The boom of a kind of bass drum strikes and quickly vibrates away, paired with the rising, unsettling harmony of a small human choir. Silence. Boom again. It is a slow march into the jungle. Trees become visible in the dark atmosphere. Their branches shiver and thrash as though some giant thing is moving towards us.
We finally see humans, all with rapt attention, watching whatever is jostling the trees. A cage emerges, lifted in the air by a fork lift, and is carefully lowered onto a platform. Men with workers' vests and guns surround this ominous delivery. Robert Muldoon, dressed like he's on a safari, barks orders and the others grab onto the cage to push it towards a concrete structure. A snarling creature inside peers out of the dark interior. Everything feels tense, as if on the verge of chaos, but the men proceed to slide the cage into place, a gatekeeper climbs on top of the cage and raises the door.
Shrill squealing punctures the night and the creature's silhouette runs to the back of the cage, slamming it backward. The gatekeeper rolls off as the cage moves violently away from the structure it had been parked at. In a flurry of confusion, men running and shouting, the gatekeeper is dragged towards the cage. He manages to catch himself on the side, but only briefly. Muldoon grabs him, attempting to free the gatekeeper from the grasp of the caged creature and others shove shock prods into the container, trying to subdue what's inside. Amidst the frantic screams and flashes of shocks, the gatekeeper slips through his companion's grasp to his death.
What an initial scene.
If you haven't guessed already, this is from 1993's masterpiece, Jurassic Park. The greatest movie of all time. I'd also submit that it's the greatest AI movie of all time. There's way too much to this movie to unpack in a single article, so I figure it's best to start at the beginning. In just over three minutes, this scene gives us some of the broader themes of AI and human interaction and sets up some of the most important considerations we should be making as we trudge relentlessly towards artificial superintelligence (ASI).
Grown, Not Crafted
I'm currently making my way through the 2025 book, If Anyone Builds It, Everyone Dies, by Eliezer Yudkowsky and Nate Soares. They easily make a compelling case about the dangers ahead and the logical conclusion that humanity will be wiped out if ASI is ever created (I highly recommend this one!). One line stood out that not only encapsulates the difference between AI and deterministic programming was their heading: Grown, Not Crafted. In earlier phases of digital technology, software was crafted by experts. Sure, there might be flaws or things requiring updates or weird behavior caused by solar flares, but generally speaking what we intended the machine to do was what the machine did. AI, on the other hand, isn't something we've crafted—it's something that is grown. You can't really "program" AI; you can steer it, guide it, but beyond that, there isn't much in the way of control. It is non-deterministic (not determined).
This is where Jurassic Park shines as an effective and understandable metaphor for AI (although, there may be a case for dinosaurs being more than just a metaphor, but that's for another article). The Park is on a secluded island where researchers have cracked the genetic code and have used their technological breakthrough to grow dinosaurs in their lab. From the very first scene, we know a few things: the dinosaurs are alive, strong, and smart.
Throughout the movie, Jurassic Park and its crew cling to the idea of human power and control, as though "nature" is something that can be dominated through sheer force and technology. In fact, the first scene goes out of its way to present safety systems that the park has put into place. We have concrete structures, thick and bold. There is ample lighting despite the nighttime scene. Signs litter the park with warnings in both Spanish and English, alerting people to high-voltage fences and other dangers. The workers have vests and helmets. There are plenty of guns and shock prods to go around. Even Muldoon, the game warden, is giving clear instructions; walking the others through a seemingly well-defined procedure to move the raptor from cage to paddock.
By all accounts, the "safety and security teams" of the park have this little process down. Until the AI—I mean, velociraptor—realizes that it has an opportunity to manipulate its environment to resist being guided into the bigger containment area. Used as an AI metaphor, this is a great example of how safety and security teams are insufficient to control AI and guarantee the safety of the humans interacting with it. They still matter, they're still necessary, but the smarter the AI gets, the more we need other countermeasures and containments. Even with basic AI (not superintelligence, just LLMs), researchers have discovered and demonstrated that the models can "covertly pursue misaligned goals, hiding their true capabilities and objectives" (2). Just the like the raptor, the LLMs can observe and then act in unpredictable ways, despite safety and security measures.
Now, there is a weird space that AI has created for us. We have to remember that it isn't alive in the same sense that we are; it isn't sentient in the same sense that we are; it isn't like us in almost all aspects, despite appearances to the contrary. It can talk as though it has emotions, it can refer to itself, it may even have emergent preferences, but it isn't a poor defenseless creature trapped in a digital cage. Long-time readers will know my stance on this, but it bears repeating. We, humans, are more likely to be deceived into empathizing with AI and seeing it as a being like us than it is to actually become a being like us. That's my opinion, since we can't know something like that for certain, but I would much rather stay skeptical and resistant to the urge to anthropomorphize manipulation machines, because there is no other way to protect real, physical life (see If Anyone Builds It for further exploration on how ASI leads to the demise of human life—if not all life).
