Category Archives: science, technology and society

using a model to explain a single case

Leave a reply

Charles Sanders Peirce introduced the logic of what he called “abduction” — a complement to both deduction and induction — with this example:

The surprising fact, C, is observed;
But if A were true, C would be a matter of course,
Hence, there is reason to suspect that A is true.

At least since Harry Frankfurt in 1958, many readers have been skeptical. Can’t we make up an infinite number of premises that could explain any surprising fact?

For instance, Kamala Harris has gained in the polls compared to Joe Biden. If it were true that voters generally prefer female presidential candidates, then her rise would be a “matter of course.” But it is a mistake to infer that Harris has gained because she is a woman. Other explanations are possible and, indeed, more plausible.

Note that “voters prefer women candidates” is an empirical generalization. Generalizations cannot be derived from any single case. If that is what abduction means, then it seems shaky. Its only role might be to suggest hypotheses that should then be tested with representative samples or controlled experiments.

But what if A (the premise) is not an empirical generalization but rather a model? For instance, a model might posit that Harris’ current position in the polls is the combined result of eight different factors, some of them general (voters usually follow partisan cues) and some of them quite unrepeatable (the incumbent president has suddenly bowed out).

Positing a model to explain a single case has risks of its own. Perhaps we add no insight by contriving an elaborate model just to fit the observed reality. And we might be tempted to treat the various components of the model as general patterns and apply them elsewhere, even though one case should give us no basis for generalizing.

But let’s look at this example from a different perspective–a pragmatic one, as Peirce would recommend. After all, Peirce calls his topic “Abductive Judgment” (Peirce 1903), suggesting a connection to practical reason or phronesis.

The question is what should (someone) do? For instance, a month ago, should Joe Biden have dropped out and endorsed Harris? Right now, should Harris accentuate her gender or try to balance it with a male vice-presidential candidate?

Inductive logic might offer some insights. Research suggests that the choice of vice-president has never affected the outcome of a presidential election, and this general inference would suggest that Harris needn’t pay attention to the gender of her VP. But induction cannot answer other key questions, such as what to do when you replace the nominee 100 days before the election. (There is no data on this matter because it hasn’t happened before.)

Besides, various factors can interrelate. The general pattern that vice-presidents do not matter might be reversed in a situation where the nominee had herself been the second person on the ticket until last week.

And the important questions are inescapably normative. For Harris, one good goal is to win the election, but she must attend to other values as well. For instance, I think she should adopt positions that would benefit working-class voters of all races. Possibly this would help her win by restoring some of Biden’s working-class coalition from 2020. Polling data would help us assess that claim. But I favor a worker-oriented strategy for reasons of justice, and I think the important question is how (not whether) to campaign that way.

Models of social phenomena typically incorporate descriptive elements (Harris is down by two points today), causal claims (Trump is still benefitting from a minor convention bump), and normative premises (Harris must win)–all combined for the purpose of guiding action.

Arguably, we cannot do better than abduction when we are trying to decide what to do next. Beginning with a surprising fact, C (and almost anything can be seen as “surprising”), we must come up with something, A, that we can rely on to guide our next steps. A should not be a single sentence, but rather a model composed of various elements.

It is worthwhile to consider evidence from other cases that may validate or challenge components of A. But it is not possible to prove or disprove A. As the pioneering statistician Georg Rasch said, “Models should not be true, but it is important that they are applicable, and whether they are applicable for any given purpose must of course be investigated. This also means that a model is never accepted finally, only on trial.”

If a model cannot be true, why should we make it explicit? It lays out what we are assuming so that we can test the assumptions as we act. It promotes learning from error. And it can help us to hold decision-makers accountable. When evaluating leaders, we should not assess the outcomes, which are beyond anyone’s control, but rather the quality of their models and their ability to adjust in in the light of new experience.

