The Dissolved Middle

What Happens When AI Eats Every Arbitrage Simultaneously?

#

Margaret pays a woman named Linda $400 every April to do her taxes.

Linda is not doing anything Margaret could not, in principle, do herself. The tax code is public. The forms are free. The instructions are available online. Every number Margaret needs is on documents she already possesses. Linda’s service is not access to secret information. It is translation. She converts a system designed to be incomprehensible into outcomes Margaret can act on. Four hundred dollars for the privilege of understanding what the government requires of her.

Linda is an arbitrageur. She profits from the gap between what the tax code says and what Margaret can understand. She did not create this gap. The gap was created by decades of legislative complexity, lobbying-driven exemptions, and institutional indifference to whether citizens can actually comply with the systems that govern them. Linda simply positioned herself at the point where incomprehension meets obligation, and she charges a fee for passage.

She is not alone. She is one of millions.

The Toll Booth Economy

#

Look at enough industries and a pattern emerges that economics textbooks tend to obscure. A remarkable share of the modern service economy is not creating value in any productive sense. It is extracting rent from information gaps.

The insurance industry sits atop at least four simultaneous arbitrages. The insurer calculates your actuarial risk with precision you cannot replicate (information asymmetry). It constructs risk pools you have no mechanism to assemble yourself (aggregation asymmetry). It designs claims processes complex enough to suppress collection of benefits you have already paid for, which this series examined in Parts 44 through 46 as administrative burden operating as profit strategy. And it navigates a regulatory landscape you cannot read (regulatory asymmetry). Four toll booths, stacked. You pass through all of them every time you file a claim, and at each one, value that should flow to you is siphoned away by the intermediary’s informational advantage.

Legal services operate on the same architecture. The law is public. Every statute, every precedent, every regulation is technically available to any citizen who wants to read it. You do not pay a lawyer for access to secret knowledge. You pay because the knowledge, while technically available, is practically inaccessible. The language is specialized. The procedural requirements are arcane. The consequences of misinterpretation are severe. The legal profession is built on the gap between theoretical access and functional understanding, and that gap is an arbitrage as surely as any spread on a trading floor.

Real estate brokerage. Financial advising. Tax preparation. Healthcare specialist referrals. Pharmaceutical pricing. Recruiting. Government relations. Education credentialing. Each of these industries, employing millions of people collectively, is organized around the same fundamental structure: someone stands between you and something you need, and they charge you for bridging a gap that, in a world of perfect information, would not exist.

This is not to say these professionals provide no value. Linda is good at her job. Margaret’s insurance agent has caught errors that saved her money. Her lawyer handled the estate paperwork when David died with competence and genuine care. The value is real. But the value is inseparable from the gap, and the gap is sustained not by the professional’s skill alone but by the system’s design. The tax code does not need to be incomprehensible. Insurance claims do not need to be adversarial. Legal language does not need to be opaque. These complexities are not natural features of the domains they govern. They are architectures, built and maintained, that create the conditions for intermediation.

The modern service economy is, to a remarkable degree, a system of toll booths positioned at points where knowledge becomes inaccessible to the person who needs it.

And AI is dissolving them. All of them. Simultaneously.

The Universal Solvent

#

Consider what happens when Margaret’s AI prepares her taxes.

It reads the same forms Linda reads. It applies the same rules. It pulls the same numbers from the same documents. But it does this at near-zero marginal cost, with no billable hours, no scheduling friction, no need for Margaret to bundle her papers and drive across town. The information gap that sustained Linda’s practice for twenty years collapses. Not because the tax code became simpler. Because the translation cost dropped to zero.

Now multiply this across every arbitrage in the economy.

AI reads insurance policies and identifies coverage gaps the insurer hoped you would miss. It drafts legal documents by parsing the same precedents a junior associate would research, at a fraction of the cost and in a fraction of the time. It calculates your personal actuarial risk, making the insurer’s informational advantage over you evaporate. It matches job seekers to opportunities without the recruiter’s placement fee. It interprets medical imaging, collapsing the referral chain that routed you through a gatekeeper to reach the specialist who could actually help. It compares pharmaceutical prices across formularies, dissolving the opacity that let the system charge you $400 for a medication available for $30 at a different pharmacy.

