Willpower and Akrasia

You set an alarm for 6 AM to go running. At 6 AM, you snooze it. The person who set the alarm and the person who snoozed it are, in every meaningful psychological sense, different agents with different preferences occupying the same body. The entire problem of willpower is the problem of getting these agents to cooperate.

The Wrong Metaphor

The traditional model of willpower treats it as a muscle or a resource — something you have a finite supply of that gets depleted with use. This is the "ego depletion" framework, and while it captured something real about the phenomenology of self-control, George Ainslie's Breakdown of Will argues it gets the mechanism completely wrong.¹

Willpower isn't a muscle. It's a bargaining game.

Each of us contains a population of short- and long-term preferences that compete for control of behavior. The will to stick to a diet has the same fundamental structure as the "will" of nations not to use nuclear weapons. It's not located in a single place — it's a distributed network of cooperation among competing interests, held together by shared expectations about future behavior.¹

The key insight from behavioral economics: people systematically prefer smaller-sooner rewards over larger-later ones, following a pattern called hyperbolic discounting. Would you prefer $1,000 today or $2,000 in a year? Most people take the thousand. Would you prefer $1,000 in five years or $2,000 in six years? Most people take the two thousand. The time delay is the same — one year — but proximity warps preference. When temptation is far away, the deliberate self dominates. When temptation is at hand, the impulsive self takes over.¹

This isn't an exotic quirk. It's the fundamental architecture of human motivation. And it makes self-control look less like exerting force and more like maintaining a treaty.

Personal Rules as International Treaties

Ainslie's framework treats personal rules — "I don't eat sugar," "I go to bed by 10" — as self-enforcing contracts between your present and future selves.¹ Each time you sit down to eat, you're putting your credibility with yourself on the line. The rule works because you believe your future self will continue to honor it. Break the rule once, and the precedent makes it easier to break next time — not because you've depleted a resource, but because you've weakened the expectation of continued cooperation.

This is exactly the structure of mutually assured destruction. Nations don't use nuclear weapons because they expect other nations won't. The continued peace depends on mutually shared belief in continued peace. If these beliefs falter, the whole system can destabilize rapidly.

For individuals: "the threat that weighs on your current self's choice is the risk of losing your current stake in the outcomes that future selves obtain."¹ When you break a personal rule, you make your future self more likely to defect too, which means your current self has less reason to cooperate, which means... the diet collapses. Not with a bang, but with a cascade of eroding expectations.

This explains why personal rules feel so all-or-nothing. An exception isn't just one donut — it's a precedent. Your short-range interests know this, which is why they attack not by direct confrontation but by "proposing exceptions that might keep the present case from setting a precedent." The voice in your head doesn't say "abandon the diet forever." It says "just this once, because it's a special occasion." The genius of temptation is in reframing defection as not-really-defection.¹

Rules Without a Rulemaker

How do personal rules develop in people who don't explicitly set them? Ainslie draws a remarkable analogy to common law. English common law has no lawgiver and no written constitution — just accumulated precedent from individual cases. Similarly, the rules that govern your behavior often crystallized through repetition rather than conscious decision. What started as one possible approach to controlling a behavior becomes the approach after years of consistent practice.¹

This is how many of our deepest commitments work:

• People experience a rule against trying drugs as the belief that they'd immediately get addicted.
• Vegetarians experience a rule against eating meat as the belief that killing animals is morally wrong.
• People who avoid violent movies experience a rule against watching them as a belief that they're disgusted by violence.¹

The rule and the belief become indistinguishable. You don't feel like you're exercising willpower — you feel like you're expressing who you are. This is a feature, not a bug. A rule that operates at the level of identity is far more robust than one that requires moment-to-moment enforcement.

Religious commandments exploit this mechanism at scale. When a believer resists temptation, it strengthens their belief in God's help, making future resistance easier. Repeated success creates a self-fulfilling prophecy of divine assistance. God plays the same functional role that "future selves" play for secular people — a third-party entity that inspires cooperation across time. Jordan Peterson and Ainslie arrived at this conclusion independently: the discovery that gratification could be delayed was simultaneously the discovery of time, causality, and the concept of bargaining with the future.¹

Hyperbolic Discounting as Rational?

