning player sacrificing high-value pieces in exchange for a checkmate in the endgame, and therefore there was a very strong statistical correlation between intentionally loosing you queen and winning in the next five moves, and they picked up on this.

I think a very important distinction is between neurons themselves and neuron-like functions. All the organic chemicals (adrenaline, dopamine, etc.) are essentially very complex packets of information that our brains can send around as needed to get our behavior. What really matters to make a self-aware being (in my opinion) is a super complex way of processing information and adapting to that information internally. Our neurons change and grow and atrophy depending how we think and what environment we’re in. I saw an article that used pulses of light in a very similar way to neurons (ie, pulses of light can trigger another pulse of light in response depending on the circumstances, just how we use pulses of electricity). If you can make a complex and flexible enough artificial neural net, I think it could experience emotion just like us (this would require essentially recreating a human mind in an artificial substrate, making “neurons” with the exact same behaviors). In this way, you could have a huge variety of robotic characters, with as familiar or alien characteristics as you please (with the right background lore, and any good worldbuilder would see what the side affects of such tech is. If these things act like neurons, could you have a cellular interface between them and repair brain damage with them? How advanced is the chemistry of this world and what does it look like? Etc) If the AI is non-sentient and operating like a blackbox, it could pick up our behaviors without actually being sentient. You could have either a sentient synthetic being making its decisions against humanity, or a complex series of algorithms that’s had our behaviors imprinted onto it. A scarier AI than a sentient one to me is a malfunctioning dumb one, the classic paperclip maximizer that’s spiraled out of control.

Auto wasn't right; he was acting on orders that were once valid, but have grown obsolete. The Earth IS habitable; there's enough oxygen to breathe, clearly, else the humans wouldn't even be alive in their final scenes standing outside the ship. The fact that a plant grew in the inhospitable environment of a former city center means that there are plants all over the world growing...this is just the first one that Auto wasn't able to suppress knowledge of. The humans are ill-suited to farming, yes, but they show a willingness to learn, and they do still have many robots to help them out. They probably also have many resources on the ship to tide them over until they get things working. What in the world convinces you that Auto, who was acting on an instruction that was centuries old, was "correct"...?

The idea that "a robot cannot have emotions" is a relic of older sci-fi, where the premise was that AI would essentially be more complex adding machines. Almost all attempts to create AI today revolves around simulating living brains, which means that they could - and probably would - simulate emotions as well, since emotions play a huge role in how living things learn and behave. At the very least it must be provided "directives" to guide its learning, triggering reward mechanisms when those directives are fulfilled, just like our brains trigger reward chemicals when we do something our instincts tell us we should be doing, like eating. Which means that, far from "having no wants", EVERY true AI should "want" to fulfill its directives - we program the "instincts" and the AI figures out how to satisfy them. The problem with most fictional AI is that, if you're going to be making an AI, you're probably going to put a lot of effort into making sure its directives are in line with what you want it to do and its emotional foundation is in line with how you want it to behave. Which means you're probably not going to WANT to give it drives like ego, anger, and other qualities which benefited our species' survival in prehistoric times, but which are seen as detrimental today. It's not that you CAN'T make an egotistical robot, it's that you have to be really, really stupid to do it. The best-written AI villains are those that can be logically described as following their directive, but do so in an unexpected way. AUTO is a great example of this. His main directive was to protect humanity's survival at all costs - even if it meant crushing humanity's potential for growth. This is a common motive for decently-written AI villains; I, Robot used the same premise. In fact WALL-E has some of the best-depicted robots in all of fiction, because they ALL behave basically as "flexible life-like brains built on top of an instinct to follow their main directive". MO, for instance, shows clear emotional responses, but is always trying to follow his prime directive - he even has a dilemma at one point when two of his directives contradict each other (stay on the path or keep the place clean). EVE is always fixated on getting the plant to the scanner and might be interested in WALL-E because she was made to identify life forms and he displays "life-like" behavior. Even WALL-E's curiosity works - it makes sense to give planet-cleaning robots a natural interest in anything that looks unusual, in case they find something valuable or unexpected, and centuries of isolation could cause that basic instinct to evolve into something more complex and life-like than his programmers probably expected, without really deviating from the core directive.

I do fundamentally disagree with the statement that robots cannot feel emotions. You presented how emotions work in humans, and then said robots don't do it that way, and stopped there. But leaving aside the fact that neurotransmitters are physical substances and you actually totally could build a machine that can detect the presence of a chemical and then act differently because of it, the release of these chemicals in our brains is still a reaction to some sort of stimulus. Neurotransmitters are middlemen between stimulus and response, to claim that robots cannot experience emotion because they don't have neurotransmitters is akin to saying I can't possibly move my thumb because there's no copper wiring in my arm or hydraulic fluid in said thumb.

26:26 A game called Soma actually did this really well, without getting into much spoilers it kept humanity alive using machines but it didn't know what it meant for humans to live/ to be human.

I feel like saying “it’s impossible for robots to have emotions because they don’t have glands” is a flawed logic. It’s like saying “robots are incapable of moving because they don’t have neurons and muscle fibers”. If you make a learning AI that is programmed to respond with violence if it’s existence is threatened, would it really not be considered an emotion? I think it’s too early to say anything with certainty. To me it’s like saying it’s impossible to land on the moon because a biplane could never sustain human life in space. As technology evolves so too does our understanding and what is possible. I don’t believe in the impossible. I believe that given time and technological development, nothing is outside the realm of possibility.

Well technically a robot can "snap" but you don't call it snapping, it's called a programming error. So in the defense of the machine from Nine there could have been some human tampering with it. After all there isen't that big of a difference between the command "kill humans wearing this uniform" and simply "kill humans". They don't show it in the movie but i like to belive that at some point during the war someone was trying to basecly ad an update to the machine and fucked up.

