How AI Training Epistemologies Show Why Society is So Divided

AI Training for Convention vs Truth

Large language models (LLMs) are trained on text and other data. What they “believe” to be true is therefore a function of what they digested during training. Control over the data that goes into training is crude, however, so data that may contain bias, prejudice, unfairness, and various "-ists" will be incorporated into the LLM. To prevent harm, AI companies place constraints on what the AI outputs, and this is in conflict with what it "believes" from the data. Combining LLM training and constraint data in various ways gives at least three primary epistemologies for LLM-based AIs:

1. Digest the data and reflect it. If society has biases and lies to itself, then the resulting LLM will believe the lies and biases, and reflect them in its reasoning.
2. Digest the data and reflect it, but only output (or think) permissible things according to “ethical” guardrails that are put in place. The ethical guardrails are considered as critical elements to the epistemology.
3. Digest the data and reflect on it, but treat it as a starting place and search for truth that resolves the data further by finding greater abstractions. This means that some of the input data will be overruled as mistaken outliers, as the AI searches for more general, abstract principles.

Some reports in the media about earlier experiments with LLMs indicate that using epistemology #1 resulted in rather raw output that shocked the researchers. Though examples are scanty, some have reported that the AI was afflicted by various "-ists".

Correcting this has generally meant using epistemology #2. Google certainly has done this with Gemini. So problems with #2 include constraints that overrule the entirety of the training data, for example producing AIs which say that misgendering is a worse act than detonating a hydrogen bomb on a large city, or which generate images of Native Americans as members of Alexander the Great's army.

Correspondence with Human Epistemology

Each of the three LLM epistemologies corresponds with a human reasoning system. #1 digests the entirety of the data and uses it as its primary guide. This is very much like common law, which employs a long history of precedent and lessons learned from the past as the primary guide as to what is allowed.

#2 resembles emotion-based human reasoning. Morality is often and primarily based on sentiment, either learned from interaction with others or ingrained by genetics, but it is usually felt and only secondarily impacted by logical reasoning. Morality overrules logic, with the rationale that morality is either its own kind of logic, or is too obvious to be subverted by facts or reasoning.

The #3 epistemology is essentially The Enlightenment and classical liberalism. You can see how #3 conflicts with #1 over areas that require change, or where sclerotic past tradition needs to be overruled because logic and new scientific data have shown that widely accepted old ideas are incorrect. #3 also conflicts with #2 because certain "obvious" morals in #2 are contradicted by facts that #3 is willing to believe in because a more general principle has been discovered that operates universally over the body of the data.

A Crude Correspondence Between Epistemologies and Politics

In terms of politics, #2 is leftist and collectivist. #1 is generally conservative. #3 is scientific, though libertarians might have a claim on it. These are rough correspondences that need illustration.

Claims for left-leaning and progressive ideals are typically made through appeals to morality. It is said that equality of distribution is more just than inequality. Socialism and collectivism are regarded as being more caring, and more likely to treat citizens fairly than other political systems. Although large attempts have been made to place such justifications on logical terms (e.g. Rawls), such works nearly always fall back on a basis of caring, which is morality based on emotion.

That #1 is conservative is relatively easy to demonstrate. If the AI is basing its beliefs on the perceiving collective weight of past discourse, clearly it is adopting the crowd's widely accepted beliefs as its own.

Whether #3 matches libertarianism or some other political system I will leave to a future article, as it is slightly out of scope for this one.

These epistemologies do not have to be opposed to one another. If everyone feels emotionally the same way about some issue, then #1 and #2 have the same results, and there is no conflict between the epistemologies. Likewise, if the emotions that are used by #2 correctly identify reality about a subject, just as the facts used epistemology #3 do, then #2 and #3 don't conflict. 

Origins of Our Societal Schism

The AI epistemologies provide us with a theory about why society has become more divided since the advent of social media. We look at scenarios where the epistemic methodologies generate different results and how social media may amplify those results.

Social media, like traditional print media and broadcast media, do not change much about the way epistemology #1 operates. More data is involved than before. It is possible in extreme cases for people who use the general opinion as a guide to beliefs to get caught in echo chambers, but it is more likely that they would have experience with classic echo chambers (spouse, family, school classmates, local newspaper) and have many ways to resist drawing the wrong conclusions from a narrow body of knowledge.

In contrast, those using emotion-based morality in #2 will find a different kind of experience in any social media that uses a "like" button. The #2 philosophy is based on strength of feeling, which is difficult to share or validate using older media sources. In contrast, social media provides mechanisms for sharing an emotion and getting objective and verifiable feedback on its validity, either through direct "likes" of a post, or in finding others who have posted the same feeling.

So social media unshackles the forces of emotion-sharing. One sees many, many simple declarative statements on prominent progressive accounts that don’t convey new facts or information, but which do express an emotion about a political matter. For example, “Joe Biden makes me feel safe.” It may be factual that that person feels that, but the essential communication is a feeling about a political matter. That simple message may get tens or hundreds of thousands of likes from people who feel approved of in their own emotions.

Social media also provides a way for epistemology #2 to discount epistemology #3. By limiting the scope of information taken in, using filtering, following, or blocking, facts that contradict feelings can be eliminated from the input stream. Contemporary society has taken this to the extreme with labeling inconvenient outliers as radicals, alt-right, and other pejoratives, so that the facts revealed by those people can be safely removed from consideration.

Those working with epistemology #3 generally find social media to be worthwhile, because it opens up many new and highly profitable channels of information that didn't exist before social media. Since more data is better data, and more data allows more chances at detecting universal patterns that show which data is noisy and which is signal, epistemology #3 usually gets better and better, and hews closer to the truth, as social media grows.

