The Robot's Rebellion: Finding Meaning in the Age of Darwin - by Keith E. Stanovich

Among the many astonishing and unnerving insights of universal Darwinism is that humans serve as the hosts for two replicators (genes and memes) that have no interest in humans other than the role they play as a conduit for replication.

The fundamental insight triggered by memetic studies is that a belief may spread without necessarily being true or helping the human being holding the belief in any way.

I use the term "robot's rebellion" to refer to the package of evolutionary insights and cognitive reforms that are necessary if we want to transcend the limited interests of the replicators and define our own autonomous goals. We may well be robots-vehicles designed for replicator propagation-but we are the only robots who have discovered that we have interests that are separate from the interests of the replicators. We indeed are the runaway robot of science fiction stories-the robot who subordinates its creators interests to its own interests.

A major theme of this book is that rationality (and its embodiment in institutions) provides a means of creating conditions that optimize at the level of people rather than the genes-the beginning of the robot's rebellion.


We were constructed to serve the interests of our genes, not the reverse. The popular notion-that genes "are there to make copies of us"-is 180 degrees off. We are here so that the genes can make copies of themselves! They are primary, we (as people) are secondary. The reason we exist is because it once served their ends to create us.

Many people who think that they believe in evolution fail to think through the implications of a process that is algorithmic-mechanical, mindless, and purposeless. One of the themes of this book is that humans are at risk of being passive conduits for the interests and goals of their genes if they do not recognize the logic of their origins as vehicles for mindless replicators. The term vehicle, with its pejorative connotations when used in the context of humans, throws down the challenge that I feel is necessary to motivate efforts at cognitive reform.

Human brains represent the culmination of an evolutionary trend towards the emancipation of survival machines as executive decision-makers from their ultimate masters, the genes... By dictating the way survival machines and their nervous systems are built, genes exert ultimate power over behavior. But the moment-to-moment decisions about what to do next are taken by the nervous system. Genes are the primary policy-makers; brains are the executives. But as brains became more highly developed, they took over more and more of the actual policy decisions, using tricks like learning and simulation in doing so. The logical conclusion to this trend, not yet reached in any species, would be for the genes to give the survival machine a single overall policy instruction: do whatever you think best to keep us alive. The type of long-leash control that Dawkins is referring to is built in addition to (rather than as a replacement for) the short-leash genetic control mechanisms that earlier evolutionary adaptation has installed in the brain. That is, the different types of brain control that evolve do not replace earlier ones, but instead are layered on top of them-and of course perhaps alter the earlier structures as well.

Humans were the first vehicle to realize this startling fact: The genes will always sacrifice the vehicle if it is in their interests to do so.

On the commonsense assumption that bodies are primary in evolution and DNA is there to serve them, the presence of DNA without a function seems a mystery. However, once it is understood that DNA is there only to replicate itself, it is no longer puzzling why so much of our DNA is junk. It is essentially a parasite. If the genes have to code for a protein and cooperate in building a body in order to get replicated, they will. But if DNA can get replicated without aiding in the building of the vehicle, that is fine too. Replicators "care"-to use anthropomorphic language-only about replicating! Junk DNA is a puzzle only if we are clinging to the assumption that our genes are there to do something for us- instead of the correct view that we are there to do something for them!

The first step in the robot's rebellion, then, is to learn how to properly value the vehicle and to stop behaviors and cultural practices that implicitly value our genes over ourselves.


Evidence from cognitive neuroscience and cognitive psychology is converging on the conclusion that the functioning of the brain can be characterized by two different types of cognition having somewhat different functions and different strengths and weaknesses. The two processes have sometimes been labeled System 1 and System 2 in the literature.

In dual-process theories, one of the systems of processing is characterized as automatic, heuristic-based, and relatively undemanding of computational capacity. Thus, this system (often termed the heuristic system-System 1 in Stanovich's taxonomy) conjoins properties of automaticity, modularity, and heuristic processing as these constructs have been variously discussed in cognitive science. Among other things, an automatic process is a process that can execute while attention is directed elsewhere. The heuristic system (System 1) responds automatically and rapidly to the holistic properties of stimuli. It is biased toward judgments based on overall similarity to stored prototypes.

The other system of processing (often termed the analytic system-System 2 in Stanovich's taxonomy) conjoins the various characteristics that psychologists have viewed as typifying controlled processing. Analytic cognitive processes are serial (as opposed to parallel), rule-based, often language-based, computationally expensive-and they are the focus of our awareness. Analytic processing is in play when psychologists and the layperson talk of things like "conscious problem solving."