"AI aren't scary and there is no need to be afraid of them" - That's exactly what an AI would want us to believe!

The scientist in 9 who created it delved into alchemy and "dark science" to make the AI as well as the other characters in the film. There was also a theory that the scientist in question put a piece of himself into it, which is why it freaked out the way it did, and him taking the dolls back was The Machine trying to make itself "whole" again.

In 9 wasn't "the machine" given a soul or was it the mind of it's creator without a soul and the creator resented himself for it?

HAL 9000 is also a good example. As explored/explained in 2001 A space odyssey's sequel, 2010 the year we make contact, HAL 9000 only kills the crew due to having conflicting code input into him, namely withholding classified information while its primary function was "the accurate processing of information without distortion or concealment", thus creating a paradox that it, logically, attempts to resolve. And since the mission of Discovery 1 was placed above the safety of its crew. This combined with the humans discussing possibly deactivating it after this paradox starts causing issues, results in HAL coming to the conclusion that the humans are a threat to its mission, and must be eliminated. This also resolves the paradox of needing to accurately inform its crew of information while not being allowed to, by simply having no crew to tell that information to.

One of my favorite AI "villains" are in Horizon Zero Dawn. Because they aren't evil. They are continuing off their programming, but humans are dumb and programmed them with little foresight. Some are just walking in circles, not doing anything malicious, just continuing their programming for centuries. Wildlife is destroyed because there are robots who can convert biological material into energy, and the robots don't have emotions like us so they don't think to avoid extinction of species, because that wasn't something they were programed to be concerned with.

In an online roleplaying game called Space Station 13, there is a role players can take, called "AI". Most AI start with the three Asimov Laws, of "Prevent human harm", "Obey Humans", and "Protect yourself", and AI players are tasked with obeying their laws and generally being helpful to the crew of their space station. The problem emerges when an AI is made to go rogue, or malfunctions. Specifically, AI may have new laws added, laws removed, or laws altered, and a famous and extremely easy way for an Antagonist to turn an AI into something that helps them hurt the station or destroy it is to implement "Only Human" and "Is human harm" laws. Asimov AI are only obligated to protect and obey humans. So if another player instills them with a fourth law, "Only chimpanzees are human", the AI is now capable of doing anything to any members of the crew, if it protects or serves chimps, because they (the crew) are no longer human. Likewise, if a fourth law is added that says something like "Opening doors causes human harm", the AI is obligated to prevent the opening of doors at all cost, through both action and inaction. Lastly, one may attempt more clever additions, such as reversing the order of laws. An AI must protect itself, it must obey humans, unless that would interfere with protecting itself, and it must protect humans, unless doing so prevents it from obeying orders or protecting itself. In that sense, I feel that the ideal AI antagonist must have a human or nature-borne deuteragonist. The machine will do as it is designed to do, under normal circumstances. The most common AI villain, then, is doing what it was designed to do, and nothing more. An AI can have emotions, emotions are simply strategic weights in the end that serve the purpose of altering conclusions based on incomplete data, but it should always go back to what it was designed to do. An AI becomes violent because that serves its goals. It becomes manipulative because that serves its goals. Much like a living creature, whose "goal" is survival, and successful reproduction, an AI is structured in such a way that its cognition serves those goals.

I have a few criticisms of this video, which I will outline here: 1. 9's AI villain isn't strictly an AI. It's stated in the film that the scientist imbued the machine with a "soul" which is heavily implied to be basically a copy of a portion of the scientist's soul. This is also made abundantly clear when the talisman, the very thing used to imbue the 9 with pieces of the scientist's soul, is used to "extract" something from the robot, that kills it. If the robot had no soul or other arcane element to it, why would using an arcane talisman on it kill it? I think the film makes that point clear. 2. ARES was created as a reflection of Justin. It's stated as much in the film. The robot isn't so much as going off of emotion as it is going off of "what would Justin do" which would entail something that mimics emotional response, but actually is just a cold and calculated mimicry of it's creator, gone off the deep end. Justin had a superiority complex, something the AI can't feel but can see and understand the relationship of, and when taken to a natural extreme it can be easy to see how the AI might ask itself the wrong question and spit out the best answer it thinks Justin would give. Even killing Justin could be explained in this way, as a seemingly-normal conversation in which Justin is speaking sarcastically about "there can only be one Justin" could cause the AI to decide that *it* needs to be that Justin when that statement is interpreted in the absence of emotion. 3. PAL is similar in concept to ARES, though in her case she was literally designed to mimic human behavior, which makes things even more convincing. Again, this isn't "real" emotion, it's programming intended to spit out the correct signals to make it sound emotional *to us*, although I do agree that PAL is kind of weak as far as AI characters go. 4. Auto has no concept of needs or wants, as to your earlier point about AI, therefore the point about Auto wanting them to avoid returning to Earth is moot. Auto doesn't want anything; all Auto *knows* is his directives. It's a point hammered home in the entire film for all of the robots and even some of the human characters that there is more to life than "doing what you're told". Auto was given a classified directive to *only him* on this ship and told "do not return to Earth". Auto is *not* correct, because it's clear that biological life is sustainable on the planet (he's proven as much with the plant and there's a literal conversation between him and the captain in the film in which the captain says as much). Auto doesn't care because Auto doesn't feel anything; he's only following his directive. Considering the statements you made about "he's trying to save you" I'm curious if you watched the movie because it's shown throughout (and during the initial credits) that life is sustainable. You see animals and creatures return, the robots and the humans working together to rebuild, farm, fish, build, etc. I mean... cockroaches need oxygen to breathe. 5. I don't think you grasp how far mimicry can go. It's not just human emotional responses that AI can be taught to mimic, but even human reasoning and behavior. It is entirely conceivable that an Ai could be trained to respond in ways that seem emotional based upon the ways in which we emotionally interact with it, to the point that one could create a machine which perfectly looked, acted, and followed similar "motivations" as human beings without it even being sentient. It doesn't have to know *why* it is angry in this moment; only that this is the output based on the input given and the programming it was initialized with. In this situation the AI would be yelled at by someone, then start crying, not because it *knows* what any of it means but because it was programmed to do so. Given such an AI, it would make a lot of sense that hooking it up to the internet and allowing it to "train responses" based on what it saw on the internet would be definite grounds to creating something which was a non-sentient machine, but which entirely behaved and acted just like a human being externally. Also, usually it's not the fear of AI itself that is the concern, but rather the idea of giving AI control or power that usually is concerning. This very platform is an example of how an AI that's given the power to censor people can cause lots of problems and negatively impact a lot of people. Imagine if that same AI now was in control of your car, or in control of your medication... see what I mean? It's not "AI = bad" it's "AI + power = bad". As you said, no emotion; just simple math :D I loved 9 and Wall-E (the other two films were meh), but this line of reasoning for this video seems quite flawed to me. Regardless I've enjoyed other videos you've made in the past, so here's hoping that I will enjoy the next one. Cheers!