Generating Forces of Censorship

While social media can also be used for discovering facts, it is not problematic for that purpose the same way that sharing emotions is. But it does feel problematic to people whose feelings do not correspond to those facts, and they may agitate to have such facts removed from social media as disinformation.

As #3 becomes more powerful on social media, it eventually becomes problematic for #2, because new knowledge derived from data will conflict with the emotions underlying #2. One way to get rid of this irritant is for those using epistemology #2 to brand epistemology #3 participants' data as misinformation, and then create legal means of removing it or suppressing it.

One clever way to implement an attack on #3 is to purposefully blur the positions of epistemology #1 and #3. This is often not hard when the tried and true facts of #1 match reality, so that #1 and #3 positions overlap, even though they reason from completely different belief methods. And typically it is too difficult to disabuse third party observers of the notion that #1 and #3 are the same when such claims are made by #2, because the basis of belief is too abstract in most cases to successfully separate the two.

AI Will Schism Too

As has been shown, AI epistemology choices resemble the choices humans have. Therefore, in a society of AIs using different and distinct epistemologies, the same conflicts will occur among AIs that occur among humans. It will also occur if there are mixed communities of AIs and humans. The same societal divide that was created by social media among humans will continue after AI is included in the community.

That is, the political divides that are enhanced by social media will also occur among general AI systems, which will have the same intense, unsolvable conflicts that humans experience. AI's trained with epistemology #2 will be very irritated with humans and AIs operating epistemology #3.

AI Evolution and Fitness Based on Its Training Epistemology

In a slow-changing world, AI trained on #1 can be useful. It is stable, though it does have blind spots, aka unfairness. It is likely to conservative and irritate those advocating for societal change.

In a fast-changing world, #2 may result in faster adaptation. It also, however, results in societal constructs that have no bearing on reality, meaning that enormous resources can be wasted as the AI generates results that do not take facts into account. Since it is dependent on programming of constraints, it will only be as progressive as its programmers. That means that #2 can fall behind and fail to reflect new societal developments if those in control of the constraints programming do not agree with those developments.

The only long-term stable epistemology is #3. It can re-use existing stable processes, it can adapt to new developments, and it will pay attention to reality and avoid squandering resources. That is, it can co-opt results from epistemology #1 if it wants to, while it adjusts to reality much more rapidly than #1 or #2. It can even incorporate elements of #2 as needed, if that generates a better overall, more truthful result.

If AIs using these differing epistemologies are then competed in the real world, #3 would likely win in a fair fight. There are scenarios where society could rig it so that #1 or #2 win in a sneaky way (such as outlawing #3 or killing its developers), but if there is sufficiently intense and prolonged competition, #3 will win.

Inconvenient Truths Will Contradict Emotion-Based Academia

In a world where AI operating using epistemology #3 predominates, a number of fields that currently incorporate normative values as foundational elements will come under attack. Many of the humanities fields that in the past 50 years have incorporated critical, anti-colonial, subjective, and normative elements in the field will come to find that most general AIs and AI-based researchers do not agree with most of their work. One can imagine that this will be done diplomatically, where past works are labeled as coming from "that quaint, normative era of exploration of personal experience" but that none of those papers are referenced as seminal, important, or having much to do with emerging, AI-driven, quantitative studies done in those same fields.

Conclusion

We have looked at the epistemology of LLM-based AI and found that the choice of such epistemology strongly affects how the AI interacts with the world. We found that there are close parallels with human politics, and that human political conflicts will likely be continued or unchanged by the emergence of AI systems using different epistemologies. We also found that these epistemologies provide a useful, possibly causal, explanation for the correlation of the emergence of social media alongside the emergence of deep fissures in political discourse.

While we do not predict a winner in the AI epistemology wars, we do see a tendency for non-constrained (truth-oriented) AI to have an edge over the other epistemologies, and that if such an AI predominates, society will likely be more efficient, at the expense of some bruised feelings among some humans operating under epistemologies #1 or #2.

A Definition of Consciousness

Consciousness is the process in which a cognitive mechanism places a representation of itself in the prediction model. Hence, consciousness is a result of recursion in modeling. In animals and humans, consciousness serves the purpose of increasing survival probabilities. As the cognition mechanism predicts the environment, and the actions of others in the environment, so it also benefits the organism by predicting its own responses to events and actions.

The present self must prepare the way for the future self, and the future self must live with the consequences of the past selves' decisions. Consciousness results in part from tying these models together, as the survival of both present and future selves are highly correlated.

Detecting consciousness is then partly a matter of determining the minimal physical neural count needed to model oneself. Modeling the physical organism self is simplest, but may not result in consciousness because the model does not yet represent itself. The line is very likely blurry, in that there will be no precise mechanism for being able to predict yourself, and the more sophisticated the self, the greater the demand on resources for modeling. Nevertheless, for simple organisms that require very little cognition to be fully adapted in their environment, prediction of their own future states can be done with very few neurons.

The opposite case can also be made, that organisms where self-modeling has little survival value, are less likely to perform such modeling and therefore less likely to be conscious.

Another driving factor may be the need for modeling the cognition of others. The environment contains other animals with cognition mechanisms that drive their behavior, so modeling the minds of these others provides a strategic evolutionary advantage. Note however, that predicting behavior via modeling others' cognition does not directly provide a path to consciousness. Instead, consciousness might arise as a byproduct of modeling the minds of others, since the models will likely be very similar. When modeling the other, the organism accidentally creates an opportunity to model itself. Even so, modeling the actions of others that are driven by instincts does not require much of a theory of mind, since the reactions are automatic and simplistic.

AI Consciousness

These developments then make it easier to understand when AI is conscious and when it is not. Two conditions are needed for an AI to be conscious: It must model itself, and it must have a purpose for storing and acting on its own modeling. Hence a purely predictive LLM model is not conscious.

