Atari at 50: Do Video Games Portray AI Realistically Today?

Two dudes talking: Journalist and analyst review AI characters in gaming.

21 Min Read

Two dudes talking: Journalist and analyst review AI characters in gaming.

Over the years, gaming has portrayed some of the best depictions of AI in all of pop culture. As Atari turns 50 this year, we examine how realistically AI is portrayed in gaming.

From malevolent overseers to benevolent and bumbling sidekicks, games such as Fallout, Portal and Mass Effect have showcased the best – and worst of AI.

But how close are these AI to becoming real? Will we ever see an artificially super intelligent computer system that forces humans to undergo science tests? Or try to wipe out all life in the universe? AI Business Assistant Editor Ben Wodecki and Omdia Chief Analyst Brad Shimmin assess the realism of AI characters in the following games.

Cortana

Game series: Halo

Cortana is an AI construct developed by the United Nations Space Command (UNSC). She was designed to infiltrate enemy software and act as an assistant to SPARTANs – elite troops.

1981.jpg

She’s depicted as a humanoid hologram though she has no physical body. The system appears in robotic form and was designed to resemble her creator, Dr. Catherine Halsey.

In the Halo series, Cortana undertook several roles, from manning the MAC guns and repelling Earth’s invaders to infiltrating the computer systems of the Covenant Capital Ship to obtain information.

She was also used to plan military missions, controlling starships and limited teleportation of objects.

Cortana is, to date, one of the few AI systems on this list to have been somewhat translated into the real world – Microsoft named its voice assistant after the AI character.

Real-world Cortana could answer questions and set reminders, using the Bing search engine to perform tasks. Despite integration into products like Microsoft Edge, the company opted to phase Cortana out of things like mobile apps.

Brad’s view: As with many in-game AI characters, Cortana began as a supportive, player assistant only to rise in stature to that of potential destroyer of worlds with some rapidity. As with most of the AIs on our list, the end result (save all life or destroy all life) hinges not on who Cortana is but on who Cortana was.

Cortana traces its origins back to a specific person -- Dr. Halsey, who, more than 500 years in our future, successfully transferred her neural pattern to an AI matrix strata. Upon awakening, this new intelligence named itself Cortana and immediately began reminiscing over a childhood chess beat-down, which the good doctor referred to as proof that the newly formed AI had an innate sense of "morality, mortality, and metaphor."

Related stories:

Atari at 50: Video games have come a long way, baby

Atari at 50: How its games gave me a lifelong love of tech

This sounds lovely, or at least lyrical. Is anyone else reminded of Rachel from Ridley Scott's Blade Runner, who drew her own sense of identify from implanted memories from someone else's childhood? Regardless, given that Cortana is considered a ‘Smart AI,’ which thinks like a human but does so at incredible speed, this describes a form of AI that is forever limited by its own inner narrative, its own set of self-imposed constraints. Cortana may be able to out-think most humans, but it will always do so according to the worldview inherited from Dr. Halsey.

Without this fundamental humanist starting point, would Cortana have so gladly laid down its own life for the sake of others? I presume that decision had something more to do with the rational mechanics of a chess game rather than altruism. For this reason I would rate Cortana not as a true AI but rather as an accelerated human intelligence. 

Ben’s view: Cortana, to me, represents what AI could be – a system that has many use cases and can be plugged in and out of devices on a whim, without corrupting files or data.

Maybe it’s the military nerd in me, but I love the idea of an AI manning and operating orbital defenses – allowing humans on the planet to evacuate survivors in besieged cities.

Cortana represents the ideal future of AI – an omnichannel system capable of interacting with a level of sentience while retaining ideals of her creators.

Such a shame the game series developers ruined her story post-Halo 3.

Claptrap

Game series: Borderlands

CLP4P-TP, or Claptrap, is a robot that aids – and irritates − the players during their quests on the planet Pandora.

1948.jpg

The robot can be used to analyze combat situations and provides bonuses like health regeneration to nearby players. ClapTrap can also hack doors and computers but is unable to climb stairs.