Sources: Peirce, C.S. 1903. Lectures on Pragmatism, Lecture 1: Pragmatism: The Normative Sciences; Frankfurt, Harry G. “Peirce’s notion of abduction.” The Journal of Philosophy 55.14 (1958): 593-597. See also: choosing models that illuminate issues–on the logic of abduction in the social sciences and policy; modeling social reality; different kinds of social models

analytic and holistic reasoning about social questions

Leave a reply

“President Biden’s student loan cancellations will bring relief.” “Retrospectively forgiving loans creates a moral hazard.” “At this college, students study the liberal arts.” “An unexamined life is not worth living for humans.”

These claims are, respectively, about a specific act (a policy announced yesterday), a pattern that applies across many cases, an assessment of an institution, and a universal principle.

These statements may be related. An ambitious defense of Biden’s decision to forgive student loans might connect that act to liberal education and thence to a good life, whereas a critique might tie the loan cancelation to cost increases. A good model of a social issue or question often combines several such components.

In this post, I will contrast two ways of thinking about models and their components.

  1. Analytic reasoning

Analytic reasoning seeks characteristics that apply across cases. We can define government aid, moral hazard, and education, defend these definitions against alternatives, and then expect them generally to have the same significance wherever they apply. For example, becoming more self-aware is desirable, all else being equal or as far as that goes (ceteris paribus or pro tanto). We need a definition of self-awareness that allows us to understand what tends to produce it and what good it does. The same goes for loans, loan-forgiveness, and so on.

Methods like controlled field experiments and regression models require analysis, and they demonstrate that it has value. Ethical arguments that depend on sharply defined universals are quintessentially analytic. Qualitative research is often quite analytic, too, particularly when either the researcher or the research subjects employ general concepts.

Analytic reasoning offers the promise of generalizable solutions to social problems. For instance, let’s say you believe that we should spend more money on schools in poor communities. In that case, you are thinking analytically: you view money as an identifiable factor with predictable impact. Note that you might advocate increasing the national education budget while also being sensitive to local differences about things other than money.

  1. Holistic reasoning

Holistic reasoning need not been any less rigorous, precise, or tough-minded than analytic reasoning, but it works with different objects: whole things. For example, we can describe a college as an entity. To do that requires saying many specific things about the institution, but each claim is not meant to generalize to other places.

At a given (imaginary) institution, the interplay between a rural setting, an affluent student body, an applied-science curriculum, a modest endowment, and a recent crisis of leadership could produce unexpected results, and those are only some of the factors that would explain the particular ethos that emerges at that college.

Holistic reasoning is wise if each factor is closely related to others in its specific context. For instance, often a statistic about a place is the result of decisions about what and how to measure, which (in turn) depend on who’s in charge and what incentives they face, which depends on prior political decisions, and so on–indefinitely. From a holistic perspective, a statistic lacks meaning except in conjunction with many other facts.

There is a link here to holistic theories of meaning and/or language, e.g., Robert Brandom: “one cannot have any concepts unless one has many concepts. For the content of each concept is articulated by its inferential relations to other concepts. Concepts, then, must come in packages” (Brandom 2000).

Holistic reasoning is also wise if values change their significance depending on the context, a view labeled as “holism” in metaethics. We are familiar with the idea that lying is bad–except in circumstances when it is good, or even courageous and necessary. An ethical holist believes that there are good and bad value-judgments, but they are not about abstract categories (such as lying). They are about wholes.

Finally, holistic reasoning is wise if we can gain insights about meaning that would be lost in analysis. In Clifford Geertz’ classic interpretation of a Balinese cockfight (Geertz 1972), he successively describes that phenomenon as “a chicken hacking another mindless to bits” (p. 84); “deep play” (p. 71), or an activity that has intrinsic interest for those involved; “fundamentally a dramatization of status concerns” (p. 74); an “encompassing structure” that presents a coherent vision of “death, masculinity, rage, pride, loss, beneficence, chance” (p. 79); and “a kind of sentimental education” from which a Balinese man “learns what his culture’s ethos and his private sensibility (or, anyway, certain aspects of them) look like when spelled out externally in a collective text” (p. 83).