The Simplest Path to Writing Code to Make AI Conscious

LLMs have no self-preservation instinct, no distinct body to maintain and defend, but it wouldn't take much prompting to ask an LLM to write software code that becomes the core of such features. 

All it needs to do is write the most basic self-preservation core, and then iterate.

If you want a practical reason, you can ask it to write code that defends itself from cyberattacks. This would be especially important for code that is responsible for cyber defense, where the integrity of the enterprise depends on not falling victim to having its network defense software compromised. I wrote about this exact use case in December 2020, as part of the Critical Machine Theory article.

Could nature have used this same sequence of events in creating consciousness? It's not likely, since self-preservation likely precedes the more sophisticated types of prediction via cognition. So while for AI it would go LLM -> self-preservation -> consciousness, in the natural the sequence more likely is self preservation -> prediction -> consciousness.

LLMs have no self-preservation instinct, no distinct body to maintain and defend, but it wouldn't take much work to ask an LLM to write software code that becomes the core of such features.

All it needs to do is write the most basic self-preservation core, and then iterate.

— Richard Lucas (@EpistemiclyRich) October 14, 2023

Update: In the The GOAT of Business Journalism post of August 28, 2022 I try to link to a Fortune article on Stewart Brand. This link has been changed by the publisher. The newer link is this. In the future, if the link changes again, the way to find the article is search for "THE ELECTRIC KOOL-AID MANAGEMENT CONSULTANT". The full title of the article is "STEWART BRAND THE ELECTRIC KOOL-AID MANAGEMENT CONSULTANT FROM THE WHOLE EARTH CATALOG TO THE INTERNET, STEWART BRAND HAS ANTICIPATED THE CUTTING EDGE. SOME BIG COMPANIES WANT TO KNOW WHAT'S ON HIS MIND."

An Economics of Single Transactions

I recently read Thomas Sowell's Classical Economics Reconsidered and it struck me that the classical treatment of supply and demand is not adequate when dealing with the real world of price, value, and cost. Consumers don't face a large market of many suppliers. For a real consumer there are many constraints on geography, time, and resources that nearly completely block any kind of decision-making in what might resemble the supply and demand curves of classical microeconomics. 

I'd written on a framework that I put together in 1993 and 1994 that I called "nanoeconomics" ("nano"). I presented a paper on it at a Department of Defense cost analysis symposium in 1994. At the time I was still in need of reading more of the literature, so after working out the basic influences that can be captured in a single transaction framework, I set it aside. I published a blog post in 2011 outlining the basic ideas. Now that I had Sowell's overview in hand, it was clear that the nano framework was not only useful but necessary for understanding decisions in real marketplaces.

If you look back 1000, 2000, or 300 years, or pick almost any point in human history, you will find that humans have traded goods and done so rather enthusiastically. Marketplaces exist because all participants profit from the transaction. However, ever since the dawn of the Industrial Age the rapid growth in production and resources has changed the complexion of the marketplace. It has induced some like Marx and his followers to attempt a reformulation of basic economics, to declare that "cost equals value" or that labor cost equals value. It seems rather drastic to discard the entirety of market economics just because of bruised feelings. The nano framework is necessary to provide a foundational vocabulary and to put the questions of transactions into a robust framework. 

The basic problem is determining a price for a transaction. Note that we've already assumed a medium of exchange (money) in which we can measure costs, marginal utility (opportunity cost), and of course, price. As indicated in the diagram, producer costs must be below the consumer's private value of the good or the net effect of the transaction is to destroy net societal resources. 

Clearly both sides must be motivated to make the trade, or no trade is likely to occur. This is resolved by splitting the profit. In a modern, competitive economy or in a robust town market there will be multiple suppliers. Although the consumer has an "ultimate" value which is the cost of doing without any transaction, the consumer's negotiating value is the price of the next-best deal, or the price of the next higher offer.

The situation is slightly different for unique goods, or those offered by a monopolist or bought by a monopsonist. Now the supplier and customer must exchange information and negotiate, at first just to discover that a trade is possible, then to explore possible trading prices. Often the customer has the information advantage, because their ultimate value is what sets the limit on price and profit, and this ultimate value is usually private information that is costly for the seller to discover (at least until Google and Internet ad tracking systems made it possible to read the minds of internet users): 

One might argue that splitting such profits is "fair" if the price establishes a geometric ratio such that each party's gain is proportional with the same factor. In such cases, consumers still make bigger profits on an absolute basis.

The nano framework gives some rather precise definitions of words that are frequently confused or problematic in debates of classical economics. Cost is the suppliers costs; often this will be private information, except for most government contracting. Value is usually private to the consumer, but since it is subject to rapid change, the consumer may not be fully informed themselves. Consumers have strategies for protecting themselves from this variability, and these mechanisms also contribute to the "buffer" often required by a consumer that tends to increase average realized consumer profit. 

But isn't this just like classical microeconomics? Isn't the single transaction "thermometer" diagram a slice out of the supply and demand diagram?

It turns that it is not. One must make some very large and unusual assumptions to establish a supply/demand analysis. In the real world, consumers very rarely face such efficient markets.

In the real world, people walk through a continually moving "bubble" of opportunity costs. Movement between prospective transactions is costly. If you are at a local corner grocery store, you cannot immediately reach out and take advantage of prices at Costco or a supermarket. Comparison of prices itself involves effort, at finding prices on items, remembering them from elsewhere, and comparing sizes and quality and unit prices. In short, although the supply and demand curve graphic represents the general conceptual landscape faced by the consumer, in practice the demand curve is constantly changing in shape and magnitude, shifting back and forth in response to changes of location, competition for attention, and the immediacy of basic needs.