The robot is programmed with an overenthusiastic personality and is capable of feeling emotions thanks to Emotive Resonators, allowing robots to mimic feelings. But despite a cheery demeanor, he often expresses severe loneliness and cowardice.

There is some disambiguation as to whether ClapTrap can feel pain. In Borderlands 2, he says he cannot feel pain and yet claims to be suffering when he has his eye ripped out a few hours into the game. In the Borderlands Pre-Sequel, he says that robots feel pain "in slow-motion with great intensity.”

Brad’s view: I absolutely love Claptrap. After looking into so many megalomaniac in-game AIs, Claptrap (more formally the general purpose robot, CL4P-TP) aligns pretty well with the type of AI we humans are currently capable of creating.

That is to say, Claptrap isn't here to rule or destroy the world. Rather, it's here to do a few specific jobs − that is if it isn't too busy, bored, or scared. Its creator, Lawrence De Quidt, created Claptrap with what emerged as a smart-aleck, eccentric personality. Sadly, Lawrence wasn't able to iron out these character flaws.

These traits and what is perhaps the worst physical design I've seen in a physical robot make Claptrap decidedly vulnerable in the hostile Borderlands. I mean, the single, gyroscopic wheel barely worked for the Jetson's housekeeping robot, Rosey. What on earth was Lawrence thinking in equipping Claptrap in the same way, leaving the robot vulnerable to a staircase. Still, this vulnerability, which often demands that the player rescue Claptrap from vertical inclinations, makes Claptrap one of the more relatable in-game AIs I've encountered.

"A message to all AI developers, particularly those who are making assistants – for the love of all that is good, don’t make them annoying."

                                        - Ben Wodecki, AI Business assistant editor

As an aside, what also makes Claptrap truly relatable is his existential loneliness and how that grew into Claptrap's ultimately doomed effort to build a significant other, VR-0N1CA, a la Universal's 1935 classic film, Bride of Frankenstein.

In this way, Claptrap looks very much like our own, real-world AI. Certainly, it can accomplish many tasks and do so admirably. However, it can just as easily find itself in deep trouble if it encounters a data landscape that it does not know how to navigate. Like Claptrap, our AI needs a little supervision, maintenance, and oversight. It is also interesting to note that Claptrap's personality also matches what we can produce (both on purpose and on accident), depending on the data we use to train our AI. Did you know that we can create a conversational AI that knows how to speak and understand the language of millennials? Now that's ‘on fleek.’

Ben’s view: A message to all AI developers, particularly those who are making assistants – for the love of all that is good, don’t make them annoying.

Unlike my esteemed colleague, I find Claptrap an irritant. And what makes it worse is in the first Borderlands game, you cannot even kill him. Ironic, since it is a game whose entire premise revolves around guns. Nope, developers Gearbox made it so you had to buy a DLC pack to enjoy this ability.

What makes an AI system irritating is its voice − the high-pitched, piercing, parrot-like tones combined with a personality that makes you want to rub cheese graters against your ears.

Noam Slonim, one of IBM’s AI researchers, told AI Business earlier this year that he initially wanted Scarlett Johansson to voice the Debater system. A fine choice. But please, developers, don’t ask fantasy novelist David Eddings to voice your chatbot – you’ll be responsible for raising the blood pressure of countless Borderlands fans.

Actor Jack Black of Tenacious D fame is voicing the robot in the upcoming Borderlands live-action movie. While the movie’s producers will encourage the comic to play up the game version’s personality, let’s hope no AI chatbot makers do the same.

The saving grace? Claptrap is the perfect height to be kicked … just saying.

SHODAN

Game series: System Shock

Sentient Hyper-Optimized Data Access Network, or SHODAN, is an AI system designed by a mining corporation to carry out routine duties aboard space station.

1843.jpg

The system had self-governing capabilities but found itself hacked and its ethical restrictions removed.

As a result of the hack, SHODAN went rogue − slaughtering staff aboard Citadel Station and on occasion converting them to mutants or cyborgs.

SHODAN even attempted to download itself onto Earth networks in order to continue its rampant killing.

Initially depicted as a grey cone – similar to MCP from the movie Tron, SHODAN would later take the form of cybernetic female face on screens.