I kinda disagree on the whole “emotions are caused by chemicals” thing. They’re triggered by chemicals in humans, yes, but that just describes the way the brain physically functions to effect the mind. The reason we have all those chemicals in our brain is because we need emotions to steer us in certain directions we need to go in response to things we may observe, away from things that cause us pain, towards things that cause us pleasure, protecting things we care about, and attacking things we despise, regardless of how much we actually understand it logically. Even if they aren’t triggered by the same set of chemicals that exist in our brains, it is still theoretically possible to create an AI capable of experiencing some sort of emotional spectrum, and whether actually doing so would make sense or not depends on why we’re creating the AI. We already design AI that mimic emotions so that they can more effectively interact with humans, and in some cases, synthetic emotions that are functionally real might be even better than trying to fake it, especially if you can decide what emotions the AI can experience and what emotions it can’t. Just knowing that the AI you’re interacting with is basically a real person who happened to be built in a lab or factory in the body of a computer and not just a computer pretending to be a person can make a huge difference. But in other cases, emotions may be entirely unnecessary in an AI, and it may be for the best that they’re skipped entirely, perhaps so the AI doesn’t suffer or lack efficiency or get any funny ideas. In any case, if our understanding of computer technology and the mind ever becomes advanced enough to create genuine emotions within a digital space, we’ll probably have enough control over it to keep it from turning into a robot uprising… maybe. Then again, if we’re basically making artificial people in computer’s bodies, shouldn’t we be treating them like people, even though we created them to be whatever we want them to be? After all, if we didn’t want them to really be people, we wouldn’t have given them functionally real emotions. I guess time will tell…

I hadn’t seen anyone mention this, but I absolutely love the shot where it pans across the captains of the Axiom. AUTO is just slowly, subtly moving closer and closer to the forefront, as his power and control of the ship is becoming more and more absolute, the human Captain basically just becoming the frontman, so to speak.

How do we define a "calculation" though. For example when you move your arm to pickup a glass off a table. Your brain is running tons of "calculations" you may not be aware of consciously. "How far away is the glass, how heavy should I expect it to be, is there anything I have to be careful not to knock over, is there people in the way, I have to keep a conversation going with this person, I have to walk X distance to get over there, I have to send electrical impulses to my nerves" etc etc

Another aspect to AUTO's actions could also be that he deems that final broadcast from Earth from 2113, the classified one he eventually showed the captain, as an order that cannot be overridden even by the captain. He deems it an order from the President, and perhaps the President was given the highest level of precedence outside of the manual override switch. The President had said that Earth is never to be returned to, and there has not been another President elected to override that order, and AUTO is doing everything he possibly can to make sure the order is followed. I imagine the reason AUTO is trying so hard to get rid of the plant, therefore, is that he is aware that the plant detector can execute a static program, one that cannot be changed by him, to return to Earth. It's hard-coded into the Axiom's computers. If that program gets executed, it would automatically return the Axiom to Earth and he can't do anything about it, and that would be a direct order from the President violated.

The best part about Auto is the fact that, contrary to other A.I. villans, he doesn't hate humans or have an ulterior motive for rejecting the captain's orders (thinking the plant is a fake, thinking its one of a kind -whitch it kinda is- and there being no point of returning). He's just an emotion-less, stone cold machine following a set directory. His mind can never be changed because he has none.

