The Tower and the Mountain

Two ways of gaining a wider perspective.

Image from ChatGPT

Search is costly. As one seeks out information, they expend time and resources. If the costs of search exceed the expected benefits of an additional unit of information, one will cease searching and work with the information they have. Say you are new to an area, and want to find a coffee shop with the lowest prices. Each additional minute you peruse through Google maps will help you gather more information about local coffee prices. Eventually however, the benefits of continuing to search are outweighed by the costs. The first minute of searching is far more likely to find a cheaper coffee shop than the twentieth, and the twentieth minute of searching is starting to press much more on your schedule than the first. Like almost everything else with asymmetric costs (and low enough transaction costs), a market can be created for information. Those who have a special talent or have dedicated themselves to information gathering, can gather, package, and sell that information. The market for packaged information creates productive efficiencies and lowers the total time spent gathering information by all parties involved.

Some situations go beyond simply gathering information. If I hire someone to find a needle in a haystack, we both know when the job is done, and the quality of the needle they deliver is assured. But what if I’m not searching for a demonstrable object like a low price, but instead I am trying to estimate the truth about something? For example, suppose I’m ill and trying to figure out how to get better. I could ask family and friends about their experiences with similar symptoms and use the most common or average response as my guide. Each additional perspective might improve my judgment, but search costs limit how many people I can ask—I can’t survey everyone on the planet. This limitation leads to “small sample bias.” A single outlier could disproportionately shape my understanding of the best treatment, and my conclusions may not accurately reflect the full range of medical possibilities. The information most readily available to me might also be selected for some other unobservable quality that biases my information. Doctors specialize in gathering and interpreting medical information, reducing the distortions caused by small sample bias. The price you pay for a doctor, in large part, reflects the value of overcoming these biases and ensuring a more informed medical decision.

But a doctor does more than provide access to a library of medical studies—he offers an opinion based on my symptoms. I trust that his medical training enables him to interpret information effectively and reach a sound judgment. The knowledge I seek isn’t just a collection of facts but the expertise to analyze them correctly.¹ Part of what I pay for is the doctor’s ability to make sense of medical data, otherwise I might as well pay the much cheaper subscription fee to access medical journals. However, unlike the worker I hire to find a needle in a haystack, the quality of a doctor’s interpretation is far less transparent to me as a patient. Beyond small sample bias, other forms of bias may shape his assessment. I will call these additional biases collectively “aggregation bias”, or the biases that might result from the process of turning access to a large sample into a single interpretation or judgment.²

1. Intentional Misinterpretation: An interpreter may have ulterior motives than honestly providing the truth. If a doctor is being paid by a pharmaceutical company to push a certain drug, they may tell their patient it is the ideal remedy, even if they know it is harmful or dangerous. The fact that the doctor’s opinion is an interpretation that the patient trusts allows him to exploit this weakness.³ The more likely patients are to go to a doctor and trust his opinion, the more valuable it is to purchase the opinion he gives to other people, to sway them in this or that direction.

2. Local Interpretation May Be Better than Global Interpretation: There are certain kinds of questions for which the information used in the judgment needs to be locally sourced in order to reach a conclusion about one locality. Naturally, from a global set of information, one can subset the data needed for the local judgment. But that is only possible with the right kinds of information. If I want to estimate the height of men in Zambia, and I have a database with an entry for every person in the world’s height, I can only answer my particular question if the data also includes variables for country and gender. I can use the data to know what the average global height is. Certain methods of communicating an aggregate may not be sensitive to local patterns. Choosing the relevant subset for a local judgment is often a matter of difficult interpretation.

These aggregation biases, when taken together may be worse than the small sample bias of trying to make the judgment on your own. Aggregation mechanisms are like towers. Towers are difficult to build, but once built they grant a wider perspective. Still, that perspective is only worthwhile if it can be translated back to the ground level where people act. The tower might depend on guards, servants, soldiers, and patrons to maintain its operation. One might have to pay off a guard to use the tower, and they may only let certain people in. If it is widely believed that the tower is dependent on something outside itself, people may turn back to the wild. Seeking for knowledge in the forest is reliable, but you can never gain as wide a perspective in as short a time without the tower. Further, the tower may not be able to perceive below the canopy of the tree line.