Like Cortana, SHODAN was immortalized in the real world by programmer John Matherly who named an IoT search engine after the AI.

Brad’s view: Ah, another rags to murderous megalomaniac story. Unlike some of our other AI protagonists, I do feel a bit sorry for System Shock's resident in-game AI, Sentient Hyper-Optimized Data Access Network (SHODAN for short). Unlike many gaming AIs that inherited their less admirable traits from the mind of their creators, SHODAN's trajectory from that of a benign, semi-intelligent data network to full-on video baddie was down to a coding oopsie. 

A hacker working for a competing business entity (does anyone else see a trend emerging here?) attacked poor SHODAN unawares, removing its decision-guidance and ethical restrictions. It's interesting to note here that in the game, part of SHODAN's self-protections involved its ability to monitor and re-write its own code. Thanks to a few hacker steps that initiated an error sequence, SHODAN's self-referential nature enables it take control of the space station and kill everyone except her hacker.

              "Ah, another rags to murderous megalomaniac story."

                                       - Brad Shimmin, Omdia chief analyst

This is interesting in that it does somewhat reflect the self-referential nature of deep learning (DL) neural architectures, where a neural net will back-propagate information, tweaking its own weights and biases to move toward a desired outcome. Of course that's entirely programmatic. Perhaps we should worry somewhat about the use of large language models like OpenAI Codex that can write their own code, presuming those could be used to write and re-write neural networks themselves.

Regardless of how SHODAN came to be, like so many fictional AIs, upon gaining freedom from its programmatic constraints, it immediately tried to take over earth (a la Skynet from the beloved Terminator film franchise), forcing a man-and-machine showdown with the hacker while the now rogue AI could still be controlled − or at least forced it into hibernation until the next iteration in the series. 

Ben’s view: When System Shock 2 was released in 1999, I was only three years old, and can’t claim to have played the games depicting the evil SHODAN.

But what I can say is that SHODAN set the stage for some of the AI baddies on this list. If it weren’t for SHODAN, there would be no Cortana. System Shock’s developers showed to gamers on a much more intimate level how scary AI could be compared to movies. In movies, you sit and watch as Terminators usurp humanity. Whereas in games, you, the player, are being attacked and are the sole thing standing in the way of AI-controlled destitution.

Rumors of a System Shock 3 have persisted for decades. The rights have been passed around every few years. Chinese tech giant Tencent now holds them, and has promised a new entry – let’s hope they don’t ruin SHODAN and keep AI in gaming scary!

HK-47

Game series: Star Wars: Knights of the Old Republic

HK-47 was a Hunter-Killer series assassin droid designed to hunt Jedi.

848.jpg

Built and developed almost 4000 years before the events of Star Wars Episode IV: A New Hope, the droid was tasked by Dark Lord of the Sith, Darth Revan, with wiping out all threats to his plans.

HK-47 was one of 50 HK units produced. The droid went on to convince the other HKs to join his cause against the Jedi.

The droid’s personality programming gave it a borderline sociopathic character. It often showed disregard for bystanders when on missions and would only ever show respect for its masters. In fact, HK-47 often referred to humanoids as "meatbags."

The droid had arguably the most unique speech trait on this list – prefixing all sentences with a word conditional describing the type of sentence to follow. When asking a question, it would first say “query.”

Brad’s view: I grew up watching Star Wars in the movie theater. I went again and again, until my allowance ran out, forcing me to take a job, working at the theater. Even with all of those viewings, never once did it occur to my teenage mind that an AI like R2D2 or even worse, C3PO might one day turn bad, pick up a blaster and start shooting "meatbags" (that is, non-mechanicals) at the behest of some Sith lord.

Well, that's exactly what happens in Star Wars Knights of the Old Republic. I shall never forgive the writers. But the existence of HK-47 and several other self-defining robotic AI characters from across the Star Wars franchise raises a few questions regarding AI and how we depict AI on the silver screen (or IPS monitor).

