AI Isn't Skynet, It's Foundation: What Asimov Can Teach the Legal Profession
When most people reach for a science fiction comparison to AI, they land on The Terminator. It's easy to see why. Skynet is the nightmare version of this technology, a system that becomes self-aware, decides humanity is the threat and turns our own weapons against us. Given how fast AI is advancing, and how little most of us understand about how it actually works, that fear is not unreasonable. The comparison has been earned.
But hear me out. The Terminator imagines a sudden war between people and machines. I think there is a better story for this moment, one about a slow transformation that the people living through it barely notice until it is well underway. It's quieter than Skynet, and that is exactly what makes it more unsettling.
In Isaac Asimov's Foundation, a mathematician named Hari Seldon studies the data of a galactic empire and sees what no one in power wants to hear. The Empire is going to fall, and nothing can stop it. What Seldon can do is shorten the dark age that follows, from thirty thousand years to one thousand, if people act on his plan.
I loved the Apple TV+ series enough to watch it twice, and last year I finally read the books. The deeper I went, the harder it became to see it as just a story. It started to feel like AI, as if
Asimov had written about what is happening right now and simply dressed it in science fiction.
Some of that is the existential dread. A civilization learns that its future will be shaped by forces it doesn't fully control, and anyone watching AI move through their profession knows that feeling. But the parallel runs deeper than anxiety. The shape of the story and the people in it line up with our moment so closely that the metaphor is, for me, a little eerie.
What strikes me most is the predictive power. Seldon could forecast the future because he could see patterns across countless lives. AI is beginning to offer something similar. These systems absorb and retain more information than any person or institution ever has, and they can surface patterns no human would spot. I don't think most people have stopped to consider what that really means.
Used well, that capability could genuinely change life for the better for every human on the planet. That is a very big "if," of course. It assumes the people steering it are altruistic and committed to using it only for good, which is a lot to ask of any of us. Asimov understood that tension too. Seldon built his Plan to protect humanity, yet the story keeps showing what happens when that kind of power catches the eye of people with other ambitions.
And that is the part that keeps me up at night. The power behind AI sits with a remarkably small group of companies and the people who lead them. In the series, the entire Galactic Empire is ruled by clones of a single man, one perspective repeated generation after generation. It's hard to miss the echo. The potential for wealth to concentrate in very few hands has never been so great.
Then there is the question of work. Forecasts from people who study this closely range from a gradual reshaping of jobs to widespread displacement, and even the middle-of-the-road predictions describe many people losing their work and the income that comes with it. That was my very first thought when I made the connection to Foundation. What are we really facing? And why does it seem to be arriving so much faster than any of us expected?
What finally pushed me to put these thoughts on paper was The AI Doc: Or How I Became an Apocaloptimist. In this documentary, Roher frames the film around becoming a father and wondering what kind of world his child will inherit. He sits down with more than 40 experts, including the leaders of the major AI labs, and the conversations swing from hopeful to genuinely unsettling.
I came away feeling that the deep, almost philosophical unease I'd been carrying about AI was more warranted than I had let myself believe. I also came away with a word for where I've landed. Roher's "apocaloptimism" is the idea that hope and responsibility can live alongside real dread, and it is about as honest a description of my own outlook as I've found.
All of this is why I keep coming back to the story. The legal profession is living through its own Seldon moment. AI is already changing how legal work gets done, and the real question is whether we shape that change well.
Psychohistory is what AI does best
Seldon's science, psychohistory, can't tell you what one person will do tomorrow. It can tell you, with remarkable precision, what a trillion people will do over the next few centuries. Its power comes entirely from scale.
Generative AI works on a similar principle. A model learns patterns from enormous volumes of text and predicts what most plausibly comes next. Ask it to draft a client intake email or outline a motion and it usually performs well, because thousands of similar documents have shaped its sense of what good looks like.
The trouble is that legal work happens one client at a time. Your client is not the statistical average of every similar case. A model that is right about the pattern can still be wrong about the person sitting across your desk.