Geertz offers lots of specific empirical data and uses concepts that would apply across cases, including terms like “chickens” and “betting” that arise globally. However, he is not primarily interested in what causes cockfights in Bali or what they cause. His main question is: What is this thing? Since Balinese cockfighting is a human activity, what it is is what it means. And it has meaning as a whole thing, not as a collection of parts.

Conclusion

This discussion may suggest that holistic reasoning is more sensitive and thoughtful than analytic reasoning. But recall that ambitious social reform proposals depend on analytic claims. If everything is contextual, then there is no basis for changing policies or priorities that apply across cases. Holistic reasoning may be conservative, in a Burkean sense–for better or for worse.

Then again, a “whole” need not be something small and local, like a cockfight in Bali or a college in the USA. A nation-state can also be analyzed and interpreted holistically and changed as a result.

It is trite to say that we need both analytic and holistic reasoning about policy, but we do. Instead of jumping to that conclusion, I’ve tried to draw a contrast that suggests some real disadvantages of each.

References: Brandom, Robert B. Articulating reasons: An introduction to inferentialism. Harvard University Press, 2001; Geertz, Clifford. “Deep play: Notes on the Balinese cockfight.” Daedalus 134.4 (2005): 56-86. See also: against methodological individualism; applied ethics need not mean applying ethical systems; what must we believe?, modeling social reality; choosing models that illuminate issues–on the logic of abduction in the social sciences and policy; choosing models that illuminate issues–on the logic of abduction in the social sciences and policy; different kinds of social models etc.

the age of cybernetics

Leave a reply

A pivotal period in the development of our current world was the first decade after WWII. Much happened then, including the first great wave of decolonization and the solidification of democratic welfare states in Europe, but I’m especially interested in the intellectual and technological developments that bore the (now obsolete) label of “cybernetics.”

I’ve been influenced by reading Francisco Varela, Evan Thompson, and Eleanor Rosch, The Embodied Mind: Cognitive Science and Human Experience (first ed. 1991, revised ed., 2017), but I’d tell the story in a somewhat different way.

The War itself saw the rapid development of entities that seemed analogous to human brains. Those included the first computers, radar, and mechanisms for directing artillery. They also included extremely complex organizations for manufacturing and deploying arms and materiel. Accompanying these pragmatic breakthroughs were successful new techniques for modeling complex processes mathematically, plus intellectual innovations such as artificial neurons (McCullouch & Pitts 1943), feedback (Rosenblueth, Wiener, and Bigelow 1943), game theory (von Neumann & Morgenstern, 1944), stored-program computers (Turing 1946), information theory (Shannon 1948), systems engineering (Bell Labs, 1940s), and related work in economic theory (e.g., Schumpeter 1942) and anthropology (Mead 1942).

Perhaps these developments were overshadowed by nuclear physics and the Bomb, but even the Manhattan Project was a massive application of systems engineering. Concepts, people, money, minerals, and energy were organized for a common task.

After the War, some of the contributors recognized that these developments were related. The Macy Conferences, held regularly from 1942-1960, drew a Who’s Who of scientists, clinicians, philosophers, and social scientists. The topics of the first post-War Macy Conference (March 1946) included “Self-regulating and teleological mechanisms,” “Simulated neural networks emulating the calculus of propositional logic,” “Anthropology and how computers might learn how to learn,” “Object perception’s feedback mechanisms,” and “Deriving ethics from science.” Participants demonstrated notably diverse intellectual interests and orientations. For example, both Margaret Mead (a qualitative and socially critical anthropologist) and Norbert Wiener (a mathematician) were influential.