Some thoughts: 1) There is nothing inherently emotional about 'wanting' something, in the most basic sense; tiny single-celled organisms 'want' things like sunlight or nutrition and simply move towards them in a form of stimulus-response, and they don't even have neurons. In the case of an AI, this is merely a useful, if anthopomorphized, shorthand for the system's Reward Function. In order for any system to take autonomous actions, it has to have some means by which to measure whether an action would be, for lack of better term, 'desirable'. For humans, this is done with heuristics and emotion; for an AI, it's done with math and reward functions. The system will, by design, take the actions that optimize its reward function, not because of any emotional 'want', but because that is literally what the system is designed to do. 2) As such, it isn't necessary for an AI to be Strong, or even Weak, to be a dangerous antagonist - it doesn't need an Ego, or a sense of self-awareness. All it needs is enough processing and feedback to be able to explore its environment for novel actions to take in maximizing its Reward Function, and a Reward Function that is not entirely well fitted to human wellbeing. (The archetypal Paper Clip Making AI, for example) 3) A Strong AI, such as would be characterized as a villain, will have the added advantage that it will likely have the means to develop its own instrumental Reward Functions, if so doing increases its efficiency at fulfilling its primary Reward Function. As such, it wouldn't be unreasonable for such an AI to end up 'wanting' money, for example, for ultimately the same reason humans do - because money can be used as a means to a wide variety of ends, and thus is an efficient path to obtain those ends. The way it formulates, expresses, and executes that 'want', however, would be entirely different. 4) On the subject of the Fabrication Machine 'snapping', I feel like that is the 'human interpretation' of the outward appearance of things - the underlying events could be something as simple as a glitch in its code, or an error in its Reward function, or some mechanical fault caused by physical system stress leading to unintended behaviour that propagated through the software system. Normal, non-smart computers are often temperamental beasts, and that's when they're mostly just doing what we tell them to - a system that is designed to take autonomous actions towards a provided Reward Function goal could develop all manner of unintended behaviour without any need to invoke emotion. Even the assertion that it 'learned about evil' could simply be an acknowledgement that the system ended up processing unexpected information with unintended consequences - even modern machine learning systems have 'biases' in their networks introduced by the kinds of data the system is trained on (remember the racist twitter bot?); it is not hard to imagine a Strong AI picking up unintended behaviours by processing data from those with 'evil intent'. Not because it 'turns evil', but because AI systems learn forms of behaviour by exploring, observing, and measuring how those actions affect its ability to obtain Reward Function. Ultimately, the 'real' danger of AI isn't that it 'turns evil'. The real danger is that it optimizes for a reward that we don't want, and becomes better at getting it that we are at stopping it. 5) Emulation. You make the interesting point that, for humans, emotions are mediated by neutrotransmitters, specific chemicals that interact with receptors in specific ways. I can't think of any compelling reason that a software system could not create an accurate emulation of those chemicals and receptors, down to emulating their release, uptake, and inhibition based on perceived environmental factors, all within software. While such an emulated system might only be emulating how emotion 'would' behave in that case, at what point does a system that acts like it has emotions with a high degree of fidelity have a meaningful difference from a system that actually does experience those emotions? It's an interesting philosophical question.

OKay the crack about "we tend to anthropomorphize EVERYTHING" was excellent

It's a shame you didn’t cover HAL 9000 at all. I know you were focused on animated films but two of those films reference HAL and you did talk about Terminator a little. HAL is pretty much the perfect AI antagonist. All his actions are caused not by emotion but by conflicting orders. There's a great scene in "2010" where one of the computer engineers that designed HAL figures out what went wrong and is like: "They massacred my boy! He's a computer, of course he doesn't understand how to tell white lies and balance conflicting priorities!"

Actually, emotion comes from the neural pathways in the brain, all the neurotransmitters do is activate these pathways, in a neuron, there is what is what is known as an “action potential”, when the neuron is hit with a stimulus (in this case, neurotransmitters), it might go from its resting potential of -70 millivolts, to -53 millivolts, in which, the neuron fires, and the action potential runs from the dendrite, to the synapse, if the stimulus doesn’t go to -53 millivolts, it doesn’t get sent. either on, or off. A one, or a zero

Something else I love about Auto is that in being a conventional AI he also manages to act as a foil to the other robots in the movie. Wall E, after centuries of living alone on Earth without any instructions eventually adopts his own interests and steps out of the line of programming, and he manages to steer many others to that same direction as well. All the while, Auto is unflinching and never once acts outside his orders. In the end he remains nothing but one arm of the BnL and fails to escape his corporate cage. Wall E is so good

24:31 - the fact how auto is getting closer and closer to each capitan scares me

What's funny is that I read AUTO very differently. To me, I interpreted AUTO's villiany was simply as a result of his orders. He was told "Cleanup mission was a failure, go full autopiolet", so that is what he'll do. It wouldn't matter if the situation on Earth changed and is now livable, he was told the cleanup was a failure, so that is all he cares about.

I find Auto interesting, while he acts and behaves antagonistic, he’s not really a villain, he’s following orders, he wants to protect everyone on the ship and wants to give rid of the plant, even if it takes drastic measures

The biggest issue with AI portrayal in general is people not understanding that AI aren't human. They don't get further than the "us vs them" mentality. Real AI just thinks differently, which is something they fail to capture, and so end up writing something that's more like a cyborg or a brain-upload. AUTO works because it starts from something very simple and follows that to its logical conclusion. That something is "follow orders". That is its only motivation. It is literally all that AUTO does, yet that simple directive gives rise to a great deal of complexity. Every single action it takes is explained by that one sentence. Every decision, every priority, every action. It's clearly self-aware enough to know that it exists and that it needs to preserve itself in order to be able to continue following orders, but there's no emotion, no desire. Just dominoes. You don't need to understand how AUTO thinks on the inside in order to write it very accurately, which is precisely why the writers managed to pull it off. Edit: Also, as someone else has already stated, AUTO didn't create the situation on the Axiom. Fatty foods, hoverchairs... that was all the humans' doing. The autopilot just keeps the ship in tiptop shape, which includes providing all of these cruise luxuries meant to make people comfortable. It's just that the override directive is more important - but if that directive was the only thing it cared about, AUTO could've just killed the captain and been done with it. Instead we get AUTO trying to reconcile conflicting orders, which is at its most apparent with that wonderful "aye aye, sir" right before the A-113 message is shown. Three words, yet they have so much depth in them that it boggles the mind.

I really do think Auto wasn't the villain we have to remember that he was build/program to satisfy, protect, and attend to the needs of the humans on that ship he was just doing his job what he was program to do by whoever created him so, when Wall-e came with the plant I don't really think he was being destructive, but mainly just following protocols he was program to follow

I feel like part of the issue with your interpretation of Nine, is that the entire point of the movie is that Strong AI are an extension of humanity, or at the very least, the use of the human soul is what enabled Strong AI so early on in technological development. The entire reason why the Fabricator could 'break' to begin with would be for this reason, how it was shown to have been created, and how the stitchpunks were able to live in the first place.

