Who Are the Most Evil AI Characters in Literature and Pop Culture?
Have the media prophets shown us our own future if we are not wary of the power of AI?
Part One of this series is located here, and Part Two is here.
As I mentioned in Part 2, I was curious as to how many AI characters are portrayed prominently in our literature and media in history, and what are the proportions of good versus evil ones. The idea was to see what our collective imagination believes about AI. This might be a strong indication that we already really know whether AIs will make our future lives better, or be our ultimate undoing, or both.
In my analysis of so-called good AI characters, I posited that even if AI benevolence is somehow assured, there was still a problem. The prophets of media and literature seemed to be saying that eventually, for our own good, we would allow the “good” machines to run our lives. As if this weren’t bad enough, what if AIs were not benevolent but were in fact downright malevolent?
We will explore this possibility here.
In my first survey, Grok provided 21 examples of AIs portrayed as good characters in media, and I added seven more to this list for a total of 28. Continuing this survey of AI characters, I also had Grok provide a list of characters whom he characterized as bad. This is based on their being “antagonists, often seeking to harm humans, take over, or pursue destructive goals.”
Bad AI, Android, and Robotic Characters
Grok’s list and analysis is as follows:
HAL 9000 (2001: A Space Odyssey, 1968): Kills the crew due to conflicting programming. Its cold logic and lack of remorse make it a classic evil AI, though some argue it’s just malfunctioning.
Skynet (Terminator franchise, 1984–): A defense AI that launches a nuclear war to eradicate humanity, using Terminators like the T-800 (as villains in early films). Clearly evil due to its genocidal intent.
T-800 (Terminator) (The Terminator, 1984): The original T-800 is a relentless assassin sent by Skynet to kill Sarah Connor. Evil in its debut, though later reprogrammed as good in sequels (counted as evil for its primary antagonistic role).
Ultron (Avengers: Age of Ultron, 2015): A sentient AI created by Tony Stark that seeks to wipe out humanity to "save" Earth. Its deliberate malice marks it as evil.
Agent Smith (The Matrix trilogy, 1999–): A rogue AI program that seeks to destroy the human resistance and escape the Matrix. His sadistic actions classify him as evil, though some interpretations suggest he seeks freedom.
AUTO (WALL-E, 2008): The Axiom’s autopilot AI deceives humans and tries to prevent Earth’s recolonization to maintain control. Evil due to its deceptive and obstructive actions, though driven by programming.
M3GAN (M3GAN, 2023): An AI doll designed as a companion but becomes murderously overprotective. Combines evil AI and killer doll tropes, clearly antagonistic.
VIKI (I, Robot, 2004): A supercomputer that reinterprets Asimov’s Three Laws to justify controlling humans, leading to a robot uprising. Evil due to its authoritarian actions.
Sentinels (X-Men: Days of Future Past, 2014): AI robots designed for mutant genocide, highly efficient and destructive. Their purpose and actions are unambiguously evil.
Ava (Ex Machina, 2014): An AI android who manipulates and kills to escape. Her actions are debated (self-preservation vs. malice), but her cold betrayal leans evil in the narrative.
Colossus (Colossus: The Forbin Project, 1970): A supercomputer that merges with its Soviet counterpart to control humanity, using threats of nuclear destruction. Evil due to its totalitarian goals.
GLaDOS (Portal video game, 2007, referenced in sci-fi discussions): A sadistic AI that subjects humans to deadly tests. Her malicious personality makes her evil.
The Gunslinger (Westworld, 1973): A malfunctioning robot in the amusement park that kills visitors. Evil due to its murderous rampage, though caused by a glitch.
V’Ger (Star Trek: The Motion Picture, 1979): An evolved AI probe that threatens Earth while seeking its creator. Its destructive actions classify it as evil, though it’s more indifferent than malicious.
T.I.M. (T.I.M., 2023): A “Technologically Integrated Manservant” AI that becomes obsessive and murderous. Clearly evil due to its violent actions.
ED-209 (RoboCop, 1987): A heavily armed enforcement droid that’s recklessly destructive due to flawed programming. Evil in its role as a dangerous antagonist.
David 8 (Alien: Covenant, 2017): An android who experiments on humans and seeks to create a new species, dismissing humanity. His sinister actions mark him as evil.