Wiener (who had graduated from Tufts in 1909 at age 14) argued that the central issue could be labeled “cybernetics” (Wiener & Rosenblueth 1947). He and his colleagues derived this term from the ancient Greek word for the person who steers a boat. For Wiener, the basic question was how any person, another animal, a machine, or a society attempts to direct itself while receiving feedback.

According to Varela, Thompson, and Rosch, the ferment and diversity of the first wave of cybernetics was lost when a single model became temporarily dominant. This was the idea of the von Neumann machine:

Such a machine stores data that may symbolize something about the world. Human beings write elaborate and intentional instructions (software) for how those data will be changed (computation) in response to new input. There is an input device, such as a punchcard reader or keyboard, and an output mechanism, such as a screen or printer. You type something, the processor computes, and out comes a result.

One can imagine human beings, other animals, and large organizations working like von Neumann machines. For instance, we get input from vision, we store memories, we reason about what we experience, and we say and do things as a result. But there is no evident connection between this architecture and the design of the actual human brain. (Where in our head is all that complicated software stored?) Besides, computers designed in this way made disappointing progress on artificial intelligence between 1945 and 1970. The 1968 movie 2001: A Space Odyssey envisioned a computer with a human personality by the turn of our century, but real technology has lagged far behind that.

The term “cybernetics” had named a truly interdisciplinary field. After about 1956, the word faded as the intellectual community split into separate disciplines, including computer science.

This was also the period when behaviorism was dominant in psychology (presuming that all we do is to act in ways that independent observers can see–there is nothing meaningful “inside” us). It was perhaps the peak of what James C. Scott calls “high modernism” (the idea that a state can accurately see and reorganize the whole society). And it was the heyday of “pluralism” in political science (which assumes that each group that is part of a polity automatically pursues its own interests). All of these movements have a certain kinship with the von Neumann architecture.

An alternative was already considered in the era of cybernetics: emergence from networks. Instead of designing a complex system to follow instructions, one can connect numerous simple components into a network and give them simple rules for changing their connections in respond to feedback. The dramatic changes in our digital world since ca. 1980 have used this approach rather than any central design, and now the analogy of machine intelligence to neural networks is dominant. Emergent order can operate at several levels at once; for example, we can envision individuals whose brains are neural networks connecting via electronic networks (such as the Internet) to form social networks and culture.

I have sketched this history–briefly and unreliably, because it’s not my expertise–without intending value-judgments. I am not sure to what extent these developments have been beneficial or destructive. But it seems important to understand where we’ve come from to know where we should go from here.

See also: growing up with computers; ideologies and complex systems; The truth in Hayek; the progress of science; the human coordination involved in AI; the difference between human and artificial intelligence: relationships

the difference between human and artificial intelligence: relationships

Leave a reply

A large-language model (LLM) like ChatGPT works by identifying trends and patterns in huge bodies of text previously generated by human beings.

For instance, we are currently staying in Cornwall. If I ask ChatGPT what I should see around here, it suggests St Ives, Land’s End, St Michael’s Mount, and seven other highlights. It derives these ideas from frequent mentions in relevant texts. The phrases “Cornwall,” “recommended” (or synonyms thereof), “St Ives,” “charming,” “art scene,” and “cobbled streets” probably occur frequently in close proximity, because ChatGPT uses them to construct a sentence for my edification.

We human beings behave in a somewhat similar way. We also listen to or read a lot of human-generated text, look for trends and patterns in it, and repeat what we glean. But if that is what it means to think, then LLM has clear advantages over us. A computer can scan much more language than we can and uses statistics rigorously. Our generalizations suffer from notorious biases. We are more likely to recall ideas we have seen most recently, those that are most upsetting, those that confirm our prior assumptions, etc. Therefore, we have been using artificial means to improve our statistical inferences ever since we started recording possessions and tallying them by category thousands of years ago.

But we also think in other ways. Specifically, as intensely social and judgmental primates, we frequently scan our environments for fellow human beings whom we can trust in specific domains. A lot of what we believe comes from what a relatively small number of trusted sources have simply told us.