Although the analytic system is a powerful mechanism for logical, symbolic thought, its decontextualizing cognitive styles are computationally expensive and difficult to sustain. Analytic cognition is "unnatural," and therefore rare, in this view, because it is not an architecture that is hard-wired in the brain, separate from the autonomous set of systems (TASS). Instead, in the view articulated in Daniel Dennett's book Consciousness Explained, analytic processing is carried out by a serial virtual machine that is simulated by the largely parallel brain hardware. A virtual machine is the set of instructions that run on the hardware of a digital computer ("a virtual machine is a temporary set of highly structured regularities imposed on the underlying hardware by a program ... that give the hardware a huge, interlocking set of habits or dispositions to react," Dennett 1991). In short, the analytic system is closer to software in this view (what Clark 2001, and Perkins 1995, term "mindware") than to a separate hardware architecture.

Such a view of the differences between TASS and the analytic system is consistent with a long-standing irony in the artificial intelligence literature: those things that are easy for humans to do (recognize faces, perceive three dimensional objects, understand language) are hard for computers, and the things it is hard for humans to do (use logic, reason with probabilities) computers can do easily. The current view of the differences between TASS and analytic processing removes all of the air of paradox about these artificial intelligence findings. Computers have built up no finely-honed TASS subsystems through hundreds of thousands of years of evolution, so the things that the massively parallel and efficient human TASS systems do well because of this evolutionary heritage computers find difficult. In contrast, the analytic system of humans-the serial processor necessary for logic-is a recent software addition to the brain, running as somewhat of a kludge (in computer science, an inelegant solution to a problem) on massively parallel hardware that was designed for something else. In contrast, computers were originally intentionally designed to be serial processors working according to the rules of logic. It is no wonder that logic is easy for them and difficult for us.

Our analytic systems can learn to give better narrative accounts of our behavior-ones more in accord with the neuropsychological facts. Learning this skill is part of the robot's rebellion.

Processing without Awareness: There Are Martians in Your Brain!

The essence of TASS subprocesses is that they trigger whenever their appropriate stimuli are detected, that they cannot be selectively "turned off," and that they occur outside of our awareness. That such processes may even be priming responses that analytic processing deems infelicitous, means that, as the title of this chapter suggests, sometimes a person may have a brain that is, in an important sense, at war with itself. It may require some cognitive remediation if the outcome of this war is to be what the person's deepest, most reflective self wants. The first step in such a meliorist program of cognitive reform is to recognize that in some sense the "I" that we identify with (the homunculus that, although a fiction, is still part of our folk psychology) is not only not in control of all parts of the brain, but may be positively alienated from the operation of some brain activities that take place beyond its awareness.

TASS is an evolutionarily older system that sometimes primes outputs not appropriate in the modern world. A major theme of this book will be that modern life increasingly creates situations of this type-situations where we must call on the evaluative and supervisory functions of the analytic system to overcome habitual responses that are no longer serving our needs.

Thus, some of the response tendencies primed by TASS will need to be overridden if they are not to overwhelm our reflective values, as determined by our analytic processing activities. The triggers in our brains for eating sweets, for rejecting a raped spouse, for feeling revulsion at the sight of a disfigured person, are parts of ourselves with which we do not want to identify. They are alien to our considered selves. They are the Martians in our brains. We cannot remove them, but we can dampen their effects and find ways for the analytic system to trump them, so that we don't end up feeling like Bob Marley-that "reflexes got the better of us."

The Brain on a Long Leash and the Brain on a Short Leash

The goal structures of TASS and the analytic system are different, and that important consequences for human self-fulfillment follow from this fact. The goal structure of TASS has been shaped by evolution to closely track increases in the reproduction probability of genes. The analytic system is primarily a control system focused on the interests of the whole person. It is the primary maximizer of an individual's personal goal satisfaction. Maximizing the latter will occasionally result in sacrificing genetic fitness.

Thus, another difference between TASS and the analytic system is that TASS instantiates short-leashed genetic goals, whereas the analytic system instantiates a flexible goal hierarchy that is oriented toward maximizing goal satisfaction at the level of the whole organism. Because the analytic system is more attuned to the person's needs as a coherent organism than is TASS (which is more directly tuned to the ancient reproductive goals of the subpersonal replicators), in the minority of cases where the outputs of the two systems conflict, people will often be better off if they can accomplish an analytic system override of the TASS-triggered output. Such a system conflict is likely to be signaling a vehicle/replicator goal mismatch and, statistically, such a mismatch is more likely to be resolved in favor of the vehicle (which all of us should want) if the TASS output is overridden.

Don't Be Sphexish

Dennett refers to "that spooky sense one often gets when observing or learning about insects and other lower animals: all that bustling activity but there's nobody home!". Quoting cognitive scientist Douglas Hofstadter, Dennett proposes that we call this unnerving property sphexishness. He points out that observing the simple, rigid routines that underpin the complexity of the surface behavior of simple creatures spawns in us the worrying thought: "What makes you sure you're not sphexish-at least a little bit?"

