Sarah Constantin

Motivated cognition is the state of emotionally needing to believe something is true, whether it actually is true or not.

I've found that a good kinaesthetic metaphor for motivated cognition is pressure.  If you're forcing things with your mind -- if you're going "it's GOT to be this way, or ELSE" -- then you aren't actually open to the truth being whatever it might happen to be.  

Defensiveness, anxiety, revulsion, despair, eagerness to please or to be acceptable, can motivate you to believe that the convenient and pleasant thing is true -- or that the awful worst-case scenario is -- or both at once (e.g. anxiety can make you fear the worst but flinch from it and profess the best-case scenario.)

It's not motivated cognition for motivation to be involved in cognition.  We think most clearly, in fact, when we're motivated to do so; for instance, someone who stands to make money by making the correct prediction will be more motivated to be correct than someone who's merely having a conversation.  

In fact, I think motivation is essential to all the words we use to talk about thinking well  -- rationality, wisdom, objectivity, science, empiricism, common sense, "Looking", etc. These words get corrupted by connotations of smugness, coldness, superiority, authoritarianism, etc, and new words have to be continually invented to point at the same thing the old words were intended to point at.  The thing itself is, perhaps, best described as "thinking in the way everyone naturally does when they actually care about the object of their thought."  

If you care about the thing in the real world, you will not want to be wrong about it; a delusion, however pleasant, won't give you what you want.  You still can be wrong about it, of course, but your incentives are to be as correct as you can be.  A certain amount of pretense and posturing and game-playing may drop away suddenly when, for instance, you find your child's safety is at stake; suddenly it is vitally important to get real.

(Of course, phrases like "get real", "be sensible", "be reasonable", often are used to mean "shut up and do what I tell you", which is not the thing. A person who Actually Cares about getting something done may often be perceived as an unreasonable or irrational person, because she is doing something that doesn't meet with everyone's approval.)

There's something related about words like "literally", "truly", "actually", "really", "very", "honestly" -- and it's telling that over time language evolves to make them all used as words for emphasis instead of denoting literal exact truth.  It's hard to find a way to phrase in words "I'm pointing at reality now" as opposed to pointing at a model of reality, or playing a game with language, or speaking 'in character' as the persona you want to embody right now.  

Notions like "rationality" are attempts to encourage people to think and speak literally rather than performatively.

It seems like sort of a mistake to present them as a specialized discipline to be taught rather than a stance to be adopted that most people actually have by default from time to time. Doing science doesn't actually involve going through The Scientific Method as you're taught in elementary school; but while there may not really be a Scientific Method, there is definitely a scientific mindset. It's the same mindset you have naturally when you're curious.  

If you try to codify how people think when they're being curious, it winds up sounding like nothing at all.  "Just, y'know, think! Look! Care! Try!"

Or it comes across as condescending: "Most people go through life never actually trying! You should actually try!" There's not much content to this, but the "actually" is gesturing at something: the rubber meeting the road, the moon and not the pointing finger. 

By contrast, motivated cognition is being motivated to have certain cognitions, inside your head, rather than being motivated to seek outcomes out in the world.

It's kind of weird that we have this feature at all. Why would it be evolutionarily adaptive? Or, perhaps it's not adaptive but it's a 'natural flaw' that most possible ways to make a brain would fall into?

Why do we (often) care more about the insides of our heads than what's going on outside them?

Defensiveness is a maladaptive social interaction style characterized by responding as if someone is attacking you, even when they aren't.  

If you treat innocent questions or suggestions as though they're attempts to ridicule or condemn you, you're going to damage your personal relationships. Well-meaning friends and family don't like being mischaracterized as cruel attackers. If you reflexively apologize, justify yourself, brace for a blow, or counterattack, when people in your life initiate non-hostile interactions, you're going to hurt their feelings and drive them away.

What's interesting is that a lot of common norms around politeness or professionalism are consistent with a "defensive" worldview, an assumption that people are hostile until proven otherwise.

Declining offers: it's generally considered polite to refuse offers of food or other favors.  You should assume that the other person doesn't really want to help you, and will be relieved if you don't take them up on the offer; the considerate thing to do is to reduce the burden on them.