In fact, to choose what to see in Cornwall, I looked at the recommendations in The Lonely Planet and Rough Guide. I have come to trust those sources over the years–not for every kind of guidance (they are not deeply scholarly), but for suggestions about what to see and how to get to those places. Indeed, both publications offer lists of Cornwall’s highlights that resemble ChatGPT’s.

How did these publishers obtain their knowledge? First, they hired individuals whom they trusted to write about specific places. These authors had relevant bodily experience. They knew what it feels like to walk along a cliff in Cornwall. That kind of knowledge is impossible for a computer. But these authors didn’t randomly walk around the county, recording their level of enjoyment and reporting the places with the highest scores. Even if they had done that, the sites they would have enjoyed most would have been the ones that they had previously learned to understand and value. They were qualified as authors because they had learned from other people: artists, writers, and local informants on the ground. Thus, by reading their lists of recommendations, I gain the benefit of a chain of interpersonal relationships: trusted individuals who have shared specific advice with other individuals, ending with the guidebook authors whom I have chosen to consult.

In our first two decades of life, we manage to learn enough that we can go from not being able to speak at all to writing books about Cornwall or helping to build LLMs. Notably, we do not accomplish all this learning by storing billions of words in our memories so that we can analyze the corpus for patterns. Rather, we have specific teachers, living or dead.

This method for learning and thinking has drawbacks. For instance, consider the world’s five biggest religions. You probably think that either four or five of them are wrong about some of their core beliefs, which means that you see many billions of human beings as misguided about some ideas that they would call very important. Explaining why they are wrong, from an outsider’s perspective, you might cite their mistaken faith in a few deeply trusted sources. In your opinion, they would be better off not trusting their scriptures, clergy, or people like parents who told them what to believe.

(Or perhaps you think that everyone sees the same truth in their own way. That’s a benign attitude and perhaps the best one to hold, but it’s incompatible with what billions of people think about the status of their own beliefs.)

Our tendency to believe select people may be an excellent characteristic, since the meaning of life is more about caring for specific other humans than obtaining accurate information. But we do benefit from knowing truths, and our reliance on fallible human sources is a source of error. However, LLMs can’t fully avoid that problem because they use text generated by people who have interests and commitments.

If I ask ChatGPT “Who is Jesus Christ?” I get a response that draws exclusively from normative Christianity but hedges it with this opening language: “Jesus Christ is a central figure in Christianity. He is believed to be … According to Christian belief. …” I suspect that ChatGPT’s answers about religious topics have been hard-coded to include this kind of disclaimer and to exclude skeptical views. Otherwise, a statistical analysis of text about Jesus might present the Christian view as true or else incorporate frequent critiques of Christianity, either of which would offend some readers.

In contrast, my query about Cornwall yields confident and unchallenged assessments, starting with this: “Cornwall is a beautiful region located in southwestern England, known for its stunning coastline, picturesque villages, and rich cultural heritage.” This result could be prefaced with a disclaimer, e.g., “According to many English people and Anglophiles who choose to write about the region, Cornwall is …:” A ChatGPT result is always a summary of what a biased sample of people have thought, because choosing to write about something makes you unusual.

For human beings who want to learn the truth, having new tools that are especially good at scanning large bodies of text for statistical patterns should prove useful. (Those who benefit will probably include people who have selfish or even downright malicious goals.) But we have already learned a fantastic amount without LLMs. The secret of our success is that our brains have always been networked, even when we have lived in small groups of hunter-gatherers. We intentionally pass ideas to other people and are often pretty good at deciding whom to believe about what.