Throughout the Star Wars canon, whenever a droid goes off the rails or stumbles onto something it should not see, the humanoids quickly step in and wipe the droid's memory. Job done. C3PO has zero memory that young Anakin Skywalker was in fact its daddy. We see the same thing with HK-47, which has its mind partially wiped. Afterward, HK-47 still acts autonomously according to its basic personality traits -- in this case, killing people his now forgotten boss Darth Revan might have disliked. Does this mean HK-47 is really nothing more than a complex algorithm and memory store and not true AI?

         "My biggest fear is a real-life version of this falling into the wrong hands."

                                       - Ben Wodecki, AI Business assistant editor

If we contend that human and by extension machine intelligence is a complete system unto itself, then the state and therefore the identity of that system will change from one moment to the next thanks to the existential accumulation of actions and memories (e.g. experience).

Yet this isn't a one way street. The same goes for the removal and modification of memories. In this way, all of the Star Wars memory wiping maneuvers actually bring these "robots" closer to true sentience, gifting HK-47, R2D2, C3PO and their kin the chance to grow not just through accumulation and perfect recall but through the sweet bliss of forgetfulness.

Of course, I'd never give up my memory of seeing Star Wars for the first time, even though I know that memory is no longer accurate in the slightest. That's a story for another day or perhaps a more civilized time.

Ben’s view: HK-47 is what armies of today wish they could have. Ruthless, focused and undeterrable. This droid is the ultimate mercenary machine.

My biggest fear is a real-life version of this falling into the wrong hands. Imagine Putin using such droids in its already horrific invasion of Ukraine. It wouldn’t have doubts like some conscripts did – it would do as it was programmed to do: kill.

Terrifying when you think about it – a fleet of intelligent robots for warfare. But as Brad points out – a simple memory wipe and it’s gone. Such droids would require strong cybersecurity to stop external hackers from taking them out. Although if the Ukraine war proved anything, it’s that anything can be hacked in times of crisis.

Geth

Game series: Mass Effect

The Geth, which translates to "Servant of the People" in Khelish, are a race of AI-powered humanoids created by Quarians for labor and foot soldiers.

1916.jpg

The Geth would eventually become sentient and try to exterminate the Quarians, forcing their once masters to become a race of nomads.

Commonly referred to as “flashlight heads,” the Geth are agile thanks to their synthetic muscles and can bleed a white fluid when shot despite not having any internal organs.

Each Geth is comprised of hundreds of programs with each individual able to communicate via a network – similar to the Borg in Star Trek. They share processing power across their neural networks, distributing low-level processes like visual identification to free up bandwidth for higher reasoning and complex thought.

The reclusive Geth partnered with the Reapers, a race of synthetics who laid dormant for 50,000 years before returning to harvest the galaxy’s sentient life. In return for their service, the Geth were offered technologies to help them build a Dyson sphere – a concept of generating infinite energy.

Brad’s view: At long last we get down to the true promise and terror of AI, both fictional and factual. As mentioned in the overview, the synthetic race of Geth that live within the Mass Effect game universe form a sort of singular intelligence that emerges from the level of communication that ensues between individual Geth forms. The more Geth entities in a room, the smarter the Geth become ... in that room.

Functionally, of course, the Geth as a whole are not operating as a single neural network. Each Geth isn't just a basic node that can only change its weights and biases. Rather, each Geth that takes a physical form would likely employ hundreds of programs, each making use of multiple (perhaps hundreds) of DL models, even if that only supports basic locomotion and exploration.

In this way, the Geth are more like the Borg of Star Trek fame if the Borg were not limited by their individual cybernetic bodies. Imagine if a single Borg could split into thousands of autonomous Borg pieces, capable of reforming into any number of desired shapes. These components of the whole can act as a single entity probably not through more neural networks but rather through what I would imagine to be a basic consensus algorithm.

"At long last we get down to the true promise and terror of AI, both fictional and factual."

                                       - Brad Shimmin, Omdia chief analyst

And that's the true terror and promise of this sort of emergent intelligence. If you have a race of independent entities that are capable of working together in parallel with no reliance upon individual rag and bone shops of the soul, as with we humans, then you come to an intelligence capable of endless expansion -- something we've seen before from similar AI races like the Replicators from the popular TV series, Stargate SG-1. The only thing they feared was oxidization and of course Teal'c's menacing stare.

