Newcomb's Paradox, The Cynics, & The Hungry Crowd / by Bryce Hidysmith

 Photo by T. Sullivan, San Francisco, March, 2018

Photo by T. Sullivan, San Francisco, March, 2018

< Soundtrack: Just keep reading, but also Albinoni - Adagio in G Minor and Zola Jesus/JG Thirlwell/Mivos Quartet - Hikikomori >

1. 

I'm on an overnight ferry crossing the Baltic, trying to grapple with Robert Nozick's Newcomb's Problem and Two Principles of Choice. I highly suggest reading Nozick’s original article, and there are plenty of available summaries of Newcomb’s Problem and commentaries on its implications—a recent favorite of mine is Scott Aaronson’s remarks in Quantum Computing Since Democritus, especially his remarks on the student with a 50% rating on the Aaronson Oracle—so I will not bother summarizing it again here. I will focus on only the version of the problem where the predictor is omniscient unless otherwise noted. Newcomb's problem doesn't seem paradoxical, or even particularly difficult to interpret from my vantage. Of course, Nozick makes the point that everyone seems to feel this way, and that the interesting thing would be trying to figure out how to get the differing perspectives to collide. These remarks are an attempt to reconcile some of the differing intuitions about Newcomb-type problems, as well as the introduction of a new Newcomb-like problem.

Nozick suggests that you should take both boxes, but does not believe that he has answered the question conclusively in the article. He seems to have just found himself assuming that because it would be possible to defect against your past self who intended to not take the second box, you would take the second box at least some of the time. This conflates the possibility of defection with its actualization. From what little I know of Nozick, this seems to be his political philosophy at its core—an assumption of opportunistic defection and yet still trying to construct palatial systems of life. This position is entirely honest, but confused. From a cursory reading of some of his works, Nozick seems to assume correctly that intentions don't determine outcomes, but implicitly fails to notice that intentions prune the space of possible outcomes that a given intelligent agent would seek. I feel like there's probably some more explicit statements on this in the sections of Invariances and Anarchy, State, & Utopia that I didn't get to read while I was in Seattle in December, but I'll have to check later. I also feel like I do need to give credit to Nozick’s assumptions here, as he was living in a corner of the Twentieth Century Anglosphere where most people conflated defection and self-interest. Even in the vise grip of that cultural assumption he and many others were able to at least hold onto the will to reason if not always overcome immense odds and construct reasonable realities for themselves and others. To quote the poet Anderson, when love is gone, there’s always justice, and when justice is gone, there’s always force. Provided that force is at least informed by decent compassion and aesthetics, not all is lost. I have often been the sort of man who, like Nozick, believes I live in that most cynical reality. It is only by the loving care of my friends that I have been able to embrace the discipline required to transcend that assumption. 

The two examples that Nozick chooses in his article—the Vaccine choice and choice of the Brothers—are both cases of an agent's need to choose a correct strategy while knowing the range of possible realities, yet not which one they inhabit. Both are well satisfied by the dominance principle. Yet, in the case of the Newcomb Problem, the consequential reality is socially constructed by the individual who chooses or does not choose to take the second box. Thus, it seems that if one is able to interpret the principle of expected utility, one should be able to be the kind of agent that possesses the self-control to take only the million dollar box. The Predictor is of course able to predict a lack of faith or an impossible greed in the agent. The problem does not specify as to which, but such specification is unnecessary as the problem is agnostic as to the identities and relationships of the box-taker and the Predictor. Following its abstract formulation, the Newcomb Problem is framed not as a moral judgement, but a rational judgement. Such is the case of all problems concerning the acquisition of currency or other transactable types of utility. There is no reason to believe that such defects would be present in an agent who had adequately interpreted the problem. Only a fool would choose anything other than simply taking only box B. The agent's mind is transparent, and the only thing to do in that circumstance is to rise to it, becoming the kind of agent that the situation demands. All other options are inferior. If one can submit to the limits of control, one is able to earn the greatest reward available.