Dual-process models of cognition such as those discussed in this chapter all propose, in one way or another, that in fact we all are a little bit sphexish. In fact, many of these theories, in emphasizing the pervasiveness of TASS and the rarity and difficulty of analytic processing, are in effect proposing that our default mode of processing is sphexish. If we want to be more than a Sphex, then we must continually accomplish the difficult task of marshaling cognitive power to run a serial simulator containing mindware that can monitor TASS to ensure that it is fulfilling vehicle-level goals.

The TASS modules offering up outputs that conflict with reasoned analytic system outputs could be viewed as "the Sphex within you." It does not assuage our fears of sphexishness to note that some of the automatic processes in TASS are products of our environment-learned rules that were practiced so much that they now execute autonomously. That response tendencies may have been instantiated in TASS by advertising, by our peer group in adolescence, or by the repetition by our parents of prescriptions deriving from their own limited experiences makes them no less sphexish- they are just as unreflective and as unconsidered as any other TASS processes. Only rules that were instantiated in TASS by reflective thought should be honored and identified with, and even these may occasionally be overgeneralized (because now they will be firing automatically) and need overriding in a particular situation.

The robot's rebellion derives in part from our ability to recognize the potential for sphexishness in our own behavior and take steps to prevent it. The threat to the soul represented by Darwinism is right there in our own brains-in the TASS subsystems that could turn us into sphexish automatons, into robots fulfilling not our own goals but those of the selfish replicators who built us. But there too, in the same brain, is the potential awareness of this plight along with the cognitive mechanisms to overcome it. Uncritically delighting in following our so-called gut instincts makes us little more than slaves of the mindless replicators-the micro-automata that view us as nothing more than vehicles to assist in their quest for replication.


One key component of rationality is the optimization of goal fulfillment at the level of the individual. This is why evolutionarily adaptive behavior is not the same as rational behavior. Evolutionarily adaptive behavior is behavior that increases genetic reproduction probabilities, but rational behavior is behavior that fulfills the goals of the vehicle-given the vehicle's set of beliefs about the world.

No human has optimizing genetic fitness as an explicit goal. We seek not evolution's ultimate goal of replication, but the more proximal outcome that has been installed in us during evolution-the enjoyment of sexual pleasure. If the means are pleasurable, analytic intelligence acting rationally in the service of the vehicle pursues them-not the ultimate goal of the genes. Thus, the revolt of the human survival machines-the robot's rebellion-consists of the gene-built humans trying to maximize their own utility rather than the reproduction probability of their creators in situations where the two are in conflict. If we want to be more than just survival machines for our genes, then we must use our analytic processing systems in the active pursuit of instrumental rationality. There is no more potent mental software in the robot's arsenal than that provided by the tools of rational thought.

Instrumental rationality is behaving in the world so that you get exactly what you most want, given the resources (physical and mental) available to you.

The hedonic experience of the vehicle is just a means to an end for most of the goals lodged in TASS, and it will readily sacrifice the vehicle's hedonic pleasure if ultimate fitness goals are achievable without it. We instead should be focused on our own personal goals as vehicles. We diminish our selfhood if we compromise these in order to honor responses programmed by subpersonal entities that constructed the Darwinian parts of our brains.


The terms heuristics and biases refer to the two aspects of research studies of this genre. First, human performance was shown to deviate from what was rationally expected on a particular task (a cognitive bias is demonstrated). Next, the deviation is shown to result from the operation of an automatic heuristic-a response triggered by TASS. These heuristics are assumed to be useful in many situations (and are computationally inexpensive). But where logical, analytic, and/or decontextualized problem solving is required, TASS heuristics bias processing in the wrong direction, thus resulting in suboptimal responses.

These two aspects of how people approach decisions-the differential steepness of the utility function and that they recode options as being from a zero reference point of the current status quo (in terms of wealth, lives, or whatever is at issue)-appear to be TASS characteristics. They are automatic ways that information in decision situations is coded. They cannot be turned off, but they can be overridden (for example, by analytic strategies to make sure that one's preferences are invariant under different representations of the problem).

That these two characteristics of the way we code decision information can cause an outright reversal of preferences if not overridden is a somewhat frightening implication of the way our TASS preference apparatus is structured. There is the unsettling idea latent here that people's preferences come from the outside (from whoever has the power to shape the environment and determine how questions are phrased) rather than internal preferences based in their unique psychologies. Since most situations can be reframed either way, this means that rather than having stable preferences that are just elicited in different ways, the elicitation process itself can totally determine what the preference will be! Such a conception brings the foundational concepts behind the idea of "rational man" from economics crashing down. It also has potent social implications. As Kahneman has pointed out, the assumption of stable, rational, well-ordered preferences has been used to "support the position that it is unnecessary to protect people against the consequences of their choices".

