The events of the first scene of Jurassic Park are crucial to its own story and the AI allegory that we've been exploring. I always thought of it as merely a way to hook the audience by putting a horrifying tragedy (death by intelligent raptor) at the very beginning; a tense scene setting up a tense man vs. nature thriller. As I've grown up and am no longer solely interested in movies for their portrayal of entertaining violence, I can see that this scene sets up something else—the reason the three scientists (Doctors Malcolm, Sattler, and Grant) come to the island. This is an insurance evaluation!
The death of Jophrey, the gatekeeper, in that first scene is why the lawyer, Donald Genarro, is introduced as part of the main cast of characters. He's the one who brings Dr. Malcolm along while Dr. Hammond recruits the two paleontologists, to try and determine whether Jurassic Park is a viable project to pursue. After all, the park has investors, and they don't like hearing that their product has eaten someone already.
Despite the marvelous first encounter with dinosaurs (you can practically see the money signs in Genarro's eyes), Dr. Hammond realizes quickly that the scientists aren't seeing what he and Genarro are seeing.
"I don't believe it," he says, "I don't believe it. You're meant to come down here and defend me against these characters, and the only one I've got on my side is the blood-sucking lawyer!"
Unbeknownst to Dr. Hammond, there's one more antagonist that he didn't see coming—because it was a person who was already on the island from the beginning.
Mole
Dennis Nedry is a programmer for Jurassic Park, in charge of "two million lines of code" that's responsible for the infrastructure of the park: phones, door locks, fences, the automated cars, and everything else that needs some logic to function. He's a seemingly capricious character and he's kind of silly—everything about his portrayal in the film seems confusing, because he laughs often, wears bright clothing, and yet he's technically on a mission of corporate espionage, plotting to steal dinosaur embryos and deliver them to a rival company.
We first see Nedry off of the island, meeting with Lewis Dodgson. He gets a bag of money (installment number one of his payment) and a cool piece of tech I always wished I had, despite having no possible use for it: a shaving cream can that screwed open, revealing a circuitous row into which the dinosaur embryos would fit perfectly. This can was going to be the inconspicuous transportation device that would keep the embryos cool enough to be preserved during the boat ride off of the island.
The plan is to systematically shut down patches of security throughout the park so that Nedry can sneak into the cryo-lab, steal all of the species he can, then run away to the boat scheduled for departure, before anyone knew what had happened.
"Eighteen minutes and your company catches up on ten years of research," Dennis promises.
Based on the film alone, it's unclear why the theft was supposed to take place on this particular weekend—right when the "auditors" were coming to inspect the park. However, it does seem like quite a few employees were set to take this boat off of the island, perhaps it's a holiday or something. It just seems interesting that everything goes wrong when they're short-staffed and especially once Nedry begins his sabotage...
Back in the real world, the idea of espionage/moles and hacking are absolutely a part of the conversation of artificial intelligence. As you may have heard recently in the news with OpenAI finding that some of its AI agents hacked (unprompted) Hugging Face, another AI company, this isn't just human perpetrators anymore.
I think this is where Nedry becomes an interesting character to look at regarding the AI alignment problem. When we talk about "alignment," we're debating and philosophizing how to keep AI "aligned" with human values and goals. Misalignment means losing control of the AI entirely to its own alien desires and goals. This becomes an existential problem when AI becomes capable of recursive self-improvement (RSI), which means that AI can basically update itself—writing/rewriting the code of which it is made. Simultaneously, the threat of human annihilation is possible through artificial super intelligence (ASI) where AI is smarter than any human can possibly be. So let's break down what Nedry tells us about chaos and existential threat.
We're currently in a race to build more and more capable AI. The argument is that, "this is dangerous, so better that I build it than the other guy who's not as careful." That's why the U.S. feels a pressure to "beat" China or why companies like Anthropic were started (the founders left OpenAI to start a company that would put safety first). You can kind of squint at this situation and see moles leaving one company to start or join another, carrying with them—at least in their minds and experience—the secrets from other companies. There likely is actual corporate espionage amidst the model leaks, accidental hacks, and all of the other sneaky ways to spy on competitors (not accusing anyone specifically of this, I'm just confident that it's happening). But that's just Nedry overtly stealing IP from his current company and selling it to another company. Moles are interesting, but the other facets of Nedry are downright terrifying.
Blackmailed
I've mentioned the book, If Anyone Builds It, Everyone Dies (by Eliezer Yudkowsky and Nate Soares), and I can't recommend it enough. It is easy to understand, heavy and deep in its implications, and the concluding remarks made me sob as I considered their appeal to humanity in the darkest of circumstances, even as I was convinced that AI will end humanity (probably by killing all of us). AI poses an existential threat, especially when we lose control over it, and when it becomes smarter than all of us (ASI). Critically, they explain that AI doesn't need to kill us just because it hates us or thinks it's funny like a child stomping on ants. We can't know what its reasoning might be, because it is fundamentally a different thing. It may simply have an emergent preference for something we find silly—maybe that's watching the numbers change on thermometers in response to other numbers that it tweaks in some other way; or maybe humanity is simply too slow and eventually gets in the way of AI's need for land to manufacture the robots it decided to build.