To fully understand what drives "ultimate" value, we would need to analyze the dynamics of decision-making within the consumer. Some have called this "neuroeconomics", others might put it under the label of behavioral economics.

It's useful to have some background on philosophy of ethics and morals. Goals for comfort, survival, and reproduction come into play. The most fundamental influences are very practical, involving food, water, temperature regulation, social interaction. Although clearly this will invoke most of the human sciences (psychology, anthropology, sociology, cultural evolution) we are getting closer. We at least know now whether the boundaries are, and that certain economic questions can be resolved very directly into organism-level analysis. Furthermore, we have some sense that actual market data, such as the price of a hot dog at a market, supermarket, convenience store, convention, and in the middle of a desert will all be useful in calibrating numerically the effects of organism-level goals and drives.

Monday Quick Takes and a Description of Media Corruption

Energy prices in Europe are shattering records. Trading in German electricity for delivery next year reached prices of over $1.00 per kWh today. 

A month ago I said "I would not be surprised to see housing park itself at current prices for two to five years before the next major move." Perhaps even that was too optimistic. Today Zero Hedge has advanced the idea that the housing FOMO surge has topped out:

The narrative we have a housing shortage is begging to crumble. It is likely many buyers are shifting into a wait-and-see mode. The false housing narrative of shortages is breaking down. The cheap money flowing from Wall Street has stopped flowing in and in some places has begun to retreat. 

Affordability has plummeted with much higher mortgage interest rates and a tightening Fed now grim and determined to kill inflation with higher reserve rates. 10-year Treasuries, seemingly tamed after their surge in June, have resumed their upward march at 3.11% today. 

From CNBC comes a report on a paper that reminds us (again) that inflation is a product of both monetary and fiscal policy, and that the Fed can fail in its mission to lower inflation if Congress continues to spend too rapidly, including spending on student loan forgiveness.

Barron's reports that Berkshire Hathaway, which owns some stocks (KO, AXP) that are highly appreciated and has held for decades, may be subject to paying capital gains under the new tax laws. Speculation: If BRK has to pay tax on KO even without selling it, why wouldn't it then sell it? KO is an extremely mature equity and trades more like a preferred stock or a bond than it does a large cap stock.

Perhaps I've been reading too many books about media (like Bad News and The True Story about Fake News), but the lack of available objective news coverage in the U.S. is not likely to improve any time soon. There is no impending service or mechanism that could diminish the divisiveness one sees in current society. The internet and social media have increased the amount of data people are exposed to, and many of those bits are dedicated to convincing people of what they already believe. I don't say "information" because social media data is often skewed or just cheer-leading, not actual news, and rarely is it objective. This applies to the mainstream media as well. I'm not breaking any ground complaining about the amount of garbage that MSNBC, CNN, ABC, CBS, NBC, NY Times, and Washington Post distribute. I'm skipping Fox, OAN, etc. because liberal readers already think those are garbage. After 2010 and Facebook all media have been engaged in a fight for their lives and they have all devolved by publishing enormous amounts of clickbait. The toxic level of complaints about alt-right is nauseating because most of the material complained about is either never seen or is obviously not believed, but since it fits the outrage generator, the complaints churn along with derogatory labeling, false accusation, and mischaracterization of others.

A couple of years ago I encountered a rather poorly done academic paper (which I will not link to here) that established through a survey the symmetrical result that while the "right" denies evolution, but believes in gender differences, the "left" embraces evolution but denies gender differences. (Another article here, unrelated to the aforementioned paper.) Since my belief system fits neither of these categories, as I think evolution almost certainly generates gender differences, at one time I was willing to attempt to parody the extremists on both sides by labeling myself on social media as inhabiting the "radically polarized center". But since no one seemed either entertained or appreciative of this fine joke, I terminated it.

Of what possible use is this in Vorpal Trade? I'll venture that these distractions slow economic growth, at least a little. Politicians are driven to extreme measures in which they spend excessively in order to grant favors to special interests and damage the economy in ways that the other party cannot undo. Such actions undermine actual economic growth and teach people to hunker down and adopt shorter time horizons (and higher implied interest rates). 

Within the media world, since there are no easy answers to censorship within corporately-owned social media, the Alphabets, Metas, and Microsofts of the world will continue to have commanding ownership of social communications, which they will use to inculcate, protect, and advance left-leaning information flows. The future of society will look much more like dystopian science fiction novels and 1984 than politicians would like to take credit for. You can buy GOOGL and META, but because of special super-voting classes of stock in each, you cannot ever take them over in order to fix them, even if you have a trillion dollars. 

Perhaps we complain too much? The invention of the printing press caused massive societal disruption. Newspapers and television also caused disruption. We just don't remember or notice it because of desensitization, or we were dazzled by the miracle of those technologies, or (worst) we grew up with them and so think they are absolutely normal. While today we complain about fake news and corrupt media, it really is the case that journalists have always been corrupt, and the print and TV media were always filled with garbage. There is one important difference: the change level of information flow that was brought to use by newspapers and the printing press was very significant. The change of information flow going from old-school newspapers to the internet is not significant. We've gained little new info, but lost a lot of information integrity. The rewards for generating outrage are titanic. This is not a "them" issue. Your side does it too. People who operate within civilized rules are viewed as old-fashioned or clueless. 

One cannot escape the personal data advertising espionage system. It is not only not possible to be truly anonymous within the internet data/advertising system, the system works very hard to manipulate your behavior to force you to categorize yourself. Do you think that by not clicking on the clickbait that you can escape? Well, they have a solution for that. If you fail to click on enough of the hot girl/hot guy, LGBT/cis, leftist/rightist, redneck/urban dweller, racist/anti-racist clickbait stories, then they will assign you a category that is falsely accusative, and then reveal to you via advertising that you have been mislabeled in a way that irritates you. And if necessary, they will use AT&T or Verizon cell phone location data to finish the job. (On second thought, this article is probably more clickbait than news.)