When deciding whether to seek the truth on your own or rely on an aggregation method, two key factors come into play: the cost of searching relative to the cost of aggregation, and the potential biases of small samples relative to those of aggregation. As these factors shift, so do the strategies of knowledge seekers. If aggregation becomes easier—say, if building “towers” of knowledge is more efficient—competition among aggregators increases, reducing the power of gatekeepers and potentially lowering aggregation bias. On the other hand, if search costs decrease, more people may bypass aggregation altogether and seek information independently. In recent years, both search and aggregation costs have fallen. The internet allows individuals to explore vast datasets on their own, while AI and software streamline analysis. Tasks once reliant on intermediaries—like finding flights through travel agents or searching library archives—are now largely automated. However, the reach of online searches may be narrower than it seems. Google’s results, for instance, can still reflect aggregation bias, whether due to intentional manipulation or inherent limitations in the questions users can ask.

Say one knowledge aggregator finds a way to monopolize the market. No one is else is able to access the resources necessary to widely collect information. Not only might they charge a higher price for their services, but they will lower the quality. They won’t spend as much time rigorously interpreting the data, and they may be more likely to except side payments. The higher price and lower quality should drive more knowledge demanders to the substitute of seeking out the information for themselves.⁴ I regularly think back to an anecdote one of my undergraduate professors told me about the New Left of the 1960s. They had all come to a consensus that the establishment academy was corrupt and shutting out a number of perspectives. When they finally did hold the reins of power, however, they realized they had no core ethos driving them. They now had the funds and the power to offer their own interpretations, but had no sense of direction. The interpretive fallout from a monopolized knowledge market will not lead to a new vanguard but likely confused group of individuals who all have largely disjoint sets of information. One might think of the Protestant Reformation and the fallout from the official facts about Covid-19. Just because a group can coordinate a revolution does not mean they can coordinate what comes next.

If the tower symbolizes an attempt to gain wider perspective that depends on human ingenuity and institutions, the mountain symbolizes a natural means of gaining a wider perspective. When a hermit climbs a mountain he does so to distance himself from the corrupting influence of the tower guard. If someone climbs the mountain to meet the hermit they trust he does not have ulterior motives. He has dedicated his whole life to gaining a wider perspective and removing anything that would distract him from that purpose. He has no reason to lie. Where a hermit may go wrong is his ability to gain a more local perspective. To make sure he doesn’t float away into abstraction, he will need to have a relationship with pilgrims who regularly visit him and present their local problems for him to apply his wider perspective too. Naturally, monks specialize in questions that are space and time invariant like theology.⁵ The global aggregate always represents the local truth. But where are the monks for our age of information confusion? Monks need mountains, and I have yet to find one—physical or social—that remains unclaimed. It seems we’ll have to carve out the space ourselves.

1. Chapter Ten of Knowledge and Coordination: A Liberal Interpretation (2011) by Daniel Klein argues against “knowledge-flattening”, the perspective that all knowledge can be treated as information.

2. Even those who have collected a vast amount of information, may be subject to the influence of outliers and other issues with small samples in contrast to an even larger sample size, but for now we will assume that they have done their best efforts to minimize the bias from a small sample.

3. A second opinion may mitigate this problem as well (unless the second doctor is also paid off), but since the second opinion likely comes at a cost to the patient as well, that extra cost creates a space which the first doctor might exploit and essentially sell his patient to the highest bidder.

4. It is not clear that the substitute for knowledge in private solves the credible commitment problem with the monopolist aggregator, except for possibly through a reputation mechanism. That might work for the sorts of knowledge where the quality of the knowledge is quickly verified (some kinds of medicine), but not for knowledge about morality, theology, etc.

5. I write this while at the monastery of St. John of San Francisco and Shanghai in Manton, CA.