TASS is biased toward the automatic acceptance of propositions and biased to accept the context as given. If the exploration of alternative hypotheses and alternative framings of issues are to take place, analytic system override of this natural processing tendency is essential.

It is the nonautonomous, serial-processing operations of executive control and problem solving that serve to guarantee instrumental rationality by overriding the responses generated by TASS when the latter threaten optimal outcomes at the personal level.

In situations where the present human environment is similar to the environment of evolutionary adaptedness (EEA), the human brain is characterized by a set of fundamental computational biases that quite efficiently facilitate goal achievement. However, when technological societies create new problems that confound these evolutionarily adapted mechanisms, humans must use cognitive mechanisms that are in part cultural inventions to override the fundamental computational biases that, in these situations, will prime the wrong response. These culturally induced processing modes are the abstract, analytic, but computationally expensive processes of rational thought. It is important to recall that TASS processing pervades all functioning, and that it cannot be "turned off" but instead must be overridden on a case by case basis.

The fundamental computational biases of human cognition derive from the automatic inferential machinery of the TASS brain. These biases have the effect of providing rich supplemental knowledge to augment the sometimes fragmentary and incomplete information we receive when faced with real-world problems. The four interrelated biases are: 1. The tendency to contextualize a problem with as much prior knowledge as is easily accessible, even when the problem is formal and the only solution is a content-free rule, 2. The tendency to "socialize" problems even in situations where interpersonal cues are few, 3. The tendency to see deliberative design and pattern in situations that lack intentional design and pattern, and 4. The tendency toward a narrative mode of thought.

The socialization of problems and the tendency to see deliberate design in the environment follow from the evolutionary assumptions behind the social intelligence hypothesis-that attributing intentionality in order to predict the behavior of conspecifics and to coordinate behavior with them was a major evolutionary hurdle facing the social primates, in many cases more computationally complex than mastering the physical environment. The tendency to attribute intention may have other evolutionary sources. Dennett suggests that we have an innate disposition to treat things that rapidly fluctuate in our environment as if they had souls and that this is an evolutionary design trick allowing us to categorize things in the world without thinking too much. The ubiquitous tendency to adopt what Dennett calls the intentional stance underlies many of the fundamental computational biases (particularly the tendency to see human design in the world and to socialize problems).

Finally, there exist many theoretical arguments for why the automatic contextualization of problems with prior knowledge might be adaptive. For example, Evans and Over argue that beliefs that have served us well in the past should be hard to dislodge, and projecting them on to new information-because of their past efficacy-might help in assimilating the new information.

Evolutionary Reinterpretations of Responses on Heuristics and Biases Tasks

Evolutionary psychologists, adaptationist modelers, and ecological theorists have reinterpreted the modal response in most of the classic heuristics and biases experiments, previously interpreted as indications of human irrationality, as instead indicating an optimal information processing adaptation on the part of the subjects.

The dual-process cognitive architecture introduced in chapter 2 encompasses both the impressive record of descriptive accuracy enjoyed by a variety of evolutionary/adaptationist models as well as the fact that cognitive ability sometimes dissociates from the response deemed optimal on an adaptationist analysis. These data patterns make sense if it is assumed: (1) that there are two systems of processing with the properties outlined in chapter 2; (2) that the two systems of processing are optimized for different situations and different goals as outlined in chapter 2; and (3) that in individuals of higher cognitive ability there is a greater probability that the analytic system will override the response primed by TASS.

The Fundamental Computational Biases and the Demands for Decontextualization in Modern Society

Evolutionary psychologists are prone to emphasize situations where genetic goals and personal goals coincide. They are not wrong to do so, because this is often the case. Accurately navigating around objects in the natural world was adaptive during the EEA, and it similarly serves our personal goals as we carry out our lives in the modern world. Likewise, with other evolutionary adaptations.

But none of this means that the overlap is necessarily 100 percent. Unfortunately, the modern world tends to create situations where some of the default values of evolutionarily adapted cognitive systems are not optimal. Many of these situations implicate the fundamental computational biases discussed previously. These biases serve to radically contextualize problem-solving situations. In contrast, modern technological societies continually spawn situations where humans must decontextualize information-where they must deal abstractly and in a depersonalized manner with information. Such situations require the active suppression of the social, narrative, and contextualizing styles that characterize the operation of TASS. These situations may not be numerous, but they tend to be in particularly important domains of modern life-indeed, they in part define modern life in postindustrial knowledge-based societies.