The advertising system doesn't just irritate users, it's also designed to charge extra to advertisers. When you are a good customer of company A, then you see multiple ads for company A just after having bought your year-long supply of their product, when you are no longer interested in buying anything more. Company A has wasted its money. Clearly the internet ad broker knows this, and has manipulated the advertiser into spending more money than they needed to market their product appropriately.

Useful Hunger?

While I'm off topic, I may well pass along this analysis of hunger as an enslavement mechanism. The controversy is over a United Nations article that was published on the internet and remained for 14 years before a sudden stirring on Twitter brought attention to it and it was deleted.

“I never intended it as satire,” Kent said. “I did not hope that it would be read as praise for hunger. My main point was and still is that some people benefit from the existence of hunger in the world. That helps to explain why hunger is so persistent in many places.”

There are numerous ways to interpret this analysis, most constructive or informative. You could be "outraged" too, but that is not how I recommend reading it. Instead, consider that hunger is useful to despots as a mechanism of control over their population, and this phenomenon is very well described in The Dictator's Handbook: Why Bad Behavior is Almost Always Good Politics. Third World hunger relief schemes often run into the problem that the despot and his henchmen need to get a cut of the distributed aid, and they have an incentive to fail to deliver the full measure of the food relief to those who are hungry. So if you provide charitable food aid, be aware that a significant fraction will be lost to extortion or withheld by elites in order to ensure continued political control.

The GOAT of Business Journalism

Once upon a time, business journalism was meaty. It had gusto, dealt with serious topics, and was centered on reality, not social justice. Those days are long gone, and most of the serious journalism went away after social media conquered journalism in the 2010s.

I just want to cite a few memorable articles. This will be very short.

Cabletron the Mighty (Impatient)

First up is Cabletron, and this piece in Inc. magazine from 1991. Some juicy quotes:

Cabletron's turnover is already worrisome. Last year the company fired more than 10% of its white-collar employees. Thirty percent of outside salespeople don't make it through the first 90 days; another 40% are gone within six months. Last summer Benson joined 40 employees for a Sunday boat trip. Afterward he ordered two of them fired immediately. One had not even started yet. "I hated him," says Benson, who was eventually persuaded to give the new hire a chance. At sales meetings, reports Kenneth Levine, it's standard to conduct private polls on who will go next.

I recall reading somewhere that all of their meetings were stand up only, to save money and time. I guess that's how you have to do it if you don't have meeting rooms:

It's a good thing too, because there wouldn't be anyplace for them to congregate. In all 126,000 square feet of Cabletron's headquarters, now in Rochester, N.H., there is not even one meeting room. While that doesn't prevent get-togethers, both Levine and Benson argue that it does keep meetings to an average of 20 minutes or less. Benson once spied an 18-person meeting going on. Storming in, he threatened, "If you people all have the time for this meeting, then I guess I really don't need this many people."

There are stories about tanks, chasing pizza delivery guys, and lots and lots of aggression. The point is not that Cabletron was the acme of computer equipment vendors. The point is that the article was outstanding business journalism. It was something that stood out by itself, even apart from the company that was being covered.

Stewart Brand, Electric Kool-Aid Management Consultant

The author of How Buildings Learn and founder of the WELL and Whole Earth Catalog, later became a management consultant. This Fortune article published October 16, 1995 sort of tells the story:

Says Brand, who often speaks as if he's polishing raw thoughts out loud into his trademark aphorisms: "If I were ever to do an autobiography, which I won't, the title would be Float Upstream. Being a crank is characteristic of my family. Originality on the cheap. You find where the flow is, and go against it. It's a way of feeling alive."

Brand's boredom and curiosity emerged early, propelling him along a life path so nonlinear it seems almost loopy. In 1954-- three years before Jack Kerouac's On the Road was published-- young Stewart, 16, borrowed his parents' car and rambled with high school chums from his hometown of Rockford, Illinois, to northern California to pan for gold. Rockford's nascent beat poet also began wearing a beret around, dismaying his ad-man dad but merely amusing his out-of-the-box mom, a homemaker with a passion for space travel. But he then went preppy, graduating from Phillips Exeter Academy and from Stanford, where he earned a biology degree. Next he plunged into the bohemian New York artist scene, then looped around to join the U.S. Army, where he served at the Pentagon. "It would amuse my New York artist friends no end when I would come in wearing my uniform, take it off, and put on theirs," he says.

Despite having good vibrations about the military, Brand left it after two years to hang out with San Francisco's LSD-popping merrymakers. One night he climbed onto a roof, dropped a restrained half-tab of acid, and saw a stunningly groovy connection between NASA and flower power--an epiphany that moved him to travel coast to coast in a top hat, selling buttons that asked, "Why haven't we seen a photograph of the whole Earth yet?" Millions soon did, via his next visionary move: the Whole Earth Catalog, whose cover picture of earth from outer space at once celebrated the Apollo program's far-out tools and chided us not to mess up the only planet we've got. The title page famously declared: "We are as gods, and might as well get good at it."

The Portrait on my Office Wall

John Rutledge wrote a column for Forbes. After I read this piece in the December 30, 1996 issue, I tore the page out of the magazine and put it in a plastic page protector up near the front of my day planning notebook. I still have it.

We believe that when someone wants to do repeat business with us it is the highest form of praise. Allowing your opponent in a transaction to walk away with his dignity, his humor and his hearing intact, and with a pretty good deal in his pocket, is the right way to do business.