The narrative that somehow one should intend to take only the thousand dollar box at the moment of prediction and then, through force of free will, take both boxes is either incoherent, or simply demands further specification from the problem. One must assume that the predictor could detect the intent to defect in the chooser, even if it is latent. The predictor knows how the chooser will respond to the environment of the problem even if the chooser does not know their own response at the time of the prediction. Depending on our beliefs about the informational structure of the world, we could imagine that an agent innocent of the potential success of defection could be introduced to the idea of taking two boxes by an outside power after the moment of prediction and successfully defeat the predictor's abilities, but this relies on the predictor somehow not being able to notice that the chooser would choose to defect and take both boxes if the idea was introduced, or that the predictor would be unable to predict that the chooser would come into contact with such an outside power. Of course, if the chooser's mind is not transparent to the predictor, then one can imagine all sorts of Mamet-esque method acted cons, but while that genre of problem brings great personal joy to me, it is hardly as interesting as the philosophical implications of the omniscient variant. 

Newcomb’s problem role in scholarship is mostly a shibboleth to test decision theories and decision theorists, the former because of the limited computational blindnesses of models like Causal Decision Theory, and the latter largely because of bias or confusion. For instance, it seems like frequently the two-box solution appeals to materialist types who are extremely concerned that bringing discussions of how parts of reality might be socially constructed opens the floodgates of theoretically infinite woo. The justification for the one-box solution—namely that the reality is socially constructed by the chooser’s actions—feels insufficiently rigorous even to me as it cannot prove why the one-boxer chooses only one box, except that they are the type of person who can interpret the reality of the situation and choose the highest expected utility outcome. This is certainly not the same level of certainty that the two-boxer feels. If one attempts to evaluate the problem from the perspective of the chooser, rather than from some kind of simulated third-person perspective, one sits uneasily with individuality's inherent uncertainty. 

My attention strays easily from potential solutions to wider implication. Newcomb’s Problem is beloved by analytical philosophy for the same reasons that it is altogether unheard of and likely entirely uninterpretable by mainstream societies that coordinate either through technical or emotional protocols. The problem describes a kind of coordination that can only be accomplished if you accept that sense of self is accepted to be unstable, which is anathema to the kind of assured consistent perspective necessary for doing most technological development that doesn’t involve selection effects. Taking the broadest definition of engineering—perhaps most easily specified as bounded optimization—one assumes that one's sense of self is stable. For, if it were not stable, then one would not be able to work towards the hoped optimization criteria, and instead would spend all their resource fearful that their future self will undo their labors for arbitrary reasons. It is remarkably well specified, and thus those with a well-cultivated engineering mindset are often able to interpret and speculate on its content and implications, whereas if it was written in a less explicit manner it would likely go the way of most other attempts at non-dualistic psychology, so often invoking a holistic worldview as a justification for abandoning systemization and rigor. In studying the problem, I only became more convinced that the many insights it generates are anathema to the sorts of decision theories that actually dominate the public sphere. While I believe that the investigation of rational decision theories is of critical importance to both human psychology and artificial intelligence research—both pure and applied—I am confused by the general failure of both analytical philosophy and artificial intelligence to engage with the variance of decision theories that have evolved in history. Certainly, all presently specified decision theories are flawed or at least incomplete, but clearly so is modern decision theory. It has always seemed to me that the necessary strategy for the application of decision theory to the real world was incomplete if it was simply a priori analysis, frequently committed only by those who were able to engage with it as a recreational practice. The empirical comparison of existent decision-making behaviors has always felt like the missing piece of any of the gains from the purely theoretical domain were to have any practical use. Indeed, as we have seen with the evolution of almost every branch of cognitive science, the historical constraints of theorists often lead them with a deep myopia, perhaps best demonstrated by the rise of behaviorism at the same time as the rise of control theory. We are dominated by the metaphors we live by more than I would like to admit. Rationalism as a tradition is less dominated by metaphor than virtually any other, perhaps only because it has chosen the least corruptible and most generalizable metaphor: mathematics and computation by way of Descartes and Spinoza, which is so universally applicable that it is likely incorrect to even refer to it as a metaphor even though a potentially descriptive way of thinking of the corpus of mathematics is the total of meaningfully descriptive metaphors. However, this does not mean that rationalism is not without its limits. In order to relate two objects, one must be able to specify the process by which one relates those two objects, leading one to need to specify which objects one is relating before the process of relation can take place. In short, rationalism can look at anything, but it can't look at everything at once. It can know that holism is a true principle—a totality must be by definition total—but a totality cannot be analyzed without abstracting away some of its content. This, I think, provides a pretty good way to think about the epistemic basis for trying to implement something like Functional Decision Theory or extensions or modifications thereof, as FDT serves as a method of circumventing immanent reactivity. Of course, a central problem of implementing specific policies in an FDT strategy remains how one generates policies, which must still contend with the black swan risk inherent in attention prioritization.  

