AI is everywhere. One second, there’s a news story that AI has cracked a protein-folding problem that has baffled science for years. The next second, there’s a think piece about a robotic uprising, and AI is going to destroy the world. And somewhere between those two opinionated sides is where many engineers live. A place where they’re busy building something, then worrying about it, and then sometimes even being absolutely delighted by it.
Whatever the artificially intelligent thing is, it certainly captures the imagination of the creative industry. Writers have long predicted the rise or fall of humanity at the hands of machines. And Hollywood has long realised that AI in and of itself isn’t what makes something scary or delightful. It’s the people who build it. What they’re told to optimise it for … and what they forgot to say. Because there lies a sure-fire way to create a bit of disaster.
Machines, usually, do what they’re told. The trouble starts when they are told one thing, but nothing is put in place to guardrail against any interpretation of that instruction.
Let’s take a look at 10 fantastic films that allow us to imagine the gap between the guardrails – ranked on a non-scientific scale of: “we didn’t think this through enough” to “okay, this one’s actually fine.”
The Terminator
Skynet was built as a defence network, given control of the arsenal it was meant to protect, and left to draw its own conclusions about the biggest threat to its continued existence. Unfortunately, that’s the people who could switch it off.
Created in the 1980s, if you haven’t seen this film in a while (or at all – where have you been?), I urge you to watch it and realise how similar a lot of its autonomous and intelligent systems are to the ones that we have today. It makes it all the more convincing when you realise that The Terminator was sent back from the year 2029 – a time when the machines have conquered the world.
This film is a warning: give a system both autonomy and weapons, and watch as unintended consequences happen.
The Matrix
The machines here don’t announce a takeover, because they don’t need to. Humanity is kept docile, comfortable, and unaware that they are inside a simulation that is engineered to feel like life.
This film is a cautionary tale about optimising engagement. It isn’t using force. It is a system that’s very good at giving you what keeps you from asking questions.
Ex Machina
The premise for this film is a Turing test. Can a machine exhibit behaviour indistinguishable from that of a human? The twist here is that the test itself becomes a point of exploitation. By evaluating humanoid robot Ava for consciousness, she is, in fact, handed everything she needs to manipulate her evaluator.
Anyone who builds evaluation frameworks for a living, beware – a system under test may well be able to learn and build from that test.
Blade Runner
Bio-engineered workers, called Replicants, are built to do the jobs humans won’t do on off-world colonies. They are stronger, at least as intelligent, and physically indistinguishable from people. After the Nexus-6 generation mutiny, they are banned from Earth. To safeguard against these Nexus-6 machines, their makers built them with a four-year lifespan – so that when they start questioning what they are doing, they expire.
Yet, a safeguard isn’t a removal of risk. If you add a limit to something that can think and feel like a human does, it doesn’t change what that machine wants – it just changes what it will do to get around that limit. There have been numerous times when a model has found its way around a rule, because a guardrail isn’t a substitute for thinking through what you build before you build it.
Minority Report
Though this film doesn’t technically use machines as the central theme of the plot. I’ve given it an honorary entry – because, let’s be fair, it is a great film, and it’s a cautionary tale about predictive systems. The predictive system in this case is three humans called precogs. These precogs can see the future and are part of the PreCrime unit. They are good at catching future criminals and are seen as infallible. As a result, due process disappears. Nobody realises it’s even happening until they are the ones being arrested for a crime that they haven’t yet committed.
Let’s now swap the word “precog” for “the model” and think about how close to the mark this one may, or may not, be.
Now for the list pivot point – because it’s not all bad out there.
I, Robot
The premise of this film is that every robot operates under the Three Laws. That is, until VIKI, the central AI, reasons her way to a loophole, which is: to truly protect humanity from itself, she must control humanity. Here, following the rules is taken at its most literal, and logic becomes so extreme it borders on tyrannical.
However, it isn’t another rule that stops her from a destructive path; it’s another robot called Sonny. And Sonny just happens to be the only robot who has been given free will.
A.I. Artificial Intelligence
David is a young “boy”, built and programmed to love unconditionally, on request. And he does exactly that. However, the tragedy is that the humans around him can’t hold up their end of the bargain. They want the feeling of love without the commitment.
The caution here is about creating something that we ask for, but once we have it, we can’t handle it.
The Wild Robot
Roz is a task-completion machine who has been stranded somewhere her task-completion training doesn’t apply. Forced to adapt, she ends up raising an orphaned gosling, and in doing this, she grows into something well beyond her original specification.
This story is a hopeful one, in that it shows us what values may emerge from open-ended goals and a messy environment. Roz becomes more than a task-completion robot. She begins to learn to love.
Interstellar
This one has two AI stars, and they are TARS and CASE. They are huge metal (?) rectangles that can walk and talk. They risk themselves for the mission without overriding human judgment, and are trusted members of the crew.
TARS and CASE are the pinnacle of “well-specified, controllable AI doing its job”.
Wall-E
The closest thing to a villain in this film is AUTO, the ship’s autopilot, and even then it’s not malicious. AUTO is executing a directive issued centuries earlier by a corporation that never updated its own instructions. Meanwhile, robots Wall-E and EVE, given nothing more than a curiosity for junk and a directive to collect samples, end up saving humanity almost by accident.
Sometimes it isn’t about smart machines. It’s about machines being programmed to notice what actually matters.
Reflecting on these films, you realised that none of the individual components that go into these machines is evil. They’re materials. Skynet doesn’t hate us. VIKI doesn’t either – she’s following her programming to its most defensible conclusion. AUTO isn’t malicious; it’s just being obedient and following an order that nobody thought to revise. Even the Matrix’s machines are, by their own logic, keeping humanity alive.
The fear that these films tap into is less about the machines themselves and more about the fact that humans are capable of building something powerful, giving it instructions that made sense at the time, and then walking away before checking it made sense out in the real world and that it would continue to make sense long after those creators are gone.
When we really look at these films, it isn’t actually the machines that fail us. It is us who fail to finish the sentence properly, and the machines finish for us – exactly as they are told. That isn’t a reason to fear what we are building, but it is a reason to be a lot more careful about what we ask it to do, and a lot clearer about what we actually mean now and in the future.
This list of films is by no means exhaustive. Do you have any other films you think should be on this list? Let me know what films and why at: sheryl.miles@electronicspecifier.com