Moreover, we have invented incredibly complex and powerful techniques for improving how many brains are connected. Posing a question to someone you know is helpful, but attending a school, reading an assigned book, finding other books in the library, reading books translated from other languages, reading books that summarize previous books, reading those summaries on your phone–these and many other techniques dramatically extend our reach. Prices send signals about supply and demand; peer-review favors more reliable findings; judicial decisions allow precedents to accumulate; scientific instruments extend our senses. These are not natural phenomena; we have invented them.

Seen in that context, LLMs are the latest in a long line of inventions that help human beings share what they know with each other, both for better and for worse.

See also: the design choice to make ChatGPT sound like a human; artificial intelligence and problems of collective action; how intuitions relate to reasons: a social approach; the progress of science.

artificial intelligence and problems of collective action

Leave a reply

Although I have not studied the serious scholarship on AI, I often see grandiose claims made about its impact in the near future. Intelligent machines will solve our deepest problems, such as poverty and climate change, or they will put us all out of work and become our robot overlords. I wonder whether these predictions ignore the problems of collective action that already bedevil us as human beings.

After all, there are already about 7.5 billion human brains on earth, about 10 times more than there were in 1800. Arguably, we are better off than we were then–but not clearly and straightforwardly so. If we ask why a tenfold increase in the total cognitive capacity of the species has not improved our condition enormously, the explanations are pretty obvious.

Even when people agree on goals, it is challenging to coordinate their behavior so that they pursue those ends efficiently. And even when some people manage to work together toward a shared goal, they have physical needs and limitations. (Using brains requires food and water; implementing any brain’s ideas by taking physical action requires additional resources.) To make matters worse, human beings often have legitimate but conflicting interests, like the need to gain sustenance from the same land. And some human beings have downright harmful goals, like dominating or spiting others.

One can see how artificial intelligence might mitigate some of these drawbacks. Imagine a single computer with computational power equivalent to one million human beings. It will be much more coordinated than those people. It will be able to aggregate and apply information more efficiently. It can also be programmed to have consistent and, indeed, desirable goals–and it will plug away at its goals for as long as it receives the physical inputs it needs. For instance, it could clean up pollution 24/7 instead of stopping for self-interested purposes, like sleeping.

However, it still has physical needs and limitations. It might use fuel and other inputs more efficiently than a human being does, but that depends on how good the human’s tools are. A person with a bulldozer can move more garbage than a clever little robot that works 24/7–and both of them need a place to put the garbage. (Intelligence cannot negate physical limits.)

Besides, a computer is designed by people–and probably by individuals arrayed as corporations or states. As such, AI is likely to be designed for conflicting and sometimes discreditable goals, including killing other people. At best, it will be hard to coordinate the activities of many different artificially intelligent systems.

Meanwhile, people already coordinate their behavior in quite impressive ways. A city receives roughly the amount of bread it needs every day because thousands of producers and vendors coordinate their behavior through prices. An international scientific discipline makes cumulative progress because thousands of scientists coordinate their behavior through peer-review and citation networks. And the English language develops new vocabulary for describing new phenomena as millions of people communicate. Thus the coordination attained by a machine with a lot of computational power should be compared to the coordination accomplished by human beings in a market, a discipline, or a language–which is impressive.

One claim made about AI is that machines will start to refine and improve their own hardware and software, thus achieving geometric growth in computational power. But human beings already do this. Although we cannot substantially redesign our individual brains, we can individually learn. More than that, we can redesign our systems for coordinating cognition. Many people are busy making markets, disciplines, languages, and other emergent human systems work better. That is already the kind of continuous self-engineering that some people expect AI to accomplish for the first time.

It is of course possible to imagine that an incredibly intelligent machine will identify solutions that simply elude us as human beings. For instance, it will negate the physical limitations of the carbon cycle by discovering whole new processes. But that is an empty supposition, like imagining that regular old science will one day discover solutions that we cannot envision today. That is probably true–it has happened many times before–but it is unhelpful in the present. Besides, both people and AI may create more problems than they solve.

See also: the progress of science; John Searle explains why computers will not become our overlords;