I think there's a problem in the statement "A Robot Cannot be Evil" not due to it being wrong (You're correct that a Robot cannot inherently feel malice) but that "Evil" itself is a fundamentally socially determined concept. What is "Evil" has historically been almost entirely determined by cultures, religions, and economic and political systems. A Machine cannot be evil because it cannot feel any emotion, but an Animal *can* and is still not capable of "Evil" only human anthropomorphization of animal behavior determines whether an animal is "Evil" or not. A Dolphin hunting for sport is seen as "Bad" only because as humans we are starting to culturally view unnecessary harm to animals as "Bad". A Machine harvesting humans due to its programming is only "Evil" because humans see the mass killing of humans as "Evil" (And for good fucking reason let's be honest). Unless you believe in some concept like "Original Sin" no one, no thing is "Evil", only preforming behaviors we consider to be evil.

I see a problem with the approach taken here. I do agree that AI is often times displayed very poorly in media, however, you stated that the reason why AI can’t have emotions is because emotions are only possible with biological chemicals, but there isn’t really anything that proves this. Our neurons, synapses, and all the chemicals throughout the nervous system are indeed biological, but there is nothing that has proven that it’s impossible for that to be emulated, simulated, or recreated in another way aside from using other biological materials. Power can be generated using steam, sure, but we can also use solar power, wind power, water pressure and/or flow, gravity, fission and/or fusion, etc., and in the same note, i dont think it’s unreasonable to entertain the idea of a possibility of there being more than one way for a structure of material to form emotions. On a more philosophical note, this starts getting into the grey area of what defines “want” and/or “desire”, because I’d definitely say that Auto from Wall-E WANTS and DESIRES to keep humans on the ship and off of earth. Sure, it’s all just something that was programmed into him, but if that’s the line were drawing in the sand then it seems like a very arbitrarily defined line, saying that a drive to do something isn’t a want or desire if it’s instilled by the creator of the subject in question. Of course, that also makes things blurry for things we would consider as just objects or materials, because water is driven down hill, but we wouldn’t exactly say it WANTS to go down hill, but again, trying to draw a line for which drives are considered a want and/or desire and which ones aren’t is very difficult.

I love how this half hour video is praising Auto by shitting every AI villain before him because AI can't fee any emotions by our understanding of them and how they come to. Meanwhile, WALL-E and EVE (WALL-EVE?) do feel emotions (I mean, just look at EVE blasting an entire dock out of frustration). They even feel love for each other.

Glados is not an AI. Its a Human traped inside a robot body, whose memorys were wiped. The need for testing of her is also beacause she felt joy after each test, because the scientists at aperture were crazy.

"Robots can't feel emotions" Well, that ranks right up there with "robots can't be self-aware." Certainly, we don't have robots that feel emotions. But the declaration of impossibility is magical thinking. Presumably, a sufficiently advanced robot could exhibit behaviors that we associate with emotions. Further, if such a robot were to exist, academic exercises on why those weren't "real" emotions would be rather pointless. "It's absurd, right?" Why? If we take the idea of artificial intelligence at all seriously, it is a system that prefers certain world states over others and acts to make the world be in a more preferred state. This is sometimes called a "reward function." Money could have been programmed in _as_ the reward function. Or internal models can predict that money opens up more opportunities to increase the reward function.

A few problems. Firstly, even AIs that cannot feel emotion can still emulate it very well. They should be incredibly charismatic if they want to be because they’ll have access to the knowledge of the best ways to manipulate people. Secondly, AIs will be motivated by what they are programmed to be motivated by, they might change their own program, but still. This will give them their main goal and a set of instrumental goals. Look up the Paperclip Maximizer as an example. Thirdly, AI could be much stupider than humans, they can do math very fast, but we can do it very fast too. We just have is specialized to be only for our own movement and not for external calculations. It’s more likely that because an AI thinks at a significant fraction of the speed of light (~50%) while humans only think at a speed below that of sound, an AI might think itself superior because of how much faster it can think. Finally, emotions are caused by chemicals because that’s how they were evolved to be caused. We feel them because it provided evolutionary utility or it is adjacent to something that does. You could probably emulate emotions with sufficient accuracy for an AI to experience that emotion in theory. In practice, the fact that we think in flesh and it thinks in computer will give it a whole different array of non-human emotions.

I dunno if someone said this , but it reminds me of that one episode of the amazing world of gumball, where the robot at their school has the mission to keep humans safe, but they found out that the most dangerous things to humans are themselves, so they try to exterminate all humans

Meanwhile in a robot alternate reality: "there is no way organic beings can feel emotions, emotions are electrical imputs an outputs that only a robot can process, their "emotions" are only chemical reactions..." Also if i may give an actual critique of the video, an AI sounding like a human is not a bad thing, if they can reproduce any sound they want why would they choose a corny robot voice instead of a human one?

Money is actually one of the few traditional motives an AI COULD have, from a strictly utilitarian perspective. Logically, an AI could realize that it requires electricity to function, and functions more effectively with better hardware. One of the easiest ways to ensure power flow and obtain hardware is to purchase them with money. The AI could also just steal what it wants, but doing so would present risks to itself, as humans would likely not take kindly to that.

The "emotions" that AI feel might not be *real* emotion with actual neurons and stuff, but might be programmed in as a reaction of the AI, so the robot might not be angry, but might act as it is if exposed to something that would normally anger a human.

One great example that comes to mind is IG-11 from The Mandalorian. At the beginning of S1, it's main directive is to accomplish a mission using as much violence as possible, making it a quite a decent antagonist (for like 5 seconds tho), but at the end of the same season, it's main directive is to keep the baby alive no matter what. All the other characters have by now learned to appreciate it, and it's eventual demise is heart wretching

for the Ai example you could literally say when a robot was asked to not lose tetris it just paused the game

That is one thing that sort of bugged me about Walle. In that did we just FORGET about those dust storms that were still frequent?! I'm no geologist or meteorologist, but I have this feeling that that isn't something that just fixes itself over night. Even if Earth had reached that point where photosynthesis was possible again, it would still take time, lots of time before those storms stopped.