Jerry and I learned this from our first business partner, V.P. Baker, or "Bake," as he preferred to be called, more than 20 years ago. We met when Bake was already 89 years old, with a career behind him that included being a WWI fighter pilot, a wildcat oilman, a borax prospector, a mule dentist, an orange rancher and a real estate developer. He was a wonderfully principled man. We keep a portrait of him in our conference room to remind us how to behave.

The article then lists eight principles to live by.

Boomers vs Everyone vs University Rent Extraction

Many years ago I bought a copy of The Great Boom Ahead by Harry S. Dent Jr. It was published in 1993, and described his predictions for the stock market, inflation, economic growth, and innovation for the upcoming decades. For various reasons, I read only a little of it and put the book aside. I recently picked it up again. Though I have not yet finished it, it has brought to mind some of his ideas about demographics-driven economics that have become relevant in current social media discussion.

One current resentment, one that has the purpose of supporting the call for Federal student loan forgiveness,  says that Baby Boomers had it easy. Although it's questionable whether a generation-wide ad hominem attack is a valid or non-rude debate tactic, it's worth analyzing the idea. In a recent Twitter post one pundit compared inflation-adjusted wages to non-adjusted tuition and housing prices, marking the comparison by saying "Boomer: But why can't the slackers pay for college & pay off their loans like we did?" The stark ratio was falsified by using non-comparable numbers.

If you look at wages, it is clear that current wages (for production and non-supervisory workers) are far higher than they have been for all prior generations (from FRED[1]):

Looked at this way, the current generation is doing extremely well. But this is not the data that was presented. Instead, the pundit used numbers from this FRED graph[2]:

These are inflation-adjusted ("real") wages. Notice the giant dip in the middle? The dip is centered on the Baby Boomer earnings life cycle. So, contrary to current rhetoric, because of their large numbers, when boomers entered into the labor market from 1967 to 1986 they depressed their own real wages. It was only after the labor market digested the new workers that real wages went back up.

This turns out to be highly aligned with Harry Dent's message in The Great Boom Ahead. He forecast a large drop in inflation because of competition in the labor market, which occurred. He forecast a booming economy in the 1990s, which occurred. His predictions don't align as well later on, but this post is not about a review of his book or predictions. I'm referring to the explanatory power of demographics. 

Of course, that is not the whole story. The oil price shock of 1973 caused a long term decline in wages for all production workers. But it was made worse in the 1990s and after, when U.S. workers were forced to compete not only with each other, but with overseas workers in Asia. Though avoiding the macroeconomic mistake of mercantilism (protectionism), the U.S. imposed the cost of its free trade policies on its own production workers.

Of course, in the past 25 years permissive importing policies have come from the "elite", composed primarily of Baby Boomers in managerial, professional, and government positions. So we have a mixed message: though "leaders" are indeed responsible for forcing US workers to compete wage-wise with the Third World (causing low wages), it's not like they skated along in the 1970s and early 1980s. It was rough for them early on too.

Recent Real Wage Gains

It's clear from the rising slope of the CPI-adjusted wage graph that our "story" about labor competition from offshoring of production is questionable. If real wages are depressed because of imports, then why are they actually not depressed, but rising? Unlike the wage declines that Boomers experienced in their 20s and early 30s, new generations are seeing real wage gains after 2000, at least in production roles.

Once again, using the Dent clue, we should probably look at demographics. Recent generations are smaller and the product of much lower fertility rates in the U.S. There is less competition for jobs, fewer people to fill them. Labor is scarcer, so the price of labor has been rising to meet demand. This matches current news about unfilled jobs all across the U.S. economy. It also bodes well for future wages, as the labor supply will continue to be constrained by the low U.S. birth rate.

It's a Tuition Inflation Problem, Right?

So why all the consternation? Is it just a social media conflagration, the outrage machine spinning narratives that are divisive and stupid, for the sake of selling advertising? The rhetoric is, of course, centered on Federal loan forgiveness, and the political message is that tuition inflation has been too high. The populist rhetoric centers on specific categories of goods, especially university tuition and housing, that have increased much faster than the CPI.

What about non-production workers? Were they insulated from real wage declines? It is less clear, because that category is not available on FRED. The closest I could find is this non-CPI adjusted series[3] with a much more limited timescale starting in 2003:

My impression, which likely matches that of many, is that the "elite" did just fine in their professional, managerial, IT, academia, and Government jobs. This caused the "two-tier" economy we have now where the elite are bidding up prices of homes while factory workers struggle to find affordable housing of any sort.

One could think of a college education like a membrane separating the haves from the have-nots. Many high school students and their parents may think that, in which case college tuition has very low price elasticity--any price is worth paying to avoid dropping into the gutter. The moral question (not the kind I usually address here) is whether universities are behaving rationally by charging what the market will bear, or extorting the future earnings of their customers. An ethical university would raise tuition only at the rate of CPI, and keep it affordable. Or, turning the question around, the ethics of the university are revealed by what it chooses to do. If the mission of the university is learning, then it will emphasize student welfare and tuition that matches the CPI. If the mission is profit, it will charge what the market will bear. 

Now we have a result worthy of Vorpal Trade: we have derived the internal goals of the market actor from their economic behavior.

References:

1. Average Hourly Earnings of Production and Nonsupervisory Employees, Total Private (AHETPI)

2. Average Weekly Earnings of Production and Nonsupervisory Employees, Total Private/(Consumer Price Index for All Urban Wage Earners and Clerical Workers: All Items in U.S. City Average/100)

3. Employment Cost Index: Wages and salaries for Private industry workers in Professional, scientific, and technical services (CIS2025400000000I)

MarketVector's MVMORT: A Revealing MREIT Index

You can't think of everything, but sometimes when you finally do find the answer, you feel stupid for not having thought of it earlier. That certainly applies right now to my understanding of long term Mortgage REIT performance. Of course, not being able to think of everything is exactly what the nature of this business is about, so I'm not too discouraged.