2. 

Rationalism as a demographic and tradition is largely in denial of the fact that the vast majority of the population considers it to be only a naive affectation. The common conception is that rationalism is fit only for those who have failed to find a place in more sponsored communication protocols, and instead eke out an uneasy existence at the margins, powered by autistic focus but somehow failing to truly understand the reindeer games of social life. At the same time, it is also the only tradition that is able to reliably describe the world, and thus some version of it must—even under ersatz attribution—be considered the source of all knowledge that is not accidental. If one were to notice a fire burning in the forest and know instinctively not to burn oneself, rational thought would be required to predict the spread of the flames, to say nothing of the type of rationality required to discover cooking, or the fact that the evolutionary process that generated those instincts was able to encode rational behavior, even only through blind selection or mimesis. It is rather pathetic that I have to write this, but knowing the reason for a phenomenon's behavior is the only method by which one could modify that phenomenon. Deal only with cause and effect. Fortune is blind to all influence: If luck is law, one could never become lucky intentionally. 

Yet, there is something of an uncomfortable class war that has never truly concluded in Western Civilization. Rational materialists have a tendency to be negged into submission by pseudorational idealists as a method of extracting their labor. This stance is taken by Chapman in Geeks, MOPs, and Sociopaths, serving as a good countermodel to Rao's Gervais Principlewhich fails to properly identify the Geek and the Clueless as the same type of worker. This leaves me with a responsibility to consider populist alternatives to reason even if, by negation, we have to understand that there is no logic more powerful than reason and thus supposedly "transrational" modes of thought are patently absurd and exist only to exploit the intellectual charity of rational intelligences. We must consider strategies that choose to take purposefully irrational action as a method to gain short term advantage. This can be explained as either simple ignorance, or a Hamiltonian Spite strategy where the instigator can more easily survive the chaos of confusion and benefit from a position of comparative advantage. I am of course asserting that all irrational action is costly, which might look like I'm ignoring situations such as the victory of King Carl at the Battle of Narva, but this is simply a case of accidental rationality through abject stupidity on the part of Carl, as conceivably with advanced enough meteorology he could have known he was making the correct choice, but with the sensors of 1700 there's no way in hell he did. Thus, one has to assume that Carl was simply ignorant, but there are doctrines—mostly theistic in nature—that might claim Carl possessed great wisdom. However, we have return to the proposition earlier in the paragraph, and note that those perspectives are either suicidal or only attempting at an instinctual standard of life in their willful inability to analyze the world, or are being exploited by cynical Hamiltonian Spite strategies that exploit their unscientific perspective. 

Unfortunately (and for obvious reasons) both pre-and-postmodernism tend to avoid systematic decision theory problems and models. Yet, recently in Rome, M. Vassar Arc introduced me to a song from the singer-songwriter Jewel Kircher’s 1995 album Pieces of You, which, with a bit of paranoid extrapolation, provides basically all the raw material necessary to formulate an alternative test to the Newcomb Problem that better depicts the supposed "transrational" reasoning popular in postmodern society. I will call this test the Sensitive Problem for reasons of clarity. For the moment, we will focus on a single clause of the song, but the rest of the lyrics simply reinforce this distillation. Let us read this according to Barthes’ principle of the death of the author and be agnostic as to whether it is Kircher speaking or an imagined character who I will call The Sensitive One, as I am going to read far more into these lyrics than was likely ever intended by their author. 

I was thinking, that it might do some good
If we robbed the cynics and took all their food
That way what they believe will have taken place
And we'd give it to everybody who'll have some faith
So please be careful with me, I'm sensitive
And I'd like to stay that way


Let's further specify this “Sensitive Problem" and make it as similar to the original Newcomb as possible. I know this is a case of somewhat manic apophenia, but bear with me.  