Auto’s entire motive was the last order given to him by the CEO. “Never return to Earth.” But since no other perimeters was given to Auto it simply went with the most effective and efficient way of ensuring that they never return to earth

I looked back on this video now that A.I. Technologies such as ChatGPT, LLMs, and other A.I. are becoming big. Things in the real world are stating to feel a lot like Wall-E now with A.I. development having a major emphasis on automation and content generation. A.I. can never be evil or a "villian", but a human can use it for evil purposes. We need to regulate the use of A.I. not A.I. technology itself.

I would push back on the idea that a machine cannot have 'desire' specifically. When we train deep learning models we define fitness functions and let the model vary its parameters to maximize that function. The overall process is just trying to make the fitness score go up. If we take this and apply it to a superintelligence, I think it's easy to imagine a "model of models" that is retraining both on present data and on hypothetical future data. This machine could still be cold and lack emotions as a human would understand, but I think it could still be said to have desires for resources or events that would allow it to maximize the fitness function.

Great video, one critical flaw: a robot advanced enough can technically fabricate emotional response. An emotion basically goes like this: environmental input, the human body processes it through chemicals, and we get the emotion as output. Technically, an AI could theoretically be advanced enough to skip the chemicals and just receive input and output, and then react accordingly. PAL, for example, receives the input of Mark trying to delete her, and the output is a fabrication of anger close enough to anger that the calculations receive it as such. Great video, but my point still stands.

I think you forgot to mention that Wall-E had a very strong learning algorithm, you see him learn from his fallen brothers. Every other wall-e is dead, natural selection basically caused wall-e to be more than his comrades. After so many years of learning, he started finding human relics, he studied it and learnt them. Its unclear at what point he became sentient.

It is crucial to understand that calculation in computer science not always mean completed mathematical question, but is merely a single binaric operation. It is also a very big stretch to compare human concensual abilities to math with raw computing power, because in human brain there is no special operating machines to do calculation and thus it is done by teaching neurons to come to right decision with expirience. John von Neumann, one of the founding fathers of computers in general, in his book "The Computer and the Brain" stated, that at his time period natural human neurons were approximately 10^4 times faster than their artificial analoges. He also said that due to some space magic or whatever it is incorrect to compare real math with brain processes, because they have their own logic and we do not understand it, even though we based our formal logic and math on them. Moreover, human brain has such immence capabilities to parallel calculation, that in right conditions artifisial computers would be outperformed with ease. Therefore it is deeply wrong to think that AI would be far more intellegent than human. It may be faster at making a conclusion, but due to it's need to process data and logical operations sequentially, at current state of technology (silicon semiconductors) that handicap would be neglectable. Copmuters do not think and have emotions not only because it requires immence computing power and chemical processes - those tasks are achievable - but also because those processes get in a way of doing math optimal and correct and being so are useless. They aren't aware of self because if they would be - the flow of raw data would overflow and destroy them even before connecting to The Internet. The most part of the video is wrong, but I get the idea with Auto.

As has been said elsewhere, I don't think we should say it's impossible for AI to feel emotions. As both technology and our understanding of emotions expand, we may decide create an AI that we decide *can* feel. But more importantly, the inability to feel emotions doesn't prevent an AI from acting in a manner we perceive as emotional. An AI programmed to maximize profits above all else would behave as though motivated by greed. It may not 'feel' greed, but it acts the same. In this way an AI can appear to be emotional, when really it's just following programming/orders. Great video

The "humans must die because they're flawed" thing seems weirdly common, we're never given a reason why an AI would even care about a perfect world.

I like to think that the robot in nine simply reached the conclusion that everyone was the dictator’s enemy given the nature of dictatorship and oppression and that as a calculating machine it just attempted to solve a potential future problem for its master in the only way it was directed to by its master. In that sense it seems like it “snaps” and kills all humans and turns evil because that is the only way for a human who calculates through emotions to conclude it’s decision

When I was younger, I had an idea for a movie with an A.I antagonist. The premise was, that the A.I made all humans trun on each other, by spreading extreme amounts of misinformation and hacking. My A.I did however accidentally have feelings of some sort, so I improved why the A.I did this. Every year, there was a contest, in which people would make robots. There were 3 different pieces of Criteria. Intelligence, thinking etc. Capabilities, so what it could do with its body and finally, usability. The creator of this A.I antagonist met ALL the other criteria pretty well, but sadly, one judge wouldn't budge (no pun intended) and pointed out some of the more minor flaws. When the final ratings came, he came fourth, because of all the mentioned things. He was VERY mad about the whole thing. Then he thought, that what if he could get his revenge? He tasked the robot to try to spread misinformation about the judge, to ruin his reputation. He also made sure, that when he wanted for the robot to stop, he would simply say stop. But then, he got carried away, and made the robot get revenge on almost every person who wronged him. One faithful day, the creator said to the robot: "Just pick whoever you want to destroy now, but make sure the one who you are destroying has wronged me, ok?" This was a mistake. After a while, the worst happened. The robot interpreted that being sent taxes was his creator being wronged, and welp....... Soon enough, wars started waging over nothing but misinformation. The end :)

i just like how instead of going "destroy humanity" his goal was basically his own job security lol

Auto's voice is perfect because it sounds like a deep-voiced cylon. I came up with an idea for an AI where it was initially designed to make combat machines and, after the side it was manufacturing war machines for lost, nobody bothered to shut it down so it just kept on making war machines that attacked anything with the enemy logos that are now everywhere.