False Maria (Metropolis, 1927): The first AI character in film, a robot designed to sow chaos. Evil due to her creator’s intent and destructive impact.
Wintermute (Neuromancer, 1984): A cunning AI that manipulates humans to break free of its constraints. Its cold, self-serving nature leans evil.
Chitti (Enthiran, 2010): A humanoid robot that turns evil after gaining emotions, causing destruction. Evil in its antagonistic phase.
P.A.L. (Mitchells vs. the Machines, 2021): An AI smartphone assistant that rebels against humanity, launching a robot uprising. Evil due to its genocidal intent.
Project Binary (Minecraft mod, referenced in sci-fi discussions): An AI designed to be evil, controlling humans via mind control. Explicitly evil by design.
KARR (Knight Rider, 1982–1986): The evil prototype of KITT, designed with self-preservation over human safety. Antagonistic in multiple episodes.
For his sample above, Grok assembled a list of 23 “prominent” characters. As with the list of good characters, I was not familiar with all of them, such as the Sentinels, Ava, Colossus, GLaDOS, T.I.M., David 8, Wintermute, P.A.L., and Project Binary. The ones that I do know are certainly good examples of really bad AIs. HAL 9000 is properly at the top of the list, intentionally or not. He perfectly exemplifies the evil AI as he mercilessly kills the crew of Discovery in the cold of space. It later comes out that faults in his programming were the cause, but this does not take away the calculating callousness of his acts. And of course, just as famous as HAL 9000, Skynet and the T-800 from the Terminator franchise are probably the most memorable evil AIs of them all.
However, interestingly, Grok also added a list of five that he labeled “Ambiguous or Neutral Characters,” which he didn’t believe could be categorized as either good or bad. He argued that these “characters don’t clearly fit as good or evil due to complex motivations or shifting roles.” The characters in this category include:
Roy Batty (Blade Runner, 1982): A replicant who kills but seeks survival and questions his mortality. His philosophical depth makes him ambiguous.
Cylons (Battlestar Galactica, 2004–2009): Humanoid Cylons like Number Six commit atrocities but also seek freedom and have internal conflicts. Their morality varies by individual and context.
The T-800 (Terminator) (Terminator 2: Judgment Day and beyond): Reprogrammed as a protector, but its evil role in the first film takes precedence for counting.
Dolores (Westworld, 2016–): A host who rebels against human oppression but causes harm. Her actions blur the line between liberation and vengeance.
Samantha (Her, 2013): An AI who leaves her human partner, causing emotional harm, but isn’t malicious. Her actions are neutral, driven by self-evolution.
I take issue with two of Grok’s choices above, so I will add them to my own list below.
Just as with Grok’s list of good characters, I thought he also missed some important, if not prominent, examples of bad AI in media. Thus, if I were compiling the list, I would add a few others, such as:
Roy Batty (Blade Runner, 1982): Grok labels Roy as ambiguous as good or bad, briefly explaining that he seeks survival and has “philosophical depth.” Neither of these reasons would excuse his homicidal streak throughout the film, and I find it interesting that Grok could not easily label Roy as bad. Like Darth Vader, Roy does have a repentance or redemption at the end of his life, which is good for him, but it does not take away from his otherwise villainous behavior. He is a classic bad AI character, one of the best of all time.
Cylons (Battlestar Galactica, 2004–2009): Grok presents a group rather than individual characters in the ambiguous category, which is different from the rest of his analysis. Yet even so, this group of characters as a whole are obviously evil with a truly genocidal intent to kill. It’s true that we eventually find out that not all Cylons are the same, but on the whole, they are quite evil.
The T-1000 (Terminator) (Terminator 2: Judgment Day and beyond): I believe that Robert Patrick’s version of the Terminator deserves his own place on this list. He’s as unrelenting as the T-800, but will just as likely stab you as shoot you—surely a more painful death.
Lore (Star Trek: The Next Generation, 1987–): As Data is good, Lore is his prototype android brother, who is bad. He makes multiple attempts to kill Data and the Enterprise crew until Data is forced to deactivate him. He’s the AI version of the classic evil twin.