The Predictor analyzes an agent and tries to see if the agent would be willing to attempt to win a fight to the death tomorrow night, against a random opponent.  If the agent is willing to try to defend its own life, it is invariably killed by the Predictor. If it would go entirely limp and not fight for itself, it survives.

The Newcomb problem tests rational self-interest in a scenario where local rational self interest isn't enough, and one has to collaborate with one's past and future selves to gain an optimal outcome. The Sensitive Problem is the opposite. One must intentionally avoid collaborating with one's past and future selves to remain alive—note that The Sensitive Problem states that it might do some good to take all of their food, not some. One must assume an intense negative valuation on future coordination, negative to a degree that it is not worth preserving any agency at all. Specifically, The Sensitive Problem claims that one should not attempt to rationally appreciate the situation, as rationally appreciating risk produces a cynical loss of innocence, equating innocence with sensitivity. This is to say, a finite circumstance with at least local zero-sum characteristics—which is to say any circumstance with meaningful choice or consequence—is a possibility-limiting environment because it fundamentally demands a response. Finite circumstances respond to one's will and action. The only alternative possible environment is a hypothetical infinite space where, probabilistically, everything happens at least once and there is no way to pressure that environment into a limited set of behaviors. I'm Sensitive asserts that any will and action fundamentally limits the potential for one’s immanent awareness of life to make the correct choice without considering the possibility of failure.

The Sensitive One asserts, in another part of the song, that she has this theory that if we’re told we’re bad it’s the only idea we’d ever have. It’s a far more extreme version of Nozick’s position, a concept called alternately Theucides’ Trap, the Hobbesian Trap, Schelling’s Dilemma, or Liu’s Dark Forrest where the recognition of the potential of defection by one party leads to the assumption of its necessity by the other. The Sensitive One implicitly asserts that the recognition of the possibility of defection is the source of defection (or abstract evil) itself, which makes the problem a test of morality, rather than a test of rationality. Intriguingly, by being a test of morality, it implicitly takes a specific perspective, with specific values, and tests whether or not the test-taker fits in with those values. Opposed to Nozick, The Sensitive One implicitly takes the stance that its not worth grappling with the problem of coordination with past and future selves, and indeed suggests that the entire problem comes from the sort of thinking that invariably arises when one attempts to act strategically across time. Quoting further from I'm Sensitive: Anyone can start a conflict, its harder yet to disregard it. In reality, there is of course room for the strategy of at least temporarily circumventing conflicts, but this is not synonymous with resolving them. If one’s will is in conflict with another’s, the only ways to resolve this are either to disregard one of the two wills by prosecuting the conflict, merging the two wills in compromise, or by capitulating and retreating out of the bounds of the conflict. The Sensitive One's theory is that somehow by retreating from the bounds of that conflict, a benevolent outcome will happen somehow by default. In practice, this translates to a moral obligation to eat the seed corn by conflating environmental constraint and social constraint, as well as entirely ignoring the possibility of low-communication environments demanding strong border defenses. There are all sorts of environments where correct action can only be informed by a clear theory of tradeoffs. Sometimes those tradeoffs effect agents other than oneself, though any sensible utilitarian would recognize the validity of other persons as moral patients and attempt to avoid externalizing harm.  

The perspective of the Sensitive Problem is the sort of thinking that I see in the type of liberal or leftist who somehow wants to infantilize men like King and Gandhi as unable or unwilling to sympathize or empathize with the logic of their adversaries. Amakusa Shirō, the leader of the Shimabara Rebellion, seems to me a much better example of the misinterpretation of King and Gandhi, for they—and especially King—were dispassionate strategists, possessed by indomitable wills that demanded the analysis of their opposition. There is certainly much to be said about their careers, but one cannot count for their ability to respond to threat an opportunity by good intentions alone. It is equivalent to assuming that the whole Christ as a strategist is included in the notion that one should turn the other cheek, ignoring all of his less passive sayings and deeds and assuming that such an aphorism held eternally outside of context. So, what does it mean to see the world from another angle and be an everyday angel, if not to embrace a sense of hope at the expense of rigorous tactics? Amacusa certainly seemed to be possessed by this hope, leading to a needless civil war committed by faithful men and women who must have been genuinely confused about the suicidal nature of their cause.  