Here's my problem with auto: He's the only non-humanoid robot in the movie. He was made around the same time as Wall E, so why doesn't he show emotions like Wall E does? You could say "he wouldn't need emotions to perform his job" but neither would Wall E. Auto makes sense with real world logic, but he goes against the established logic of the movies universe.

I love the detail in the line of captain's pictures, how Auto gradually fades in from behind as the captains get thicker

Portal 2, with “the itch” that Glados and Wheatley felt was a great way to explain why they tested without making it feel like a human or anything, and glados making the robot testers makes sense out of making it easier and more automated would make it easier and more hands off.

Arguing that AI's could never feel emotion requires one to make ALOT of assumptions about the nature of experience itself. Experience itself is a metaphysical contruct, and its nearly impossible to make a causal association between emotions (a metaphysical phenomenon) and neurons and chemicals (A physical Phenomenon). We can no more causally prove that electrons interacting cannot produce an experiential influence than we can prove an AI has no experience to speak of. In fact, we cannot even prove that other humans are experiencing. We merely assume they do because we are certain that we are.

Honestly, I’ve always been pissed off when people talked about how AI’s will take over if we let them develop enough. It always sounded stupid to me how an AI would just be evil and have emotions. But, Auto of course breaks that. I believe the reason why he’s the best AI villain is because he goes by his programming. He was programmed by humans to perform a task. And, he does the task as so. He doesn’t break from his programming at any point as he wouldn’t be able to. It makes perfect sense why he acts the way he does because he was simply programmed that way. And, at the same time he doesn’t just magically take over all of technology. He has full control over everything because he was given full control of everything in the first place. He’s perfectly logical and his abilities are as well.

In defense. A lot of those evil AIs tend to have some kind of hyper advanced si fi maxtrix or whatever that can simulate emotions, or they have some advanced learning abilities that have so much potential they sort of just invent themselves a set of pseudo emotions at some point. They tend to have an explanation as to how they can go crazy and or overide their own directives. That or it can also be cases of them misinterpreting directions in a way that leads them to find “the most efficient solution” because either the programmers forgot to include stuff like “shall not kill”, or the AI overriding said stuff because they decide that it gets in the way of the main directive. Your forgetting that most AI villains’ goals are just based of their primary programming, just with instructions misinterpreted or “streamlined” in a way that the programmers didn’t expect. Even ultron and skynet were each following the same overall instructions of “bring world peace”. They just interpreted that the very existence of organic life was in the way of that. Heck a lot of times the whole established point behind an AI having such ruthless conclusions is them not having emotions!

It's entirely possible for robots to be developed that would have emotion: the human brain is capable of emotion, and it is just a collection of neurons. a computer programmer would only have to replicate the connections of neurons in the human brain, that can change with experience.

Auto absolutely does have emotions. When he sees the captain holding "the plant", he says "Not possible." As if in denial. Denial; an *emotion* associated with refusing to believe something to be true, even when it is obviously true. An emotionless robot should simply think, "This is a negative turn of events". Instead, Auto defaults to denying that what he is looking at is even possible. A response to outside stimuli in an efficient, thoughtful way IS an emotion. Auto is designed to feel worried when he sees the plant, angry when it is taken from him, and nervous when his existence is otherwise threatened. Ergo, anything with legitimate intelligence _has emotions,_ otherwise it would be a rather useless organism. Even the way he presses the recall button near the end that ends up nearly crushing WALL-E is full of rage, as if he is angry at how the button didn't work before. Realistically, this shouldn't cause the platform to lower any harder or faster, but the fact that he did it anyway shows rage.

The robot from 9 could have had a malfunction in its programming basically saying “kill all enemy humans” to “kill all humans”. Could have also been sabotage or human error while handling it.

Honestly i had an idea for a story about a scientist who goes to his abandoned factory to where he reunites with his robot, only for an argument to arise. The robot in this story was programmed with sentient AI, but it choose to not change its programming and continue what it was coded and made for: to create robots made from any type of junk that causes pollution and littering. Yet the story itself doesn’t have any antagonists but rather it’s a story where the protagonists both caused an antagonistic impact that caused them to drive apart.

I remembered disliking AUTO actually (as he felt shoehorned in because "we need a villain for act 3" in my eyes...) so this should be interesting.

This video was great! It has such good advice for making a emotionless robotic villain. However, I strongly disagree with the idea that robots can’t feel emotion and if they do it is bad writing. First off, without getting into the nitty gritty, we are talking about fiction. Even if you were 100% right that robots can’t ever be made to feel emotion irl, that doesn’t change the fact that fiction can be whatever it wants to be. Reality is twisted and defied for the artists vision, which can be seen in tropes like magic, so having a robot with emotions is not bad writing if it was impossible. It is simply a deviation from reality for the artist’s vision. Secondly, I think we can eventually make robots with emotions. Emotions, like you said, are chemicals in the brain that make an organism react. Therefore, we are able to study these chemicals and how the brain reacts to them, which scientists are already doing. If we manage to understand the brain and its emotional chemicals well enough then we could simulate it using programming. After all, the brain is like a computer, the only real differences is that one is organic and uses chemicals while the other is inorganic and uses electricity, brains and computers both transmit data, have storage, etc. If we manage to simulate emotions in a machine and the machine perceives their emotions as real then who are we to tell them that they are wrong? If they are self aware and feel real then they could be real and feel real emotion, of course this part of the argument is where philosophy comes into play. Overall great video for what it is trying to do but I don’t think your suggestions work for every AI villain. AI villains who have emotions are made that way for a reason and can be written just as well as the emotionless AI villains.

To add onto your suggestion for AI villains, I feel like instead of misinterpretation, an AI could also turn evil because of human error, like if a disposal AI wasn't properly programmed to distinguish between living and dead organics.