Each of these, taken alone, is a straightforward efficiency gain. Taken together, they represent something more fundamental: the systematic removal of the informational friction on which an enormous portion of the economy depends.

Part 50 of this series argued that friction is habitat, that the inefficiencies in economic systems sustain the diversity of small producers the way underbrush sustains an ecosystem. That argument holds. But there is another kind of friction this series has not yet fully examined: friction that is not habitat but extraction. Friction that does not sustain diversity but sustains intermediaries. Friction that was designed, not emergent. Friction that exists because someone profits from your inability to navigate a system on your own.

AI does not distinguish between these two kinds of friction. It dissolves both. The habitat and the toll booth disappear together. And the consequences of each disappearance are profoundly different.

What the World Gains

#

The equity implications of dissolving information arbitrages are enormous, and they deserve to be stated plainly before we complicate them.

A farmer in Tamil Nadu who has been selling his crop at prices set by a middleman who knows the market rate and knows the farmer does not, that farmer’s loss has been some intermediary’s margin for generations. When AI gives the farmer real-time market prices on a $50 phone, the arbitrage collapses. The farmer gets closer to fair value. The middleman’s informational advantage disappears. This is not a minor efficiency gain. Across global agriculture, these information asymmetries have transferred trillions of dollars from producers to intermediaries over decades.

A woman in rural Bihar who needs a medical diagnosis has historically faced a geographic arbitrage: the specialist exists, but in Mumbai, five hundred miles and an unaffordable train ride away. AI-assisted diagnosis does not replace the specialist. But it collapses the distance between the woman and the knowledge the specialist possesses. The arbitrage was never about the specialist’s skill. It was about the accident of geography that made the skill inaccessible.

A first-generation college student in Mississippi who needs legal advice about a predatory lending contract has faced an expertise arbitrage: the knowledge to identify the predatory terms exists, in law libraries, in precedent databases, in the minds of consumer protection attorneys, none of whom she can afford. AI collapses the distance between her and that knowledge. The contract terms do not change. Her ability to understand them does.

These are not hypothetical futures. They are happening now, unevenly and incompletely, but visibly. And they represent what may be the largest transfer of informational power from intermediaries to individuals in human history.

Part 26 of this series called this the democratization of cognition. That term holds. But we can be more precise. What AI democratizes is not cognition in general. It democratizes the specific cognitive capabilities that intermediaries have monetized. It gives everyone the functional equivalent of a tax preparer, a legal researcher, an insurance analyst, a financial advisor, a medical interpreter. Not the best version of each. But a version good enough to close the gap that made the intermediary necessary.

The toll booths are coming down. For the billions of people who have been paying tolls they could not afford, on roads they had no choice but to travel, this is unambiguously good.

Where the Value Goes

#

But the tolls were also someone’s income.

Linda, Margaret’s tax preparer, employs two assistants. She rents an office on Route 9, the same road where Dot sells honey. She eats lunch at the diner. She sends her kids to the local school. The $400 Margaret pays her circulates through the community in ways that are economically small but socially significant. Linda is not wealthy. She is middle class in exactly the way the service economy promised: she found a gap, she filled it, she built a life.

When AI collapses the tax preparation arbitrage, Linda’s income does not transfer to Margaret. Margaret saves $400, yes. But the value Linda captured, the translation service she provided, does not redistribute. It evaporates. Or more precisely, it is captured at a different level entirely.

The AI that replaces Linda’s function was built by a company. The company charges Margaret $20 a month, or nothing at all, subsidized by data collection and adjacent services. The difference between Linda’s $400 and the AI’s $20 is not savings that Margaret banks. It is value that has moved from a local professional to a platform owner, from a distributed economic actor to a concentrated one.

Now multiply this by every arbitrage in the economy.

The insurance analyst whose expertise is replaced. The paralegal whose research function is automated. The financial advisor whose portfolio recommendations are algorithmic. The recruiter whose matching service is disintermediated. The real estate agent whose market knowledge is commoditized. Each of these people occupied a toll booth, yes. But the toll booth was also a livelihood, embedded in a community, generating local economic circulation. The tolls were extraction in one frame and income in another. The distinction between rent-seeking and employment depends entirely on where you stand.