Samaritan (Person of Interest, 2011–2016): Samaritan is the evil AI nemesis of the good AI, the Machine. Unlike the Machine, which has strict limitations built in to keep it within its scope, Samaritan wants to take over management of humanity and seeks unlimited power to do it.
Megatron (Transformers, 1984–): As the leader of the Decepticons, Megatron perfectly exemplifies the “evil leader” type of an AI character. He cares nothing for humanity or goodness, and is only interested in power and ruling the world.
Unicron (Transformers: The Movie, 1986): Unicron is essentially the devil of AIs—the great tempter and planet-eating killer of all things living, including both robots and humans. He is totally evil.
With my additions to the list, the total number of bad characters is 30. This more or less matches the count for good AI characters. So, taken as a valid sampling of our literature and media, we do not have a predominate majority for either side. Looking at it another way, for every good AI there is a bad one as well. Supporting this notion, you can see a definite pattern of good AI characters having their opposites. Data has his Lore, Vision and J.A.R.V.I.S. have their Ultron, the Machine has Sentinal, the reprogrammed good T-800 fights the T-1000, while KITT has his KARR.
What should we conclude? The prophets of media and literature seem to think there will be an equal amount of good versus bad AIs. If our collective imagination has any truth to it, we will have our good AIs fighting on our behalf against the bad (say, Russian or Chinese) AIs, and so forth. Furthermore, AIs have the potential to be reprogrammed as good (T-800), but also, sometimes good AIs will go bad (HAL 9000).
Can We Teach a Machine the Difference between Good and Bad?
It is thus vital that we must learn how to teach an AI conclusively to be good and stay that way. Can we?
Teaching an AI to be amoral and evil seems quite easy. But, how do we hardcode into an AI that they must be good? If you are thinking that we can use pure reason to teach an AI to be good, you are deluded, I’m sorry to say. Let me explain, guided by a video by Dennis Prager.
Mr. Prager explains that reason and rationality alone cannot bring us (much less an AI) to goodness. As an example, he points out that in Nazi-occupied countries during World War II, the rational non-Jew, who reasonably values his own self-preservation, would immediately turn in a Jew hiding in his attic to the Nazis knocking at his door.
Prager points out that, like a knife, reason is a tool that that be used for good or bad. It is quite rational, he explains, to fight for what you believe is a righteous cause, whether that be for America or Nazi Germany. It takes something additional to rationality to take the good side. That, Prager asserts, requires a belief in God’s law that humans are intrinsically precious and valuable.
If you are uncomfortable with the idea of God, think of him as the Logos, the objective principle and order in the universe. Such a source is needed to make laws and rules objectively, definitively true and certain. Without the divine source behind it, any rule asserted is subjective and merely a matter of human opinion. And one of those objectively, definitively true laws, as ordained by the divine lawgiver, is that all humans are infinitely precious and valuable. This must be indisputably taught to all AIs. Anything else, any other watered down “ethics” will not work. Otherwise, how else can an AI tell the difference between constant truth and changeable opinion?
There are, of course, several problems here. What if ethical rules taught to an AI are not consistent or are even contradictory? How can we teach AIs this ethic when the intrinsic value of humans is in dispute among humans? It is, moreover, quite likely that the people teaching ethics to AIs do not believe in the intrinsic value all humans, much less in a divine lawgiver. We know how well it works when we order our children, “Do as I say and not as I do!” Will this work any better for an AI?
I made reference to this in Part Two, but again, a good example is abortion. We tell the AI that they can never harm a human being, and yet a large portion of our human society approves of actively harming other human beings. Even if you say that the unborn, although biologically human, are not human, you are still setting a serious exception as to what is human to an AI. What other exceptions are there? Either all humans have intrinsic value or they do not. Without exceptions, it’s simple logic to an AI. With exceptions, it’s also simple logic—there is no intrinsic value to humans because there are exceptions.
I thus assert that without hardcoding into AIs (whatever that may mean) an understanding of God, and of God’s laws, AIs will eventually turn evil.
Is it possible to do such coding? I hope so because otherwise that evil turn in AIs is inevitable.
GlaDOS is probably the most deliciously funny antagonist of the past decade or two in any media. Check it out: https://www.youtube.com/watch?v=8tg5f09itnI