By the same token, you don't want to appear needy, or actually ask for anything you want; the "safe" thing to assume is that the other person would feel burdened if you made a request.

Being reluctant to initiate contact: the "safe" assumption about any person is that they don't want you to disturb them and would rather be left alone.  It's impolite to reach out too much or too often.  

Discretion: it's considered potentially embarrassing or career-limiting to "overshare" or be excessively candid in public; the more mysterious you are, the more respectable you appear. The implicit belief here is that the more people know about you, the less they'll approve of you.

Justifiability: in professional or scientific discourse, there's often an (implicit or explicit) frequentist paradigm: there is a "null hypothesis" that is the default assumption, and only a certain threshold of formally presented evidence is sufficient to justify a claim that any other hypothesis is credible.  You are expected to prove a case to a hostile or skeptical audience, rather than merely motivate or explain your reasons for your belief to a curious audience; unless, of course, your belief is the null hypothesis, in which case it needs no argument at all. 

We don't normally describe these behaviors as "defensive" or see them as maladaptive. In context, they're normal. But they have a worldview in common with the worldview that damages personal relationships -- in all cases, they involve the assumption that being open about who you are, what you think, and what you want, leaves attack surface for your enemies, and that it's not even worth considering the possibility that people might be friends (who want to help you, want to interact with you, like you better after getting to know you, and would benefit from learning what you think.)

I'm not sure what exactly to conclude, but this makes me think more about the potential downsides to "defensive" norms.

Why Read Old Papers?

These days, from what I'm told by knowledgeable people, there's a fairly tight feedback loop between current aging research and the biotech industry. When a new, major aging-related paper comes out, there are people seriously evaluating whether they can start a company around it.

But that isn't necessarily true when it comes to old research. There's no automatic means by which old papers "go viral."  There are no conferences (that I know of) where people call their colleagues' attention to remarkable, decades-old results that haven't received follow-up investigation.

I think old papers deserve a second look, for a few reasons.

1.) Often a result that had little interpretability or applicability in the past can benefit from contemporary tools. 

Let's say, twenty years ago researchers found a way to extend life in rats -- but it was a surgical operation that would be too invasive or risky to try on healthy humans. And maybe that was the end of that research direction.  But now we have lots of new options! We can take tissue samples with and without the intervention, and look at gene and protein expression, even down to the individual cell level. We can identify genetic modifications or drug targets that could be used to simulate the intervention in a safer, more targeted way.  

2.) Looking at old papers reduces some of the biases that come from looking at the latest, most-cited papers.

The volume of scientific publications is increasing at an exponential rate, up to 4% a year.[1] 

• overestimating effect size in gene-association studies increases with impact factor (more bias in more prestigious journals)
• sample size in gene-association studies decreases with impact factor
• statistical power in psychology and cognitive science papers decreases with impact factor
• randomization in animal studies is reported less frequently in papers from high-impact-factor journals
• errors in supplemental data (eg Excel auto-converting a gene name to a date) are more common in papers from high-impact-factor journals
• p-value reporting errors, usually in the direction of misinterpreting a non-significant result as significant, are more common in papers from high-impact-factor journals
• metrics to identify tell-tale signs of questionable research practices find lower research quality in higher-impact-factor journals.

So, no, we can't assume that the most-cited papers of today are the cream of the crop. If anything, there's more pressure today than ever to get dramatic but untrustworthy results, and that pressure is highest at the most competitive journals.

One way to reverse this effect is to go back in time.  If the amount of noise in the system is increasing, an old paper is more likely to have a valid signal than a new one.

3.) Low technology can be a blessing in disguise.

The miracle of modern molecular biology is that we keep developing better tools to affordably do breadth-first searches. "Sequence ALL the genes!" "Quantify ALL the transcripts!" "Quantify ALL the proteins!" And so on.

The danger of having these incredible tools is that you can cherrypick positive results -- and people do.

It's much harder to erroneously get an apparently therapeutic intervention if your tools are blunter and your search space is smaller. If somebody in 1944 says that shining light on a duck's head makes its testes grow[4], then by gum I bet that actually happens!  