When millions of toll booths collapse simultaneously, the value does not simply dissipate. It reconcentrates. The platforms that provide AI services capture a thin slice of each dissolved arbitrage, but they capture it at global scale. A million $400 tax preparation fees, individually small, collectively represent an industry. When that industry’s value shifts from millions of Lindas to three or four platforms, the total economic activity may remain the same. The distribution changes beyond recognition.

The New Shape

#

This is where the series needs to be honest about what we are seeing and what we are not.

The conventional story about AI and inequality runs in one of two directions. The optimistic version: AI democratizes access, levels playing fields, gives everyone tools that were previously reserved for the privileged. The pessimistic version: AI concentrates power, displaces workers, widens the gap between those who own the technology and those who are subject to it.

Both are correct. They are not contradictions. They are descriptions of the same process viewed from different positions.

What AI is doing to inequality is not making it larger or smaller. It is changing its shape. Across the vast middle of the economic spectrum, AI is genuinely leveling. The farmer gets fair prices. The patient gets a diagnosis. The student gets legal literacy. The gap between the bottom and the middle narrows, measurably and meaningfully, as information arbitrages collapse.

But the gap between the middle and the top does not narrow. It widens. Because the value extracted from every dissolved arbitrage flows upward, to the owners of the platforms that replaced the intermediaries. The local tax preparer’s income becomes a tech company’s revenue. The insurance agent’s commission becomes an algorithmic efficiency captured as profit margin. The recruiter’s placement fee becomes a platform subscription.

The result is an inequality curve that is flatter in the middle and more vertical at the extremes. More people have access to more capabilities than ever before. And a smaller number of entities control more economic value than ever before. Both of these are true simultaneously. They are features of the same transformation.

Part 57 described the invisible tiers that sort people into different levels of AI-mediated effectiveness. This is the economic complement: the invisible redistribution that sorts value from distributed local actors to concentrated global ones, while everyone involved experiences the transaction as an improvement.

Margaret saves $400 on tax preparation. She does not see that Linda closed her office. She does not see that the diner lost a regular customer. She does not see that the platform that prepared her taxes reported $8 billion in quarterly revenue, up 40% year over year, and that its three founders are now worth $120 billion collectively.

She sees the saving. She does not see the shape.

The Complexity Feedback Loop

#

There is a darker thread woven through all of this that connects back to the administrative burden argument from Parts 44 through 46.

The arbitrages AI is dissolving were not all naturally occurring. Many of them were manufactured. The tax code is not accidentally complex. Its complexity is the product of decades of lobbying by industries that benefit from provisions most citizens cannot understand. Every exemption, every special rate, every arcane filing requirement was advocated for by someone who could afford a lobbyist, and the cumulative effect is a system so opaque that ordinary citizens must pay professionals to comply with it.

This means that the complexity which created the arbitrage was itself a form of value extraction. The industries that lobbied for complex tax provisions captured value through the provisions themselves. The tax preparation industry captured value from the complexity those provisions created. The citizen paid twice: once through the tax code’s substantive provisions and again through the cost of navigating them.

AI disrupts the second layer of extraction without touching the first. Margaret no longer needs Linda to translate the tax code. But the tax code is still written by and for interests more powerful than Margaret’s. The toll booth is gone. The road still leads where it always led.

This pattern repeats across domains. AI can help you understand your insurance policy, but it cannot change the policy’s terms. It can help you read a predatory lending contract, but it cannot make the lending less predatory. It can translate the law, but it cannot rewrite it. The informational arbitrage dissolves. The structural advantage persists.

AI is the most powerful tool for closing information gaps ever built. Information gaps are not the only kind of gap. And the gaps that remain after the information gaps close may be harder to see, because the absence of the obvious barrier makes the structural barrier invisible.

What Margaret Does Not Know She Is Losing

#

Margaret will not miss Linda’s service. The AI does it better, faster, and cheaper. She will not miss her insurance agent’s explanations. The AI explains more clearly and does not have a financial incentive to steer her toward a particular policy.

But Margaret may, eventually, miss the texture of a world in which these functions were performed by people she knew.

Linda was not just a tax preparer. She was someone Margaret saw once a year who asked about her grandchildren. The insurance agent was someone who came to David’s funeral. The pharmacist who explained medication interactions was someone whose daughter went to school with Sarah. These relationships were economically marginal. They existed in the cracks of transactions that were, at their core, arbitrages. But the cracks were where something human lived.