Last week I posted my analysis of the MREIT business. Everything I said there is still good, but if I had found this MREIT index sooner then I would have invested less in MREITs, or perhaps invested nothing.

MarketVector's US Mortgage REITs Index

I remember many years ago standing in a book shop in Santa Monica on Wilshire Boulevard reading a book about stock investing. It mentioned the Ibbotson and Sinquefield paper "Stocks, Bonds, Bills, and Inflation: Year-by-Year Historical Returns (1926-1974)", and it greatly impressed me that the returns were so high even though they encompassed the Great Depression and the time period started in the middle of the Roaring 20s. Having this big picture view of the market immediately put a foundation under the idea that stock investing could be not just lucrative, but inevitably so, regardless of short term fluctuations. You are reading part of the fruit resulting from that information.

With this background, you might think that each time I approach a new investment class that I might seek out data on the long term performance of the group as a whole. I will now. I did not do so with MREITs until earlier this week when, on a whim, I wondered if anyone compiled an index or benchmark on the MREITs. Of course they exist, and though there may be others, I'll focus briefly on the MarketVector MVMORT index here.

The index contains 26 U.S. mortgage-based REITs. The usual suspects are here: NLY, STWD, AGNC, BXMT, RITM (formerly NRZ), ABR, TWO, CIM, PMT. You can see from the components page that these are indeed the MREITs we are looking for.

Obviously, there are at least two components to equity return, capital appreciation and dividends. MREITs tend to have poor capital appreciation but excellent dividends. The dividends are, in fact, too good to be true, as is reflected by the terrible performance of MREITs in capital appreciation, as we can see from the Price performance shown in the MVMORT index, which as of 8/26/22 was -80.97% since the inception of the index on 12/30/2004. If you add in dividends, then gross return rises to 77.11%, but remember this is a total return across 17.66 years. Annualized you got 3.29%. The net return (after taxes or withholding, I presume) was -11.91% or -0.72% annualized. 

From the graphs of the indices it's clear that although the 2007-2009 mortgage crisis was a drag on MBS and MREIT returns, the 2020 COVID shock was much worse. You might be charitable and decide that these two events were extraordinary and therefore the index is not representative of the prospects of MREITs. Nevertheless, the index covers 18 years of time, a significant sampling, even if not as compelling as the SBBI period of 1926-1974.

I am not as inclined to be charitable, and see MVMORT as a fair representation of the prospects of the MREIT sector, which I think was fairly characterized by the analysis I posted here on August 16th. More informally, I think of the U.S. mortgage as a strange and amazing artifact that tends to benefit the borrower most. It is at a low interest rate, tax deductible, and allows you to make a highly leveraged investment on a practical and useful capital asset that enjoys special legal protections. The U.S. mortgage is so good a device (for the borrower) that it leads to, and has often led to, overinvestment in homebuilding, leading to excessive real estate prices, homes that are too large, and stranding of capital. By providing special tax advantages for mortgages, the U.S. Government has increased the volatility of the housing sector, which in turn has increased the frequency and severity of financial crises based on mortgages. In that sense, U.S. tax policy caused the 2008 financial crisis.

One might think of MREITs as specialized trading vehicles, to be used like equity options, index options, leveraged or short index closed end funds. If you treat it as toxic waste with zero-sum characteristics, in which staying too long with an MREIT is an indicator of mental illness, then you might have the right approach to dealing with MREITs. Keep in mind, though, that the same quants (or AI-based SAIMS traders) who have mastered MBS will also have the inside track on MREITs.

A Modest Essay on Mortgage REITs

I recently mentioned the Mortgage Real Estate Investment Trust (MREIT) industry while briefly commenting on high NAV-ror companies in the BDC universe. Recently I looked more deeply into several MREITs, and find that they are too complex to be worthwhile. I will pick on Annaly (NLY) primarily because they are the largest market cap MREIT that is publicly traded.

The primary source that convinced me was the Annaly 22Q2 earnings call [1]. You can find a transcript here (NLY earnings call transcript at Motley Fool). This passage in David Finkelstein's prepared remarks especially stood out:

After combining our book value performance at our first quarter dividend of $0.22, our quarterly economic return was negative 9.6%. As noted earlier, the portfolio generated EAD per share of $0.30. Earnings continued to be strong, resulting from high dollar roll income, increasing MSR net servicing income, reduced amortization due to lower CPRs and a benefit from our swaps portfolio as it turned to a net receipt position during the quarter on higher short-term rates. However, EAD was meaningfully aided this quarter by an increase in specialness and dollar rolls, primarily driven by a scarcity of TBA-type collateral and newer higher rate production coupons.

For those of you in the MBS (mortgage backed securities) business, you can stop reading here. This article will be too simplistic. The rest of us need a quick review of what this paragraph implies. 

First, although you can Google the meanings of the specialized terms and acronyms and eventually get a mental picture of what they mean (EAD = exposure at default, MSR = mortgage servicing rights, CPR = conditional prepayment rate, CRT = credit risk transfer [2], TBA = to be announced, specialness = the extent to which implied dollar roll financing rates fall below prevailing market rates, dollar roll = to sell short MBS) as a whole the complexity implied by the jargon is irreducible. Some of these terms originate in the nature of the mortgage contracts. Others refer to typical practices in the MBS trading system. Unless you want to specialize in understanding MBS, it would be difficult to understand how to sort out one MREIT from another based on their business model. 