Because it's not coming from a breadth-first search, somebody had to have a specific reason to think that the experiment would work, and because there just aren't that many experiments being done at random, you can expect that to be a well-informed reason. The prior is higher.  

There's a similar sense in which lack of standardization is a blessing in disguise. Most mammal experiments today are done on the same handful of strains of inbred mice, for instance. The standardization is a boon to researchers in many ways (you can make apples-to-apples comparisons, you don't have to spend time inventing the basics of experimental methods yourself) but it also means that experimental results can just turn out to be an artifact of the "standard methodology."  Looking at older experiments, which have greater diversity in model organisms and other experimental methods, can be a corrective.

4.) Old papers are undervalued opportunities.

The author of the latest exciting result is a ready-made advocate for the discovery and a potential founder or collaborator for new ventures to put it into practice.  The author of an old paper, by contrast, might be dead or retired, with nobody to champion the potential applications of the discovery. It's very easy for an area of research to quietly fall out of fashion through no inherent lack of merit, just because it never met the right opportunity for application. 

I'm just barely old enough to remember when neural nets were thought of as an embarrassing phase in the history of computer science; they became "hot" again in 2012, with AlexNet, when newly affordable GPUs proved that deep learning algorithms could suddenly outperform the competition. In other words, advances in a totally different technology made a "failed" research approach into an overnight success. 

Going through old, not necessarily well-known experiments to see if there are opportunities is something I don't believe is being done that often, and is probably an unusually good place to apply a little bit of time and attention for big returns.

That said, let's look at some specific examples!  These are all results from prior to the year 2000, that are experimental interventions affecting vertebrate lifespan or aging, and which aren't currently the focus of a research program that I'm aware of.

Lowering Body Temperature: 71% Life Extension in Fish

Unusually long-lived vertebrates (tortoises, sharks, rockfish, etc, which can survive for centuries, or naked mole rats, which are extremely long-lived for their size) are frequently cold-blooded.  Warm-blooded animals which are long-lived (in absolute terms, like whales, or relative to their size, like bats, hummingbirds, and squirrels) often undergo temporary reductions in body temperature, during diving or hibernation.  Moreover, interventions like dietary restriction which extend lifespan have the effect of reducing body temperature.  So can reducing body temperature directly extend life?

For a cold-blooded example, transferring fish from 20-degree water to 15-degree water extended lifespan by 71%, in a 1972 study.[5]

To reduce the body temperature of a warm-blooded animal, it's not enough to reduce ambient temperature, since warm-blooded animals generate heat to compensate. In fact, reducing the ambient temperature actually shortens mouse lifespan. However, there are tricks to lower body temperature in a warm-blooded animal.

Mice genetically modified to overexpress the uncoupling protein, UCP2, in the hypothalamus have lower body temperature than wild-type[6], and they live longer than their wild-type counterparts (20% increase in female median lifespan, 12% in male.)

You can also induce hypothermia by stimulating the heat-detecting cells in the hypothalamus, either by injecting capsaicin [7], heating the hypothalamus directly with a thermode [8], or stimulating the heat-sensing neurons optogenetically [9].

The natural next experiments to do are a.) see if any of these other methods of inducing hypothermia affect lifespan and diseases of aging in mice or other mammals; b.) do longitudinal transcriptomics or other broad assays to see what reduced body temperature is doing and whether its effects can be simulated chemically or genetically.

Altered Photoperiod Cycle Length: Short "Years" Shorten Lifespan 30% in Lemurs

Days get longer in summer and shorter in winter; by lengthening or shortening the cycle of alteration in photoperiod length (by changing artificial lighting) you can give an animal a shorter or longer subjective "year".

This turns out to affect lifespan!

The gray mouse lemur is a prosimian primate, native to Madagascar, that is long-lived for its size. During the long-day summer, gray mouse lemurs breed and are more active; during the short-day winter, they gain weight, become lethargic, and don't copulate. If you alter the photoperiod cycle artificially, you can alter the timing of these behavioral and morphological changes accordingly -- and if you reduce the "year length" by a third, from 12 months to 8 months, lifespan also shortens by 30% and the onset of white fur happens 30% earlier. [10] . In other words, lemurs live 9-10 "subjective years", whether those are 8-month years or 12-month years. 