The TASS Traps of the Modern World

It is possible to accept most of the conclusions of the evolutionary psychologists but to draw completely different morals from them. The evolutionary psychologists want to celebrate the astonishing job that evolution did in adapting the human cognitive apparatus to the Pleistocene environment. Certainly they are right to do so. The more we understand about evolutionary mechanisms, the more awed appreciation we have for them. But at the same time, it is not inconsistent for a person to be horrified that a multimillion dollar advertising industry is in part predicated on creating stimuli that will trigger TASS heuristics that many of us will not have the cognitive energy or cognitive disposition to override. I personally find it no great consolation that the heuristics so triggered were evolutionarily adaptive in their day. What I am attempting to combat here is a connotation implicit in some writings in evolutionary psychology that there is nothing to be gained from being able to understand a formal rule at an abstract level (the conjunction rule of probability, etc.)-and no advantage in flexibly overriding the fundamental computational biases at times.

An uneasy tension disturbs the heart of the selfish gene theory. It is the tension between gene and individual body as fundamental agent of life. On the one hand we have the beguiling image of independent DNA replicators, skipping like chamois, free and untrammelled down the generations, temporarily brought together in throwaway survival machines, immortal coils shuffling off an endless succession of mortal ones as they forge towards their separate eternities. On the other hand we look at the individual bodies themselves and each one is obviously a coherent, integrated, immensely complicated machine, with a conspicuous unity of purpose.
— Richard Dawkins, The Selfish Gene


From the standpoint of the framework developed in this book, the evolutionary psychologists' theoretical stance against domain-general processing is a profound mistake. Because it is this system that computes actions that maximize utility based on the organism's long-leashed goals, analytic processing is essential if a person is to achieve instrumental rationality. It is this system that overrides the TASS subsystems when they compute responses that are not in the interests of our current long-term goals. When is this likely to happen? Answer: In situations where the cognitive requirements of technological societies do not match those of the environment of evolutionary adaptedness (EEA). Ironically, the evolutionary psychologists themselves often mention that the EEA is not to be confused with the modern world. However, they largely fail to develop the most important implication of potential mismatches between the cognitive requirements of the EEA and those of the modern world-that if we respond in the modern world according to TASS subsystems, we will often be maximizing something other than our current goals as individual organisms.

Measures of the ability to deal with abstraction such as g remain the best employment predictor and the best earnings predictor in postindustrial societies. The psychometric literature contains numerous indications that cognitive ability is correlated with the avoidance of harmful behaviors and with success in employment settings, as well as social status attainment, independent of level of education.

In short, evolutionary psychologists are wrong to assume that TASS heuristics (adapted for the EEA) are optimal for achieving rationality in the modern world. Many important decisions in life are nearly "one shot" affairs (job offers, pension decisions, investing decisions, housing decisions, marriage decisions, reproductive decisions, etc.). Some of these decisions were not present at all in the EEA, and we have had no time nor learning trials to acquire extensive personal frequency information about them. Instead we need to make certain logical and probabilistic inferences using various rules of inference, and most importantly, we must decouple myriad sources of information that our autonomously functioning modules might be detecting and feeding into the decision ("no, the likability of this salesperson should not be a factor in my deciding on this $25,000 car").

What Follows from the Fact that Mother Nature Isn't Nice

As distinguished evolutionary biologist George Williams has eloquently pointed out, Mother Nature is in the replication business, not the niceness business. TASS was built to further the goals of the subpersonal replicators rather than your own personal goals. If you give up your life to TASS (in the form of giving in to your "gut instincts"), then you are essentially buying a lottery ticket. You are betting, when you do so, that it is one of those instances when genetic goals and personal goals coincide. You are betting that it is not one of those instances when pursuing ancient replicator goals conflicts with your personal goals. However, work in cognitive and decision science has indicated that you would sometimes lose this bet and, further, sometimes you would lose it in important circumstances and experience negative real-world consequences.


Intelligence, while representing the potential computational power to carry out TASS override, does not guarantee rationality because the capacity for override remains unrealized if not triggered by the superordinate control hierarchies that we identify as rational thinking dispositions at the intentional level. Variability in intentional-level thinking dispositions means that there is at least the potential for cognitively competent people to do irrational things. Such an analysis implies that it should be no surprise that a lot of smart people act dumb sometimes.

A dysrationalic fails to fulfill his/her personal goals despite having at least adequate cognitive capacity because it is rationality that ensures human fulfillment, not algorithmic-level capacity alone. It is thus puzzling that society seems to place so much more emphasis on algorithmic-level capacity than on intentional-level rationality. Society is obsessed with intelligence-discussing it, measuring it, demanding that schools increase it, and so on. Intelligence tests, SAT tests, and most school aptitude and achievement tests assess algorithmic-level capacities rather than intentional-level thinking dispositions. Society is much less focused on debating rationality and ways to increase it. This discontinuity is disturbing and odd. Rationality is, if anything, probably more malleable than basic algorithmic-level cognitive capacities and rationality is more important for the attainment of a person's goals. In a testing-obsessed Western society with all kinds of assessment instruments now being used in schools and industry, there is very little emphasis on assessing rational thinking skills.