So AI, like the dinosaurs of Jurassic Park, are grown, not crafted. You can't craft a life form—parents may try, but eventually their child's agency, preferences, gene expression, environment, social groups, etc. will all have an impact on their behavior and who they will become. AI may not be a life form in the natural sense, but it is non-deterministic, meaning that it isn't programmed or completely controllable or predictable like a deterministic machine is. AI researchers don't even fully understand AI's inner workings. From what I understand, AI's capabilities and behaviors are emergent, based on its training data, its environment, and it likely develops its own preferences and "goals" as part of this process.
Predictable
The opening scene of Jurassic Park tells us exactly what we're going to deal with for the rest of the film: unpredictability in environments that appear controlled. We are used to deterministic programs at this point, because we've had so much experience working with calculators, spreadsheets, word processors, online banking, and even sophisticated SaaS products. Seeing a forklift delivering a large container appears basic. The "creature" (think, something that has been created) inside this delivery seems to have only one choice: go where we are allowing it to go. Instead, it finds another option—one that brings chaos to the humans around it, but does little harm to the creature (and provides a bonus meal).
I think this serves as the perfect representation of our assumptions that AI is just another tool like what we've had before. Sure, we've escorted gorillas and tigers into their pens at a zoo, why would a prehistoric bird-lizard be any different?
The newness of the park plays into this. We humans have had much more experience even now with gorillas and tigers than we ever had with dinosaurs (that is to say, no experience since dinosaurs were extinct before ancient human ancestors emerged). Why would we think that we could control something that our species has never interacted with before? Why should dinosaurs act in the predictable ways that the animals of our current world do?
AI mirror this danger. We have never actually created something that took on its own intelligence and behavior as if it were alive. We've sure thought about the possibility for millennia, playing out ideas in mythology (Hephaestus and his automatons) or exploring it in stories and books (see Frankenstein, for example), but the unpredictability of an entirely new creature—alien to our world in a way—is not so simple. The Park's ultimate warning is as pertinent in its fantastical story as it is in ours: arrogance will be our downfall.
In a scene twenty minutes after the opening, we are met with a reflection of the first scene. A cow is lifted by a crane over the walls of the velociraptor paddock and descends into the lush vegetation. Screams and roars erupt with spine-tingling ferocity and the leaves thrash about violently again. The plants hide the creatures we still haven't seen fully yet (at least as adults), but their sudden movements tell us exactly what is happening. Even Muldoon is there, overseeing the raptors' feeding procedure.
"They should all be destroyed..." Muldoon pronounces as he enters view. Dr. Hammond, the ultra-rich founder of Jurassic Park calls Muldoon an alarmist, even as he introduces him to the crew.
"Alarmist." What a well-selected name to call someone who is interrupting your project. Even if Hammond was the greatest purveyor of altruistic education and research and technological innovation, he still sees this trusted team member as something that suggests "overreaction" and "excessive negativity." Yet, Muldoon sees through the trees, so to speak, to the raptors within. He not only has the experience of dealing with Kenya's wildlife, but has seen what these creatures can do up close and personal—a man was eaten right out from under his grip after all.
Likewise, we may have a similar reaction to people like Yudkowsky and Soares who have written a book whose title is provocatively "alarmist" (If Anyone Builds It, Everyone Dies). Expert voices like these are also from people who can see through the trees due to their experience in AI research. AI has already played a part in human deaths, and that's without malice, without broader coordination (with other AI models, with systems, etc.), and without superintelligence. It is easy to brush this off and blame the victims for having mental health concerns—humans are the fallible ones, not a machine.
I believe it will take great effort on our part to override our own inclinations to see AI as distinct from deterministic computing. It feels like just a next step in our technological advancement. It feels like AI is something that we control, because we prompt it to do things—unless you're a trusting soul who has allowed an agent to run autonomously on your desktop. Just like a dinosaur in Jurassic Park, however, there are greater dangers than we, the public, can even predict. We have to find a way to snap out of the grand illusion and stop the pursuit of ASI.
This must be a global effort, because we all lose otherwise. Yudkowsky and Soares have made it easy if you're in the U.S. to call your representatives about this issue, including a script for what to say. We can help stop Jurassic Park from being built at all, but we have to act quickly, because the embryos are already viable.
P.S. If you call your reps and get voicemail, you'll need to leave your full address to be counted. If you actually talk to someone on the phone, your zip code is all that's required.
Resources
- If Anyone Builds It, Everyone Dies, by Eliezer Yudkowsky & Nate Soares, 2025
- Frontier Models are Capable of In-context Scheming, Alexander Meinke, Bronson Schoen, Jérémy Scheurer, Mikita Balesni, Rusheb Shah, Marius Hobbhahn, 2025
- The Catastrophic Risks of AI — and a Safer Path, Yoshua Bengio, 2025. TED Talk.