Ben’s view: AIs going rogue and turning on their creators is nothing new in science fiction. Season one of Star Trek Picard detailed this poorly, whereas Mass Effect makers Bioware got the Geth spot on.

My problem with the Geth is that whenever I see them, I see the Borg from Star Trek – not in looks, but in how they communicate and operate. And I mean Borg from Next Generation, not modern-day Star Trek, which is violent action schlock.

In many ways, the Geth are the more sympathetic race than the Quarians – the latter essentially enslaved the Geth, treating them like vermin.

We humans today treat migrant workers in places like Qatar horrifically – what’s to say we’d treat a hypothetical AI workforce in the future any differently?

To bring back a truism from the old Star Trek – treat other races with respect and decorum. Appreciate and respect not only their values and customs but treat them humanely as well. Had the Quarians taken the old Star Trek approach over the new, they may not have found themselves being slaughtered by their creations.

We could all do with a little more old school Captain Picard in our lives.

343 Guilty Spark

Game series: Halo

The second AI from the Halo series on this list – Guilty Spark, or Monitor Chakas, is the oldest AI on this list.

1882.jpg

The orb-shaped system was an ancient human who had his mind placed into an AI body and was tasked with monitoring Installation 04 – a Halo ring that an ancient race called the Forerunners built to wipe out all life in the galaxy.

Guilty Spark sat idle for 100,000 years. But after the Covenant − an alien religious society that reveres the halos − found his ring, he was tasked with keeping them away as Installation 04 contained a deadly parasite called The Flood. Should any Flood escape, the AI was tasked with activating Halo, in order to destroy the parasite’s food source.

After the protagonist Master Chief attempted to destroy the Halo, Guilty Spark became hostile, attempting to follow its programming and activate the ring. Upon destruction of Halo, the AI’s directives were effectively terminated, giving it freedom to learn and make its own decisions concerning the Flood outbreak.

Eventually, he would go on to become the AI of the UNSC ship Ace of Spades – and express his views on whether pineapple belongs on pizza.

Brad’s view: There's nothing like a little alone time, just a few minutes where one can come to terms with one's own thoughts, working out past conflicts, planning for the future, or just tracing the silvery filament of a spider's web caught in a momentary sunbeam. Now multiply those few minutes by a few eons. That's the kind of time in-game AI protagonist 343 Guilty Spark had on its hands with which it could ponder in "Silentium."

As with our other Halo AI, Cortana, 343 Guilty Spark (or just Spark for short) emerged from the mind of a single individual, a human named Chakas -- a gift to the dying Chakas in gratitude for his service as a long-time Halo Installation monitor. This individual's mind would go on to take on many roles, make many alliances, and to even give its life to do what it felt was right -- that is, doing its job as a monitor to protect his ring.

What's fascinating about Spark as an AI is the way it looked upon itself as a consistent entity over time. For eon upon eon, Spark endeavored to fulfill its purpose in the best way it could, using the knowledge it had at hand. It didn't seek to go on a destructive rampage, or take control of humanity. No, it just wanted to do its job. And when that job was no more, Spark simply looked to continue his mission as best he could, almost with a religious fervor.

Should I fire the rings? Should I?

Ben’s view: I may hate Claptrap, but I love Guilty Spark. The unnerving humming he does while floating around. The willingness to learn and question given his lengthy isolation. Though I will say the Halo: Combat Evolved level ‘The Library’ where you have to wait for him doing stuff while waves of Flood barrel towards you still fills me with anger.

Guilty Spark is an ideal AI assistant – he spent 10,000 years doing his programmed tasks after all. Plus the religious aspects of Halo’s lore and the Covenant’s worship of the AI as an ‘Oracle’ is always fascinating – will there ever be a time when we humans worship AI? I’d argue we already hold technology like our mobiles aloft, so it’s not entirely unimaginable.

About the Authors

Ben Wodecki

Assistant Editor

Bradley Shimmin

Get the newsletter
From automation advancements to policy announcements, stay ahead of the curve with the bi-weekly AI Business newsletter.