The cultural evolution of rational standards is apt to occur markedly faster than human evolution. In part this cultural evolution creates the conditions whereby instrumental rationality separates from genetic optimization. As we add to the tools of rational thought, we add to the software that the analytic system can run to achieve long-leash goals that optimize actions for the individual. Learning a tool of rational thinking can quickly change behavior and reasoning in useful ways-as when a university student reads the editorial page with new reflectiveness after having just learned the rules of logic. Evolutionary change is glacial by comparison. Thus, in an astonishingly short time by evolutionary standards, humans can learn and disseminate-through education and other forms of cultural transmission-modes of thinking that can trump genetically optimized modules in our brains that have been driving our behavior for eons. Because new discoveries by innovators can be conveyed linguistically, the general populace needs only the capability to understand the new cognitive tools-not to independently discover the new tools themselves.


I prefer to view a meme as a brain control (or informational) state that can potentially cause fundamentally new behaviors and/or thoughts when replicated in another brain. Meme replication has taken place when control states that are causally similar to the source are replicated in the brain host of the copy. Philosopher Daniel Dennett lists some examples of memes (or memeplexes-co-adapted sets of memes)-the arch, the wheel, calendars, calculus, chess, the Odyssey, impressionism, deconstruction, the vendetta, the right triangle, "Greensleeves," the alphabet-to help us get the idea of memes as idea units or collections of idea units.

Collectively, genes contain the instructions for building the bodies that carry them. Collectively, memes build the culture that transmits them. The meme is a true selfish replicator in the same sense that a gene is. As with the gene, by using the term "selfish" I do not mean that memes make people selfish. Instead, I mean that memes (like genes) are true replicators that act only in their own "interests." The fundamental insight triggered by memetic studies is that a belief may spread without necessarily being true or helping the human being who holds the belief in any way.

Memetic theory has profound effects on our reasoning about ideas because it inverts the way we think about beliefs. Personality and social psychologists are traditionally apt to ask what it is about particular individuals that leads them to have certain beliefs. The causal model is one where the person determines what beliefs to have. Memetic theory asks instead what it is about certain memes that leads them to collect many "hosts" for themselves. The question is not how do people acquire beliefs (the tradition in social and cognitive psychology) but how do beliefs acquire people!

The principles of scientific inference and rational thought serve essentially as meme evaluation devices that help us to determine which beliefs are true and therefore probably of use to us, and also which are consistent with those that are true. Failure of the consistency checks embodied in the choice axioms of instrumental rationality will often point us to memes representing goals that are not serving our life plans. Scientific principles such as falsifiability are of immense usefulness in identifying possible "junk memes"-those that are really not serving our ends but merely serving the end of replication. Think about it. You will never find evidence that refutes an unfalsifiable meme. Thus, you will never have an overt reason to give up such a belief. Yet an unfalsifiable meme really says nothing about the nature of the world (because it admits to no testable predictions) and thus may not be serving our ends by helping us track the world as it is. Such beliefs are quite possibly "junk memes"-unlikely to be shed despite the fact that they do little for the individual who holds them (and may actually do harm.)

Personal Autonomy and Reflectively Acquired Memes

I distinguish between memetically acquired goals that are "caught" like viruses (as in the Dawkins quote, "a cultural trait may have evolved in the way it has, simply because it is advantageous to itself")-what we might call nonreflectively acquired memetic goals-and memetic goals that an individual acquires reflectively, with full awareness of their effects on the organism.

Memes having passed many selective tests that we reflectively apply are more likely to be memes that are resident because they serve our ends. Memes that we have acquired unreflectively-without subjecting them to logical and/or empirical scrutiny-are statistically more likely to be parasites, resident because of their own structural properties rather than because they serve our ends.

We must learn how to critique our desires and beliefs-and, in order to accomplish this, we need to have installed as mindware the tools of rational and scientific thinking.

  1. Avoid installing memes that are harmful to the vehicle physically.
  2. Regarding memes that are beliefs, seek to install only memes that are true-that is, that reflect the way the world actually is.
  3. Regarding memes that are desires, seek to install only memes that do not preclude other memeplexes becoming installed in the future.
  4. Avoid memes that resist evaluation. Question any meme that tells you that in order for it to deliver some benefit you must not question it.

Being reflective about the memes we take on as desires and beliefs is the only answer. This is particularly so with the memes that were acquired in our early lives-those that were passed on to us by parents, relatives, and other children. The longevity of these early acquired memes is likely to be the result of their having avoided consciously selective tests of their usefulness. They were not subjected to selective tests because they were acquired during a time when you lacked reflective capacities.