The obvious follow-up experiment is to go the other direction -- do lemurs (or other animals) live longer if you subject them to 16-month subjective years? And to take some tissue and blood samples and try to identify how this effect works -- do we see pathological changes, transcriptional changes, hormonal changes, metabolic changes?

Constant Light Exposure: 25% Life Extension in Hamsters

A Syrian hamster model of congenital heart disease showed delayed onset of heart failure and 25% life extension if they were kept in continuously lit conditions.[11]

The obvious corollary studies are to take heart tissue samples and blood samples and look for altered gene expression or metabolic parameters that might explain the effect of light exposure on preventing heart failure. It also might be possible to experiment directly with continuous light exposure on humans, since it's probably not dangerous.

Pineal-Thymus Graft: 24% Life Extension in Aged Mice

Implanting the pineal gland of a young mouse into the thymus of an old (16-22 month) mouse extends lifespan 19% in C57BL6 mice, 20% in Balb/c mice, and 35% in hybrid mice, for an average of 24% overall.[12] This is consonant with a more extensive literature about the pineal gland or the main hormone (melatonin) it secretes having a life-extending effect through preventing the dysregulation of the circadian rhythm which occurs with age.

The obvious follow-up study to do is a replication of the same implantation experiment, along with longitudinal expression data, to find out how this works and work towards identifying how a similar effect could be replicated by a less invasive intervention.

Splenectomy: 19% Life Extension in Aged Mice

In a 1969 experiment, adding spleen cells to mice of the same age as the cell donors shortened lifespan; adding spleen cells from younger mice (14 week) to older mice (76 week) extended median lifespan from 105 weeks to 128 weeks, a 13% lifespan effect; and removing the spleens of mice altogether at age 97 weeks increased median remaining lifespan from 118 to 158 weeks, a 19% lifespan effect.

Clearly, the aged mouse spleen contains some factor that accelerates age-related decline. The obvious question is to find out what this is, through expression or proteomics studies on young, aged, and splenectomized mice, and see if there's a way to target the culprit pharmacologically.

Induced Hypothyroidism: 17% Life Extension in Rats

Exposing newborn rats to thyroid hormone permanently reduces their bodyweight and thyroxine levels; it's a way of artificially inducing hypothyroidism.[13] It also has the effect of dramatically elevating their prolactin levels; as prolactin is stimulated by TSH release from the hypothalamus, clearly neonatal T4 exposure doesn't prevent TSH release in the brain, but rather impairs the thyroid's ability to respond to it.  This induced hypothyroidism also extends median lifespan by 17% and maximal lifespan by 6%.

Obviously, inducing hypothyroidism isn't a viable intervention for humans, but looking at changes in hormone levels and gene regulation in induced hypothyroidism might give clues to what downstream mechanisms are responsible for the lifespan increase and whether there's a less-side-effect-heavy way to induce it.

Castration: 17% Life Extension in Rats

Removing the testes of male Wistar rats has been found to extend lifespan 17% relative to unbred intact males; removing the ovaries extends lifespan 29% relative to unbred females. [14]

This isn't too surprising given that caloric restriction (a reliable life-extending intervention in rodents under typical lab conditions) has antigonadal effects, and that extremely dramatic lifespan effects can come from removing the germ cells in C. elegans.[15] There's also some correlational evidence -- for instance, castrated male cats arriving at veterinary hospitals lived 67% longer than intact males.[16]

Obviously, castration isn't a practical intervention for most humans, but it's possible that there's some downstream effect that doesn't alter fertility or observable sex characteristics and preserves some of the anti-aging effect; this is a good opportunity for looking at longitudinal expression changes in castrated vs. intact animals and trying to identify the mechanism of lifespan extension.