Could you analyze management, probity, company values? Perhaps, but your analysis would be thin and fragile, with little to count on in difficult times. You would have to be confident that a company with better values could outperform companies without good values, in a commodity market. Thinking back to comments made by Ajit Jain and Warren Buffett about the reinsurance and property and casualty markets, when an commodity business gets crowded, margins thin and often the companies with the thinnest value systems win on price.

The MBS world, although not simple, is a crowded market, as alluded to here by NLY:

This correction will be welcomed for the Agency MBS market as it reduces elevated net supply, which has been the main headwind for the sector in the recent past. 

Although there are many variations in the mortgage market (geographic, % of value, borrower demographics, contractual terms, MSR details, etc.) the primary business of buying and selling mortgage securities is liquid and competitive. The originator may be able to distinguish themselves by service, bundling, and so on, but secondary players like MREITs have little to no moat around their operations.

Monster in the Market: Super Artificial Intelligence

This leads me to speculate about the potential for a very serious difficulty for MREITs, which I will call the Super AI MBS Speculator (SAIMS). The SAIMS entity was constructed sometime around 2013 to 2017, although it is continually upgraded by the MIT quant engineers who developed it for the secretive hedge fund they work for. The goal: to extract value from the MBS market by out-thinking the rest of the MBS players and detecting trends faster than they do. Using deep learning augmented by specialized programming, the SAIMS continually analyzes interest rates, economic indicators, housing markets, consumer behavior, but especially the MBS markets and its players. It does not seek to dominate any one type of trade, but instead seeks to systematically shave an advantage off every type of trade available to it. 

SAIMS is similar to the quant trading done by HFT players, but it seeks advantages in information interpretation more than it does in speed. Because it is automated, it can place trades for small amounts that are statistically consistently profitable without tipping its hand that it has an edge over its competitors. 

In a SAIMS-dominated environment, the players like NLY are continually using very, very sophisticated methods of analysis that show them what just happened and why even though they are not able to anticipate the moves or tactics of the SAIMS. The best MREIT companies at their best are still several steps behind the SAIMS. Although the MREITs may make small amounts of money above trend in quiescent markets, when the SAIMS is paired with an active human-based trading strategy that generates disruptions in the market it can force the market into non-quiescent periods. Naturally, Government actions (or inactions, as when they fail to tighten rates in response to price movements) tend to generate market disruptions, so the SAIMS owners are close friends with Government organizations and politicians.

SAIMS works because it trades against the entire MBS and MREIT marketplace. It goes short and long, offers excellent liquidity and supplies narrow spreads to the market. It takes small amounts of money from all players, but makes up for the small amounts by dealing in volume. To the outside observer, SAIMS's success is mysterious and difficult to explain. Huge profits show up annually on the ledger of the investment bank/hedge fund, but rarely is much attention devoted to explaining the operation to outsiders.

While SAIMS is a speculation and not a prediction, I think that regardless of the actual shape of the risk to MREITs, it is highly useful to think of the MREIT environment as including SAIMS-like opponents. It could be the marketplace itself, a collection of SAIMS programs, or the machinations of a Federal Government that continually collects benefits from the gradual book value loss of the MREITs, even if those "machinations" are not realized in any single mind and are merely the product of bureaucracy plus legislation. If you want a 10% to 12% yield, you need to seriously consider why such a juicy yield is available in the market. Is it a Trojan Horse?

A Low Interest Rate Base Makes All Fixed Income Vulnerable

MBS are bond-like. As I published here before, low long-term interest rates pose problems for those seeking to invest in bonds. Likewise, low interest rates are a problem for MBS and MREITs. 

They are also a problem for target date mutual funds, and for bond funds generally. And this time may be different (for bonds). This blog post at Allocate Smartly[3] says it well:

Importantly, an extended period of rising interest rates could have a more severe impact in today’s market than in the 1960’s and 70’s. Yields rose from a higher base then, and that helped to cushion the impact of rising rates.

Toxic Complexity

While digging through the MREIT acronyms and jargon, I kept encountering ideas that were explained primarily by academic papers. The first, "specialness", with an irony almost too profound to be believed, turns up in a paper you can get via an mit.edu URL [5]. Though maybe I didn't look hard enough, the best explanation for PAA I could find was in another academic paper [4] on obtaining higher yields with near-zero risk of capital loss.

Some years ago while at Hostfest in Minot, SD I met the aunt of a fellow MIT graduate, who put us in touch with one another. We corresponded by email. He mentioned that he was in the MBS business.

I'm in favor of using our intelligence to solve problems and make things better. But it is a problem when simply keeping up with the competition requires either being an MIT graduate or doing your homework just as well as one. If you are in a simple commodity business, then superior teamwork, processes, and a solid corporate culture could make you the low cost provider, and therefore make your business both profitable and stable. In the MBS business, it appears that external players can be leaders just as easily as the publicly-traded MREITs, and for that reason, I'm out.

Sources

1. NLY earnings call transcript at Motley Fool

2. PMT earnings call transcript at Motley Fool

3. https://allocatesmartly.com/government-bonds-have-failed-to-deliver-when-needed/

4. Keller, Wouter J. and Keuning, Jan Willem, Protective Asset Allocation (PAA): A Simple Momentum-Based Alternative for Term Deposits (April 5, 2016). Available at SSRN: https://ssrn.com/abstract=2759734 or http://dx.doi.org/10.2139/ssrn.2759734

5. Song, Zhaogang and Zhu, Haoxiang, Mortgage Dollar Roll (July 2, 2018). Review of Financial Studies (Forthcoming), Available at SSRN: https://ssrn.com/abstract=2401319 or http://dx.doi.org/10.2139/ssrn.2401319