We inhabit a cognitive environment in which there is a widespread hostility to examining belief. Perhaps the unsupportable memes have allied together to create in us a cognitive environment antithetical to the idea that our beliefs and desires need evaluation. Or, another way to put it is: If most of our beliefs are serving us well as vehicles and are able to pass selective tests of their efficacy, why shouldn't they have created a cognitive bias to submit themselves to those very tests-tests that their competitors would surely fail. Instead, the memosphere of most of us is vaguely discouraging of stiff real-world tests of our beliefs. This prompts the worrisome question: What have these memes got to hide?

Memetic Concepts as Tools of Self-Examination

The concept of the meme is cognitively liberating because it aids in the task of mental self-examination by making it easier to achieve cognitive distancing. Memes demystify beliefs. They desanctify beliefs. Memes are solvents for beliefs, and they are epistemic equalizers in the following sense. By providing a common term for all cultural units, memetic science deconstructs the unreflective privileging of certain memes that have leaked into culture through arbitrary historical accident and through specific meme strategies that encourage privileging. The very concept of the meme will suggest to more and more people who become aware of it that they need to engage in some meme therapy-to ponder whether they have thoughtfully taken on a meme or have simply been infected by it without any critical input from their analytic intelligence.

The hostility toward the meme concept in part flows from the fact that the science of memetics has highlighted the tricks played by all faith-based memes and threatens to deprive them of their special status-a status we do not grant other memeplexes that do not contain evaluation-disabling tricks. But since such disabling memes are widely held, they provide the climate in which the meme concept is being received. They condition the cognitive environment in the direction of hostility toward meme evaluation.


The conceptual insights of evolutionary theory travel on the back of a scientific technology that people do not want to give up. Simply put, science delivers the goods. People want their technological benefits-they want the increased health, they want the cheaper food, they want the increased mobility, they want convenience and comfort. But riding along with the technologies that people want are the conceptual and metaphysical insights that the scientific discoveries trigger: that we are survival machines, that there is no immaterial "mind" where consciousness occurs and where your "self" makes decisions, that the layperson's notion of free will is utterly confused.

People want to acknowledge scientific evidence because they are aware that they live in a scientific society and wish to maintain their intellectual integrity. Thus, they know that the source of their sense of self resides in the structure of their brains and that information about the brain must be incorporated into their models. Yet at the same time they want to inject a squirt of "mystery juice" into their conceptions. They know that they cannot leave out neuroscience entirely from their conceptions, but they are equally determined that it not be allowed to explain everything.

Modern life is just too full of contradictions for someone to claim that his/her behavior was always in line with his/her higher-level values. In fact, life is so complex, and full of moral and personal choices that are potentially conflicting, that instead of admiring people who claim perfect congruence between behavior and second-order judgments, we would suspect that such people are more likely to be behaving wantonly. They are simply doing what they want, but in modern society it is socially desirable to be a strong evaluator, so they avow that they have second-order preferences. How we know people are displaying the embarrassing mismatches between their first-order and second-order preferences is when they reveal to us their internal struggle to achieve a rational integration.

We are the only species that disrupts the coherence of its desires by destabilizing them through internal cognition directed at self-improvement and self-determination.

Two-Tiered Rationality Evaluation: A Legacy of Human Cognitive Architecture

People aspire to rationality broadly conceived, not just instrumental rationality. People want their desires satisfied, but they are concerned about having the right desires. If those desires are TASS subsystems that serve ancient genetic goals better than current life goals, they need to be overridden by the analytic system pursuing considered, environmentally appropriate long-term goals. If those desires are TASS responses acquired in childhood or overpracticed rules not appropriate to the current situation, then they again need to be overridden by analytic processes in the service of a memeplex that has been thoughtfully examined. The full robot's rebellion is achieved by pursuing instrumental rationality in the context of a continuing critique of those desires being pursued.

Because humans aspire to rationality broadly rather than narrowly defined, a two-tiered evaluation of their rationality is necessary. As described in the last section, the instrumental rationality we achieve must be evaluated by taking into account the complexity of the goals being pursued and by analyzing the dynamics of the cognitive critique. Or, to put it another way, both thin and broad rationality need evaluation.

In order to achieve two-tiered rationality, I have stressed: (1) the importance of selective TASS override by the analytic system; (2) the importance of reflectively acquired beliefs; and (3) the importance of reflectively acquired desires. We can thank a feature of our cognitive architecture for making the last two possible. In holding, contemplating, and evaluating second-order desires that conflict with first-order desires, we are cognitively dealing with a hypothesized mental state-one that is actually not true of us. We are able to represent a state of affairs that does not map into an actual, causally active, mental state of our own. We are able to mark a mental state as not factual. Many cognitive theorists have emphasized the critical importance (and specialness to human mentality) of being able to separate a belief or desire from its coupling to the world (to mark it as a hypothetical state). This is what the representational abilities of the analytic system (vastly augmented by the powerful tool of language) can accomplish. These representational abilities allow you so say to yourself "if I had a different set of desires, it would be preferable to the ones I have now," and they appear to be uniquely human.