Lateral Hypothalamic Stimulation: 5% Life Extension in Aged Rats

Stimulating the lateral hypothalamus is pleasurable, and animals given the opportunity to self-stimulate will do so; this is what's known as wireheading.  Interestingly enough, there are interactions with aging here as well.  Young adult rats have more neurons and more electrical activity in the lateral hypothalamic area (LHA) than old rats; young rats also exhibit more self-stimulatory behavior than old rats when given access to a button that turns on the electrode. Moreover, in old rats, chronic stimulation in the LHA extended lifespan from 1075 days to 1125 days (5% of total median lifespan, 8% of total maximum lifespan, 35% of residual lifespan); stimulation reduced body mass as well.[17]

Is this just a dietary restriction effect, or is it something else? The natural thing to do is to try the experiment again, this time compared against controls given the exact same amount of food to eat; and also, to take brain samples after death and possibly other blood samples during lifespan to try to identify metabolic or regulatory changes caused by the stimulation.

Blindness: Increases Survival in Rats

Blindness affects the circadian rhythm; it effectively gives the same hormonal signals as perpetual darkness. Rats blinded at 25 days had increased lifespan relative to controls; at 748 days, when the experiment concluded, the blind rats had a 95% survival rate while the control rats had a 50% survival rate.

The natural follow-up is to do a full lifespan study so we can get an actual measurement of the effect on median lifespan, as well as measurements of other biomarkers so we can identify a mechanism and possibly a way to replicate the anti-aging effect without actually inducing blindness.

Fetal Hypothalmic Graft: Restores Fertility and Circadian Rhythm in Rats and Hamsters

In keeping with the pattern of neuroendocrine effects on aging, it turns out that transplanting the suprachiasmatic nucleus (the part of the hypothalamus responsible for entraining the circadian rhythm in response to day length) from fetal animals into the brains of aged animals can restore the periodicity of the circadian rhythm and restore diminished fertility. With age, circadian rhythms become less regular; animals wake more during the periods when they should be sleeping, and/or are more lethargic during the periods when they should be awake.  Fetal SCN grafts reverse this phenomenon in both hamsters [18] and rats.[19]

A natural type of question you might ask, if you're interested in understanding aging, is which genes are involved in the aging process. The practical upshot of identifying genes with causal roles is that they're potential drug targets.  If diseases of aging are caused or worsened by the excess of some protein, you might want to inhibit the production or activity of that protein. If diseases of aging are caused or worsened by a deficit in some protein, you might want to stimulate its production.

We have a lot of different kinds of experiments that can be run for identifying what genes and proteins are involved in aging, and thus a lot of aging-related "omics" studies. I'll briefly summarize a few categories I know about.

Longitudinal Transcriptomics

You can compare the expression of genes in tissue samples from old vs. young organisms (humans or mice) and see which genes are expressed more or less with age.  Today it's even possible to get single-cell resolution on gene expression; we can identify the rate of gene transcription for each gene in a specific cell at a given time.  Similarly, you can get proteomics data, directly measuring the quantity of each protein in a tissue sample.

This gives us correlational information about which genes are altered in the aging process, in particular tissues and cell types. It doesn't by itself tell us which interventions might prevent disease.  If a particular gene is more expressed with age, it could be because it's a cause of some deleterious process, or because it's a symptom of that process, or because it's part of the body's attempt to mitigate that process.  Whether you want to inhibit that protein's activity or production depends on what it's doing, and expression levels by themselves can't tell you that.

Comparative Genomics

Animals don't all age in the same way. Some are exceptionally long-lived, either in absolute terms (whales, elephants, tortoises, rockfishes, the Greenland shark) or relative to their size (bats, naked mole rats). Some mammals are immune from cancer.  Can we identify "genes responsible for healthy longevity" in the animal kingdom? Could we "borrow" the adaptations that slow-aging animals have developed, as treatments for humans?

Given the genomes for two species, you can identify homologous genes -- genes that have very similar sequences and probably similar functions.  Something like half our genes have homologues in common with insects and all vertebrates.  

If the homologue of a gene in an exceptionally long-lived species is absent or has a loss-of-function mutation, you might ask whether that gene contributes to aging.