Even as I condemn The Sensitive One for doubting the promise of rationality, I must remind myself to sympathize with its fear. Reason implies that one could isolate clear cause and effect. One promises to one’s self or others that action may follow observation, either by the senses or the intellect, and that one may observe one’s own actions to confront and perfect them. The trouble is that this covenant is so often broken when individuals fail to summon the courage to confront the ever-multiplying province of implication. Convenient myths often justify action far more interpretably than causes that are either obscure or incommunicable. Perhaps if one was alone and engaged in an abstract activity, composing music say, and keeping it secret or releasing it with the cold comfort of total anonymity, one could justify every deed. Yet, in the social context, gaslighting sets in. The reconciliation of multiple perspectives is imperfect at best, and it is far easier to abandon the often disappointing sanity of probabilistic knowledge in favor of the mutability of definite mysticism. One must grapple with one’s uncertainty in order to know anything at all, and seldom is anyone up to the task every time it is demanded. I see no other alternative to this simple humility in light of the fundamentally finite character of our understanding. Yet, while I find it absurd and frankly perverse, I can understand the desire for a logic more powerful than reason. Philosophy is indeed incoherent when it fails to confront the fact that it is always limited by its own perspective, something that everyone from Kant, to al-Ghazali and Chomsky reliably remind us of. 

The Sensitive Problem focuses on the political, so we will remain focused on that scale of relationship for the moment. Even when a society is meaningfully divided, the divisions are rarely appreciated as constraints and are instead assumed to be immutable realities, demanding factional war. Such factions are rarely so mutually incompatible as to demand a violent response, and yet as history tells us such violence has arisen time and again. Though somewhat incomplete with a number of geopolitical considerations, The Nazis’ manufacture of both internal and external threats is the canonical example of pseudorationalist crimes in the modern consciousness, but there are countless other examples that need not be listed again here. The fear of such pointless brutality that animates blind hope in the everyday angels is also the emotional drive that forces me to try and confront the reasoning of the brutalizers. The trouble is that the everyday angel perspective assumes that distinctions between persons itself animates the will to violence, rather than asserting that the will to violence is independent of differences between persons.

Yet, from history to date, it is rather unclear if a state of war is the rule or the exception in human society at large, though one could of course make far better claims regarding individual regions. However, in the case of the Sensitive Problem, a historical analysis is hardly persuasive to the perspectives that themselves find the Sensitive Problem persuasive. The sensitive problem asserts that it would do some good to eliminate the entire discipline of precedential analysis, as the cynics have to have made a judgement that it is correct to be cynical, and precedential analysis would be entirely meaningless if only positive judgements were considered admissible. We are left with an ahistorical mode of analysis. Whether out of weak thought or potent thoughtlessness, humanity often does decide that cruelty is a solution to conflict. Such depravity is not eliminated by ignoring it, but only by correcting the errors in reasoning that justified it. Such cruelty only emerges when we lack the imagination or bravery to introspect and develop better plans. In the case of the Nazis or any other criminals who were able to organize mob violence, a justification that is good enough to produce a sense of group-unity against a scapegoating target is enough to mobilize coordinated aggressive action simply by making it difficult to consider another option. Visibly questioning the party line simply makes you easy prey. It doesn’t matter if the justification makes sense, it matters only if it makes enough sense to make it difficult to question yourself.

Troublingly, this type of offensive violent action is powered far more often by moral judgements similar to the Sensitive Problem than the Cynics judged by that problem. The Sensitive Problem’s Predictor creates an in-group by testing for a similar violent intent while pretending its intent is pacifistic. While the pretense of enforcing a prohibition on cynical accumulation of resources makes the lyrics a modicum more tolerable, the violence is still there. Again, note that it says all not a surplus of the cynics' food. This is not to say that the paranoid stockpiling of resources at the expense of collective economic or ecological utility is always justified, but the position of the Sensitive Problem is unjustifiably extreme. The Sensitive One asserts that somehow lacking faith that one’s needs would simply be provided for without intentional preparation is defection against the human spirit, and that concern for one’s future autonomy is the origin of conflict rather than simply an attempt to endure it. The mind, in the account of the song, is not the sort of instrument that is supposed to be used to make tradeoffs. Indeed, the capacity to do evil is confused with the capacity to accomplish anything at all besides the construction of a unifying state of hope. There is much to be said for unifying states of hope, but the murder of all populations that harshed the vibe by tracking the potential for disaster is irrefutably insane. Passing the sensitive problem is equivalent to culling the entire demographic that keeps records, knows history, and is willing to act with strategic independence.  