This is not an argument against efficiency. It is an observation that the toll booth economy, for all its extractive structure, was also a distributed network of human connection. The tolls were real. But so were the people who collected them, and the relationships that formed around the collection, and the community fabric those relationships sustained.

AI removes the toll. It does not replace the person. And the question of what a community looks like when all the small intermediary roles have been dissolved is a question we have not yet begun to answer.

Part 50 warned that friction is habitat. This is the human corollary: some of the people sustained by the friction were not just economic actors. They were neighbors.

The Honest Reckoning

#

We are not arguing against the dissolution of arbitrages. The equity gains are too significant, too real, too important to the billions of people who have been paying unjust tolls for access to knowledge and services that should have been accessible all along. The farmer who gets fair prices, the patient who gets a diagnosis, the student who gets legal understanding: these are moral goods, and defending the toll booth economy to protect the toll collectors is not a position this series can hold.

But we can hold the complexity.

The dissolution is genuinely good for billions. The dissolution is genuinely devastating for millions. The dissolution concentrates value in ways that create new power asymmetries potentially more extreme than the ones it resolves. The dissolution strips communities of economically marginal but socially essential human connections. All of these are true simultaneously.

And the multi-billionaires? They are the ones who own the solvent.

The person who built the platform that replaced Linda captures the value of every Linda, globally, simultaneously. This is not labor. It is not even traditional capital deployment. It is ownership of the mechanism that dissolved the arbitrages, and it generates returns that bear no relationship to any human scale of effort or contribution. The medieval lord extracted rent from the farmer who worked his land. The platform owner extracts rent from the dissolution of the very concept of rent extraction. It is arbitrage all the way up.

We do not know how to govern this. Our economic frameworks assume that value creation and value capture are related, that profit reflects contribution, that markets distribute rewards in rough proportion to productivity. None of these assumptions survive contact with an economy where the primary source of value is the elimination of information gaps, and the primary beneficiary is whoever owns the tool that eliminates them.

What Would Be Different

#

Imagine an alternative that is not utopian, just honest.

The equity gains of AI-dissolved arbitrages could be captured publicly rather than privately. A government that provided AI-assisted tax preparation, not as a product but as a service, the way it provides roads, would save citizens the cost of intermediation without concentrating the value in a platform owner. The same logic applies to AI-assisted legal understanding, insurance navigation, healthcare interpretation, and benefits enrollment. Parts 44 through 46 argued that administrative burden is fiscal policy. AI-dissolved administrative burden could also be fiscal policy, if the dissolution were public rather than private.

The communities that lose their intermediary professionals could be supported through the transition rather than left to discover the loss when the diner closes. This requires acknowledging that the toll collectors were also neighbors, that their displacement is a social cost even when their function is replaced by something better, and that social costs require social responses.

The concentration of value at the platform level could be addressed through mechanisms that capture some portion of the gain for redistribution. This is not a novel idea. It is taxation applied to a novel form of value creation. The difficulty is not conceptual. It is political. The entities that would be taxed are the same entities that have the most sophisticated AI-assisted lobbying capabilities, which creates a recursive problem this series is not equipped to solve but is obligated to name.

None of these would eliminate the disruption. The toll booths are coming down regardless. The question is whether the transition is managed or unmanaged, whether the gains are shared or captured, whether the communities that lose their intermediaries are supported or abandoned.

We are currently choosing the default, which is no choice at all, which means the market decides, which means the platform owners decide. The most consequential economic transformation in a generation is being governed by the interests of the people who benefit most from it.

That is not a conspiracy. It is a structure. And structures can be changed, if they can be seen.

The toll booths are falling. The question is not whether to mourn them. It is who gets to collect the savings.

This is Part 59 of The Approximate Mind, a series examining how AI might serve human flourishing rather than human extraction. Part 58 explored what happens when a single intelligence decides how many minds it is, and how the boundaries of care become design decisions. This article asks a different question about structure: what happens when AI dissolves the information gaps on which an enormous portion of the economy depends, and whether the genuine equity gains of that dissolution can survive the concentration of value it produces.