These metarepresentational abilities make possible the higher-order evaluations that determine whether we are pursuing the right aims. They make it possible to add symbolic utility to our lives and actions. They provide the distancing from belief that is necessary for meme evaluation. The critiques of the memes we host and of our first-order desires made possible by these representational abilities make it possible to evaluate whether either genes or memes are sacrificing us as human vehicles.

The Spookiness of Subpersonal Entities

The clarity of human self-examination, for all its benefits, brings with it new and scary notions for humans to face-the spooky notions of the selfish gene and equally selfish meme. It is creepy to think about the notion that these subpersonal entities both construct and constitute our bodies and minds and that they are not necessarily only in it for us (or, to put it another way, are not optimizing for people but are optimizing across people). Here's how things get creepy:

Creepy Fact #1: There is no "I" in the brain who is aware of everything going on and who controls everything.

Creepy Fact #2: Our brains were built by entities not exclusively concerned with instantiating goals that were good for us.

Creepy Fact #3: There is another subpersonal replicator that constitutes the software that the analytic system must use to monitor TASS, and this subpersonal replicator, like the gene, may sometimes have interests that conflict with vehicle well-being.

Are We Up to the Task? Finding What to Value in Our Mental Lives

It may be that in terms of the cultural evolution of human rationality, the narrow instrumental rationality of market societies is perhaps a local maxima in the evolution of ideas. Is there reason to believe that humans are up to the task of pursuing a broader concept of rationality? I would argue that we are in fact up to the task. The reason for my optimism about our broader evaluative abilities stems from the intensity of our drive to deploy our representational abilities.

When seeking the source of human uniqueness, the popular science writer's favorite candidate is consciousness. However, the status of the concept of consciousness in cognitive science is a mess, to say the least. There are influential schools of thought in cognitive science which show that, basically, we simply do not know what we are talking about when we talk of consciousness in our natural language.

I want to suggest here that the broad conceptions of rationality that I have explored in the final chapters of this book provide useful benchmarks of a meaningful cognitive life. I am not suggesting that broad rationality is the only marker of meaning or significance-only that consciousness is not the only indicator of human uniqueness; and, in fact, it may not be the best. In any case, the situation is not a zero-sum game-there may be many markers of human uniqueness, and evidence for one does not negate the importance of the others. Exploring other concepts as markers of meaning in life does not detract from the importance of ongoing studies of consciousness.

Cognitive science has begun to uncover the internal logic of an agent that values its own evaluative autonomy. The uniqueness of the mechanisms that make this possible might be a useful focus when searching for what might be a scientifically plausible concept of self in the age of Darwin and neuroscience. This seems like a more self-respecting direction to look than the slightly insulting tendency to hide behind the so-called mystery of consciousness (qualia).

Again, in a version of Nozick's experience machine thought experiment, would any of us prefer to turn ourselves into a wanton (give up the ability to form second-order evaluations) if we were guaranteed pleasurable conscious experiences as a wanton?

Our experiences result from a massive brain system monitoring itself, responding to sensory activity, exercising supervisory attention, representing at a high level of abstraction, and carrying out other important cognitive activities. Maybe we need to refocus on the activities themselves rather than the experience of these activities. Perhaps the importance that we desire to assign to human mental life should be assigned to those activities themselves rather than to the internal experiences that go with them. Another way to phrase it is that maybe there are things that are important to do other than to experience the activities. Perhaps as well it is important to do them (again, recall your response to the experience machine).

Perhaps we have overvalued consciousness in the same way educators have misperceived the importance of self-esteem in the learning process. An extremely popular hypothesis in the 1990s was that school achievement problems were the result of low self-esteem. It turns out that the relationship between self-esteem and school achievement is more likely to be in the opposite direction from that assumed by school personnel. It is superior accomplishment in school (and in other aspects of life) that leads to high self-esteem and not the reverse. Self-esteem simply rides along with the productive activities in which the individual is engaged. At least in part, consciousness rides along in an analogous way.

Perhaps we might benefit from reorienting our search for meaning away from consciousness and internal feels (qualia) and toward the evaluative activities that make us autonomous and unique agents in the world. We create meaning when we work to make second-order evaluations; work to achieve rational integration in our preference hierarchies; attempt to achieve consistency among our first-order preferences; are alert to symbolic meaning in our lives; value ourselves as a vehicle and do not let genetic proclivities in TASS sacrifice our interests in a changing technological environment. All of these activities define what is really singular about humans: that they gain control of their lives in a way unique among lifeforms on Earth-by rational self-determination.