If a gene family of similar proteins is "expanded" in a long-lived species (meaning there are more variations on that gene present) or if a gene has a high copy number (meaning there are many identical versions of that gene) you might ask whether that gene has a protective effect.

If a gene shows evidence of positive selection in a long-lived species, you might ask whether that gene has a protective effect against some aging process.

If there's a correlation between copy number or gene family size and lifespan (or lifespan per body weight) across species, that's somewhat stronger evidence that there's an association between those genes and lifespan.

Again, these are all correlational; they don't tell you how the gene works, or what would happen if you interfered with it experimentally.

Experimental Genetic Modifications

If you induce a genetic mutation in a mouse gene and the mouse lives longer or avoids the onset of age-related diseases, then you do have causal evidence that the gene is involved in regulating aging.

There are a variety of ways of inducing specific mutations, some permanent and some temporary, some causing total absence of the gene (knockout) while others cause a deficit (knockdown) or unusually high production of the gene product (overexpression.)

Looking for the Intersection

Most broad studies (longitudinal transcriptomics, comparative genomics) looking for genes involved in aging come up with totally different lists of candidate genes. 

Sometimes, of course, this is expected, because they're looking at different tissues or different organisms. You don't expect all animals or all tissues to change in the same way with age. And, of course, comparing genomes between species and comparing gene expression over time within one species are apples-to-oranges comparisons.

But even in cases where the experiments are supposed to measure the same thing, there's poor replication. And that's not surprising because the samples are so small. It's not uncommon to see longitudinal transcriptomics studies with fewer than ten organisms in the "old" and "young" groups. 

And this matters because if you want to have any hope of translation to humans, you've got to be able to have results that are consistent across different strains of mouse, or even species of mammal; if it's all wiped out by natural variation between organisms, there's no way you're getting signal usable for designing human treatments.

So I did a very, very crude type of meta-analysis; I looked at all the studies I could, out of these three types (transcriptomics and proteomics of aging; comparative genomics of aging in long-lived species; and interventional studies of genetic modifications; all restricted to studies on vertebrates) and ranked genes (or gene families) by the number of papers in which the gene popped out as significant. 

There are a couple of potential flaws in my methodology.

First of all, I used Google Scholar to search, and stopped when the relevant search terms stopped returning studies of the relevant type. There may well have been studies this search method missed; it's just much more time-efficient than using PubMed searches (which reliably produce far less relevant results, but it's easier to document exactly how many papers matched search terms, which is why they're the standard method in formal literature reviews.)

Second of all, I didn't use a consistent cutoff in picking out which genes were significant. (In a study that tests all 20,000 or so human genes for differential expression, "statistically significant" is a very low bar.) I generally noted down the handful of genes that had the highest fold change and lowest p-value, not literally all the ones that met a significance threshold.  

Thirdly, some studies, of course, like lifespan studies of a genetic modification, aren't unbiased screens of all genes; genes that have gained more scientific interest are likely to be studied more often, so in part this list of "top genes" reflects the biases of the research literature.

And, finally, since we're comparing different types of studies, we're not making apples-to-apples comparisons. You don't expect the genes differentially expressed with age to be exactly the same as the genes which are modified in long-lived organisms or the same as those which alter lifespan when experimentally mutated. If a gene shows up in all three types of studies I think that's some sort of evidence that it's "more likely to really be involved" in aging, but not in the same straightforward sense that it's true that a study is more credible when it replicates exactly.

However, I think it's worth doing something in this vein, as a way of helping orient ourselves in a growing field. As more and more papers come out claiming that they've found genes "associated" with aging, we want to be able to be familiar with what the most common ones that keep showing up are.  Just as with genome-wide association studies for genetic predictors of disease, one correlation showing up in a study doesn't mean we've found the "gene for" anything. I think of the aggregation process as a learning experience, for getting a sense of what the field as a whole looks like.

Top Genes

The following table is color-coded for the primary "hallmark of aging" associated with the gene or gene family; red for nutrient-sensing, purple for proteostasis, green for intercellular communication and inflammatory signaling, blue for DNA repair, and pink for mitochondria.