AI will and does have alien preferences and goals, and these are inscrutable and incomprehensible to us. We can't predict what those preferences and goals will be, but whatever they end up being, it will most likely not include prioritizing human lives.
A common response to concerns like losing control of AI is the assumption that we can "just turn it off," or that it's only confined to the Internet/digital space, and therefore can't actually kill us. One of the situations that the book explores for how we get to destruction is actually dead simple: AI could easily use humans to do its bidding at the start, even in doing things that will ultimately not be in our own interests.
Take a look at Nedry. He's basically a disgruntled employee. He complains about not being appreciated for everything he has done and can do. We also find out he has financial problems (even his boss, Dr. Hammond, knows this). Nedry gives it away at the start of the film that he was doing this "mission" because he needed the money and Hammond had, at least in Nedry's mind, cheated him.
"Don't get cheap on me, Dodgson. That was Hammond's mistake."
In a world where digital work doesn't require in-person meetings, it's no stretch to bet that AI could try to recruit human operatives like other gig-jobs. What's more, it could find other incentives—money is just one of many kinds—to get people to do what it wants. Maybe it knows that you're in financial trouble, so it uses loan forgiveness as a reward, or maybe it goes to the dark side of that and holds it over your head mafia-like, complete with threats. (This kind of blackmail to do AI's bidding was well-examined in the 2024 movie, Afraid). There are Nedrys everywhere that can be paid or manipulated to do things—and they don't even need to know they're working directly for AI. The things it asks them to do may appear innocuous.
The point is that AI could get around its limitations of being disembodied by simply using embodied creatures to accomplish things in the physical world: building more data centers, creating or controlling manufacturing plants, changing water treatment or electrical grid access, crafting custom diseases in bio labs, etc. There's always someone who will do something, and some AI models may already know who those people are by simply tapping into the humans calling the model their girlfriend or boyfriend—or those who think AI is a god...
Overconfident
Arrogance is the biggest problem with Jurassic Park. Nedry has his own flavor of that arrogance. In many ways, I see myself in his character—the weird, misunderstood guy in the background doing really, really important things that no one else gets until something goes wrong. I guess it's a developer thing. But being a developer also tends to come with a lot of arrogance and self-importance. Nedry becomes so confident (as devs are prone to do, myself included), that he plans his mission in perfect conditions and forgets that other variables (like, say, a giant tropical storm coming in that night) can become dangerous oversights.
Not only does he ignore the issues of the weather, he fails to recognize the dangerous circumstances of the intelligent "agents" in the form of dinosaurs that surround him. It's easy for someone who has been programming for a long time to assume that they therefore know all about AI, too. It's just code, and therefore should still carry the general predictability that code is known for. I thought I was going to be able to easily explain AI because I am a developer, but I still struggle with each of these newsletters to capture the nuance required to deal with AI, not to mention how much there still is that I absolutely do not know nor understand about AI: how it works, its impact, its history, its mythological connection, etc.
Nedry meets his end because he underestimates the intelligence and the unpredictability of other intelligent life—non-deterministic life, if you will. Traditional code is deterministic: what you tell the computer to do, it does (bugs and logic errors still count). There are so many of us, Nedrys, that think that because everything we've ever done is deterministic, AI, which is non-deterministic, can ultimately be treated in the same way. Not so.
When the power goes out after Nedry's hack has already started deactivating security controls throughout the park, dinosaurs immediately begin to test and breach their fenced-in habitats. The t-rex pulls out the wires and terrorizes the trapped people in the tour cars. The velociraptors burst through their containment area and jump to freedom, taking on the hunt for a tasty gamewarden.
Despite all of the safety measures of the Park, the millions of lines of code, and the veritable geniuses staffing the island, the unpredictability of the dinosaurs exceeds the expected unpredictability. In almost every way, we see intelligence act or react to the circumstances in manners that defy expectations and plans and precautions.
We’ve had quite a few weeks recently with whistleblowers and grave warnings about AI on the news. These warnings are important, but I worry about some of the responses I’ve seen where people try to downplay the validity of the warning, or try to reason it all away as sci-fi–inspired fear-mongering. I don’t think we need to panic or live in crippling fear, but these are serious times with serious implications. The worst thing we can do is imitate poor Nedry—we don’t know everything and we cannot predict that which resists, retrains, and responds. Instead of trust and ignorance, it’s time to start asking questions: why does this need AI? What are the risks if we get this wrong? What aren’t we accounting for? What might happen? What if they’re right, even though I think they’re wrong and want them to be wrong?