The Revolution Beneath the Surface
The AI revolution may not arrive as one spectacular breakthrough, but as billions of small judgments and actions beneath the familiar surface of everyday life.
In 2011, the American venture capital firm Founders Fund summed up its criticism of Silicon Valley in a single sentence:
We wanted flying cars, instead we got 140 characters.
The line became closely associated with Peter Thiel. It worked because you could picture the whole disappointment. The future was supposed to be hanging above us in the shape of flying cars, but instead we were standing there with Twitter, writing short messages about our lunch.
When I first saw it, I thought it was completely mistaken.
It judged the internet by what could be seen on the screen. Beneath those 140 characters was a global network that allowed people and machines to connect almost regardless of distance. It connected businesses with customers, software with other systems, and people with knowledge, markets and one another. Much of our modern society and economy has since been built on top of that connection.
A flying car would have been hard to miss. It would have changed the sound of the city and the appearance of the sky. The internet changed the world more quietly. The houses stayed where they were. People still went to work, shopped for groceries and waited for the bus. But underneath it all, the world’s connections were being redrawn. Founders Fund saw the interface and missed the infrastructure.
I find myself thinking about that sentence again because we risk making the same mistake with artificial intelligence.
We Are Looking at the Top
Much of the conversation about AI is about the top. How intelligent is the best model? Can it solve a mathematical problem that has resisted researchers for decades? When will AGI arrive? When will the robot walk among us?
These are not trivial questions. I have written enthusiastically myself about how AI can find connections we overlook, and about machines that can search spaces of possibility no human has time to explore. The large models move the boundary of what is possible at all.
But very few of us carry one of mathematics’ great open problems in our pocket. We have a message that needs sorting. A customer who needs the right answer. An appointment that needs booking. A document containing three pieces of information that must end up in the right place.
It makes me wonder whether the AI revolution will resemble the internet more than the flying car. The breakthroughs happen at the top, while the upheaval spreads from below through millions of small uses that each look insignificant.
Three Small Decisions
Jev from TypeSafe is a good example. Jev is not built to write novels, hold long conversations or solve everything. The model does three bounded things: It can choose among options, score something on a scale, and answer a yes-or-no question with a probability.
Is this inquiry sales, support or spam? How urgent is the case? Does the text contain sensitive personal information?
These are small judgments. But they can be made quickly and very cheaply, which means they can become an ordinary part of software. The program does not need to hand the entire task to a large agent and hope it follows a complicated instruction. The code still owns the workflow. It simply calls on a small amount of judgment where traditional rules fall short.
TypeSafe describes Jev as a programming primitive. That word matters. A primitive is not a finished product that people will necessarily notice. It is a small building block that can appear in a thousand places: an inbox, an approval process, a customer system, or the logic that decides what should happen next.
Jev is interesting precisely because the model does not promise to be everything. It does a few things well enough, fast enough and cheaply enough for developers to use them again and again. That is how something as modest as a yes, a no or a ranking can become part of a much larger change.
The Boring Tasks
Meta approaches the same space of possibility from another angle with Muse. Muse is a personal agent that works in its own virtual computer with a browser, files and apps. Meta highlights tasks such as booking appointments, filling in forms and dealing with customer service. The agent can keep working, ask for approval before critical actions and leave a record of what it has done.
This is not the version of the future that looks best on a stage. Nobody has dreamed of a machine that waits on hold for us. But the phone queue exists. So do the form, the calendar puzzle, the price comparison and the small follow-up we have been putting off since Tuesday.
In When Stories Grew Hands, I described the movement from AI as an answering machine to AI as an actor. Muse makes that movement strikingly ordinary. The agent does not enter society to take over a nation or invent a new physics. It tries to reschedule a dental appointment.
It is easy to smile at that, just as it was easy to smile at 140 characters. But if the agent succeeds, something has shifted between layers. A human intention has been translated into action without the human having to carry out every step.
It also connects to When the Interface Became Optional. When the technology understands the intention and can work through the systems itself, the buttons, tabs and forms become less important to us. On the surface, we may simply have written: “Find me a new time on Thursday afternoon.” Beneath the surface, a machine has navigated between calendar, clinic, login and confirmation.
Compound Interest on Decisions
One rescheduled dental appointment does not change society. Neither does one sorted email or one correctly classified customer case.
But small decisions rarely stand alone. The first judgment determines what the system presents next. The next choice is made in that new situation and changes the starting point for what follows. When this happens millions or billions of times, something like compound interest on decisions emerges.
A company replies a little faster, spots a few more unhappy customers and lets employees spend a little less time sorting. A private individual gets three things done that would otherwise have been postponed. A piece of software begins to understand cases that previously had to be sent to a human. Each improvement is too small to look like a revolution, but it becomes part of the conditions for the next one.
Ten years later, the difference can be enormous without there ever having been a single day when the world was turned upside down.
The early pages of the internet were slow and often rather banal. What mattered was not each individual page, but that the connection spread and became a layer on which almost everything else could be built. AI may be becoming a similar layer. Not only a network that moves information, but a layer of cheap judgment and agency that sorts, interprets and responds beneath more and more parts of everyday life.
The Top and the Bottom
This does not mean that the large models are a distraction. Jev and Muse do not emerge independently of the advances at the top. They are built on the same development that makes models better at language, reasoning and tool use.
But there is a difference between creating a capability and making it transformative for society. A breakthrough can happen in a laboratory. A revolution requires the capability to become cheap, reliable and common enough to enter people’s actual lives.
Electricity did not change the world only through the largest power station. It did so through the wall socket, the small motor and all the appliances that gradually became so ordinary that we stopped thinking of them as electrical technologies.
In the same way, the most consequential AI systems may turn out to be the ones we no longer call AI. A judgment in an insurance system. An agent that moves an appointment. A small model that notices a case should go to a human. The capability came from the top, but its effects emerge as it sinks through the whole of society.
That applies to the errors as well. A spectacular failure in a large model is noticed and shared. A small bias in billions of rankings can be harder to see. Compound interest works on the bad decision as well as the good one. The more invisible the intelligence layer becomes, the more important it is to know which judgments we have placed inside it.
That does not make the quiet revolution less promising. It simply means that its significance cannot be measured only by how impressive the individual task looks.
What the World Rests On
Ten years from now, the street outside my window may look much the same. There will still be cars on the road, people in the supermarket and someone running late for an appointment. Perhaps that will be taken as proof that the great AI revolution never arrived.
But beneath the familiar surface, billions of small judgments and actions may have moved into a new layer of intelligence. Not all at once and not as part of a single plan. One customer case, one calendar appointment, one choice and one yes or no at a time.
It was difficult to see the modern internet in 140 characters. It may be just as difficult to see the AI society of the future in an agent filling in a form, or a model choosing among three categories.
The deepest technological revolutions do not necessarily change what the world looks like first. They change what the world rests on.
Sources and Further Reading
- Founders Fund: What Happened to the Future?, 2011. The manifesto became known for the line “We wanted flying cars, instead we got 140 characters.” See also Jennifer Schuessler’s conversation with Peter Thiel and David Graeber about technological stagnation, The New York Times, 2014.
- Meta: Muse — AI that gets it done.
- TypeSafe: Introduction to Jev, System One and Models.
- AI Finds the Connections We Overlook — on AI at the boundary of human knowledge.
- When Stories Grew Hands — on agents moving from language to action.
- When the Interface Became Optional — on intentions being translated directly into digital systems.