In short, I have been analyzing the lyrics of a pop song that demands a pogrom. What is a pogrom if not a scapegoating raid on an outgroup that has kept records of the world when the dominant culture has not? In famine, a hungry crowd invariably emerges, and the potential solutions are either charity, a jubilee, a pogrom, or a bacchanal that takes the whole of the society back to a state of nature. It matters not what causes the famine—it could just as easily be a famine of validation as a famine of calories—it matters only that the crowd has failed to care for itself, and looks towards easy prey. Often, the easiest prey is those persons that spent their time tracking the behavior of world instead of aspiring to high rank in the ugly meritocracy of violence. It is far easier for a ruling class to direct the rage of the hungry crowd at a scapegoat than to attempt to remedy the situation. The ruling class blames the hunger of the crowd on the act of keeping account of the process that led to the hunger in the first place. This act of bad faith ignores the possibility of learning from the past and avoiding further devastation in the future. Instead, it attempts to obliterate history, and see the world from another angle: that of spite and of hope, of good and evil, but never the inviolate knowledge of cause and effect. Those who have focused on the object-level mechanisms that determine the behavior of the world are so often blind to the metagame of threat and opportunity that distributes power and leverage.

It seems like the central question for political philosophy in the twenty-first century is simply how to make the conflict between game and metagame, between text and subtext, visible to those nerds and materialists who have thus far successfully avoided being scapegoated through strategies of either trade or invisibility. I'm not sure how to accomplish that, but it seems highly unlikely that it involves amputating the ability to successfully complete the million dollar solution to Newcomb's Problem in order to avoid death in the Sensitive Problem. It strikes me that the attachment to try and look for a logic more powerful than reason comes from a misinterpretation of the structure of risk and reward in the world. As we near the close of the second decade of the twenty first century, we exist in a period of normalized Lovecraftianism, with various specters of doom available for any given political persuasion. Many of those who have seen the promise of rationality broken see present progress as debt financed, with repossession just around the corner, or as accidental where no individual has anything approximating better than random agency over the physical world. The potential for total extinction leads the mind into a confusion that all progress might be inherently risky unless, somehow, it could make disaster entirely impossible. One undervalues the future, assuming that the potential of black swan events implies that we live in a world of finite reward and infinite risk, and that in that environment the correct thing to do is to stop playing.

The question is, why? This is madness, certainly, but it must come out of something. I believe that it comes out of invalidating one's own perspective and assuming that another perspective is somehow able to validate behavior by its fiat, rather than one's own experience. The anxiety of individuals who interpret the world through a social lens of praise and blame rather than cause and effect is understandable, if not itself worth experiencing. If words crush things that are unseen, this implies that the Sensitive One and others like it are affected by other individuals perceptions far more than would be adaptive as an individual. If one does not identify with one's own perspective, and instead with private or public opinion of another individual or collective, then indeed the correct thing to do is to avoid all potential blame and ignore the promise of rational action. It is always in the interests of the validator to praise and blame according to their own interests if the individual who they affect, for that individual is no longer an individual and is instead an extension of their will—a Pavlovian thrall.

I will terminate this essay at the moment, but will leave off with one final question that I do not currently have an answer to: What is the legal status of Pavlovian thralls? At what place does agency start and end? When one is animated by the will of another, who is responsible for the action? I am inclined to think that the question is incoherent. Agents, as individual processes, are almost as absurd to think about as inviolate taxonomies of the world as fully discrete species as described often by pop ecologists. Yet, if one is to attempt to think of correct action in the world, isolating the local cause of incorrect action and shifting it seems the only potential method for restorative justice. All justice systems are limited by their ontology, and it seems highly unlikely that the West will be able to reform its laws in the coming centuries if it still takes as immutable a concept of individuality derived not from the ability to reason, but from Roman property rights.