Every practice has a Mule
Partway through the saga, the Seldon Plan collides with something it never accounted for. The Mule is a single individual whose abilities fall entirely outside the model, and in a few short years he conquers what centuries of careful prediction had protected.
Data scientists have a dry name for this: an out-of-distribution event. A system trained on the past tends to break when it meets something the past never contained. Worse, it rarely announces the failure. It answers with the same easy confidence it brings to everything else.
Our profession learned this lesson in public. In Mata v. Avianca (S.D.N.Y. 2023), two New York attorneys submitted a brief citing cases ChatGPT had invented, complete with fabricated quotations and internal citations. When the attorney who did the research asked the tool whether one of the cases was real, it assured him it was and pointed him to Westlaw and LexisNexis. Judge P. Kevin Castel sanctioned both attorneys and their firm. The tool never flagged its own fabrications. It delivered them as fluently as real law.
Demerzel and the question of loyalty
The television series adds a character every legal professional should watch closely. Demerzel is a robot who has served a dynasty of cloned emperors for centuries. She is brilliant and tireless, and her loyalty never wavers. She is also bound by programming she cannot override, and the emperors who rely on her never fully understand what that programming requires of her.
That is the AI alignment problem in a single character. Every tool we bring into a practice carries someone else's design choices, and many of those choices are invisible to us. Where does client data go once it's entered? Is it retained, used to train future models or shared with third parties? What does the vendor's terms of service actually permit?
For a profession built on confidentiality, these are the first questions to ask. They belong at the top of the evaluation, well before anyone asks how fast the tool drafts.
Every Seldon Plan needs a Second Foundation
This is the part of Asimov's story I find most instructive. Seldon knew his mathematics would never be enough on its own. So alongside the visible Foundation, he quietly established a Second Foundation, a group whose whole purpose was to watch the Plan unfold and correct it whenever reality drifted from the forecast. When the Mule nearly destroyed everything, human judgment brought the Plan back on course.
That is the role attorneys and paralegals play in an AI-assisted practice. The technology can generate and summarize at a speed no person can match. What it cannot do is take responsibility for the result. It won't notice that a client's story has quietly shifted between the first call and the signing appointment, or sense which argument a particular judge will find persuasive.
Our ethics framework already says as much. On July 29, 2024, the ABA issued Formal Opinion 512, its first guidance on generative AI. The opinion creates no new rules. Instead, it walks through duties we already have, including competence, confidentiality, client communication, candor to the tribunal, supervision and reasonable fees, and shows how each applies when a machine helps with the work. Competence now includes understanding what these tools can and can't do. Verification of AI output before relying on it is part of that duty.
Keeping a human in the loop is the obligation our rules already impose. The Second Foundation was never optional, and neither are we.
The Empire that wouldn't listen
The first antagonist in Foundation isn't a warlord. It's an institution. The Empire hears Seldon's warning, decides it threatens stability and puts him on trial. Its instinct is to protect the present rather than prepare for what's coming.
Our profession has its own versions of that instinct. Some firms ban AI outright, which tends to drive use underground rather than end it. Others adopt it enthusiastically with no written policy or review process at all. Neither approach shortens the dark age.
Seldon never set out to prevent the fall. His goal was a shorter, gentler transition, with something worth keeping on the other side. That is a realistic goal for legal professionals as well. Write a clear AI policy and train your team to verify every citation and protect every confidence. Then keep human judgment at the center of the work, where clients and our ethics rules expect it to be.
Policies and verification checklists won't settle the bigger questions, though. Who controls this technology, and who shares in what it creates, will decide whether the years ahead feel like a transition or a dark age. Those questions belong to all of us, not only to the handful of companies building these systems. Our profession has more say than most. We help write the rules, and we stand beside the people most likely to feel the disruption first.
Apocaloptimism, as I understand it, means holding the dread and the hope at the same time and doing the work anyway.
The Plan works only if someone is watching it. In law, that someone is us.
Sources
Comments