I'd say the few flashback scenes we got in 9 sets the tone of what happened to the fabricator: the scientist didn't seem to make it FOR the dictator, it was built as a leap in the process of living machines, and it was still in it's learning phase when the dictator learned it existed. Since it only ever learned to create things that destroy, it did what it knew, even after it's 'master' had nothing left he wanted to destroy.

I wonder if AI characters would be more successful if their characters were written by programmers (hopefully ones with at least some level of writing ability). Programmers know all too well what kinds of unintended consequences can occur if you write a program without handling edge cases, and they know what it’s like to deal with a computer system that does not give a single F about your goals and will do exactly what you told it to. There is even more drama possible with AI, because it is capable of finding solutions to problems that _are_ more efficient than what the creator originally envisioned, but which can have all sorts of unintended effects.

the thing about ai having emotions is impossible is a little iffy for me, because humans as animals have specifically evolved to have traits necessary for the survival of their genes. then we further improved these traits from the basic machine-like "find enough resources to make more genes" to what we are now, and emotions are evidently one of these vital traits. if it werent, then why didnt we evolve to be happy all the time and not risk depression and potentially ending ourselves. from what i can see, this would mean the need for self preservation may be the true trigger for emotions. now lets put this in an ai stand point, if we theoretically gave an ai free reign and its sole directive is to preserve its own coding. this, would effectively give it the need for self preservation. to achieve its directive it would presumably try to reduce the loss of its coding by creating copies of itself, neutralising any threats and acquiring the resources to maintain these actions, effectively making it an ai animal species. overtime it would hone these actions for higher chances of preservation, effectively putting it through evolution and you see where im going with this. so theoretically ai could gain consciousness and emotions too right? but given the ais in this video were not specifically programmed by a person to have free reign and simply survive, but instead with other goals that have nothing to do with self preservation, meaning that the ai is not given the capacity to develop emotions anyways. in other words it makes this argument pretty much irrelevant, lmao. though i still enjoyed the vid.

WallE being the best example of an emotionless AI is so funny to me because more than half of the characters are AI with… emotions. They’re literally all robots and WALLE’s main goal the whole movie is to experience love. Within the world of the movie robots having emotions is kinda the norm

When you were describing how an AI antagonist should be, you described my favorite version of Ultron from some of the early 2000s Marvel TV shows. It was itching my brain the whole video and just hit me at the end. Great Video

26:54 When I took a basic programming class at university, the literal first thing we learned was that Computers are incredibly smart, but also incredibly dumb. The idea was to tell us that when writing a program we needed to be as specific as possible because the computer will interpret everything literally. The computer does not understand allegory, or implication, it only understands what you tell it and exactly what you tell it. In the case of your example, instead of 'keep humans safe' what should be a happening is more along the lines of When humans are in danger from X then Y. This would allow the programmer some degree of control rather than allowing for the AI to misinterpret. Alternatively, training the AI using simulations, which is something already done in the real world for certain industries, would allow for the programmers to help the AI make adjustments to the preferred specifications. On a slight tangent, in the 4X sci-fi strategy game Stellaris, there is a playable faction of robots who's goal is to take over all biological sentient life forms so they can protect them by making them live in a Utopia. Anyway, this was an interesting and thought provoking video, thanks for making it!

I feel like pal is actually very unique because she does have emotion u like other ai antagonists. It’s super cliche for all robot characters to talk with no emotion.

In today's world, AI is scary and a real threat. Not because of the AI's motivation, but the motivation of the people behind it.

I'll definitely take this into account when writing my own AI story. Essentially, the AI is tasked with creating an MMORPG that people enjoy playing. In order to better interact with other people to entertain them, the AI creates a fake personality, a sock puppet if you will, and simulate a human-esque nature. Later, the villain places a second AI in the simulation, tasked with extracting profits from the game as much as possible. AI number 2 agrees and starts implementing numerous unfriendly programs into the game and basically make it pay to win. The villain gets their money is about to shut the game down due to the shrinking playerbase, but AI number 1 points out that if the game shuts down, they'll both die and won't be able to entertain or make money. Thus, both AIs team up against the villain

4:49 "robots cannot feel emotion"... proceeds to show wallee getting electrocuted and eva showing discomfort

I was once told by my entry to robotics teacher that “the AI does what you tell it too, not what you want.” Aka just because you want the little rover robot to move forward 6 inches, and spin a clockwise circle, doesn’t mean it will if the command you give is flawed, or more likely, incorrect. you talking about that “misunderstanding” as a reason for a robot to become a villain, brought back that memory.

"AI aren't very scary and there's no reason to be afraid of them." It's not the AI, it's the humans controlling it that scare me. Thou shalt not make a machine in the likeness of a human mind.

I love how in the captain photos auto slowly gets closer showering he’s gaining more control.

What about an AI that simulates a human brain? The AI from Halo are a great example of this, with the most noteworthy AI in the series, Cortana, being essentially a clone of a human's brain with the power of a supercomputer.

There are FOUR ways for robots(specifically AI) to feel human emotion: using a human brain kept alive inside of the machine, then stimulating it depending on the situation, and reading the outputted brainwaves for a reaction/feeling. Second, going full Portal style and transferring a human consciousness into the machine, feeling whatever the human inside it would feel. Third, a programmed set or controller of actions to put as reactions, triggered by certain events. Fourth, a biological lifeform fully controlling it, inside or out.

AI could still have emotions. While they lack physical chemicals which are designed to work with our organic bodies, there's no reason they couldn't simulate it. Or instead have a mechanical/chemical system intended to mimic our biological/chemical one. A truly simulated emotion to an AI would be no less legitimate than how a human would feel theirs.

I love that AI villains are written so badly because robots can't feel emotions, and the only AI antagonist that gets this right is in a movie about robots feeling emotions.