There's a counterintuitive argument that hyperbolic discounting — the very thing that generates akrasia — may be rational under uncertainty. Farmer and Geanakoplos showed that if future interest rates follow a geometric random walk (plausible in real markets), then hyperbolic discounting is actually the optimal strategy, outperforming exponential discounting dramatically over long time horizons.²

At 20 years, the two approaches give similar values. At 100 years, exponential discounting values the future at 0.020 while hyperbolic gives 0.051. At 1,000 years, exponential gives essentially zero (4 × 10^-18) while hyperbolic still assigns meaningful weight (0.005).² If this is right, it has enormous implications for environmental economics and any domain where very long-term consequences matter.

It also suggests that the "irrationality" of hyperbolic discounting might be a case of conflating two different kinds of rationality. Exponential discounting is optimal for a world with predictable, constant returns. Hyperbolic discounting is optimal for a world with uncertain, fluctuating returns — which is the world we actually live in. Our brains may be wiser than our economic models.

This connects back to Cognitive Biases and the sunk cost discussion: if hyperbolic discounting is rational under realistic conditions, then sunk cost behavior — which tends to counteract hyperbolic discounting by keeping you committed to projects your past self chose — might be doing useful work that looks irrational only when evaluated against the wrong normative standard.

The Marshmallow Illusion

The biggest experimental claim about willpower — that delaying gratification at age 4 predicts life outcomes — turns out to be another casualty of the replication crisis. Tyler Watts and colleagues ran the first serious replication of the marshmallow test with ten times the original sample size and found that the correlation between delay and later achievement was half as large. When they controlled for family background and early cognitive ability, it almost vanished.³

Worse, most of the test's predictive power came from whether children could wait just 20 seconds. Waiting 2, 5, or 7 minutes added nothing beyond the first 20 seconds. That's hard to reconcile with a story about sophisticated self-control strategies, and easy to reconcile with a story about background variables — intelligence, trust in institutions, home stability — that predict both marshmallow performance and later success.³

This matters for the Ainslie framework because it removes the main piece of evidence that individual willpower is a stable, measurable trait independent of context. If "self-control" is mostly a proxy for background conditions, then the practical levers aren't about training willpower but about building the kind of stable, trustworthy environments where commitment strategies can take hold. The bargaining model survives; the willpower-as-muscle model doesn't.

Attention as the Real Battleground

James Williams, a former Google advertising engineer turned Oxford philosopher, argues that modern technology constitutes a "denial-of-service attack" on the human will — not by blocking information but by flooding the attention system with micro-decisions that crowd out deliberation.⁴

He distinguishes three types of attention under siege. The "spotlight" is immediate task-focus, disrupted by notifications and clickbait. The "starlight" is your sense of who you want to be — the longer-term goals that give day-to-day tasks meaning — eroded by habitual distraction that makes you forget what you were aiming at. The "daylight" is metacognition itself — the capacity to define and evaluate your goals in the first place — degraded when you spend enough time in reactive mode that you lose the habit of reflective thought.⁴

The connection to Ainslie's framework is direct. If self-control is an intertemporal bargaining game, then anything that strengthens the impulsive near-self at the expense of the deliberate far-self shifts the balance toward akrasia. Attention-capture technology does exactly this: it routes decisions through the fast, impulsive system by design. Nir Eyal literally wrote a book called Hooked teaching designers how to exploit this. Snapstreak, infinite scroll, and outrage-bait aren't bugs in the attention economy — they're the product working as intended, systematically undermining the bright lines and precedent-awareness that Ainslie's model identifies as the foundations of self-control.⁴

The Ego Depletion Controversy

Scott Alexander's review of Baumeister's Willpower captures the state of the debate before the full collapse.⁵ The book's pitch was seductive: willpower is a limited resource depleted by use, restored by glucose, and — unlike IQ — trainable. The marshmallow test proved it mattered. The glucose studies proved it was metabolic.

Every piece of this has since been challenged. Robert Kurzban pointed out that the metabolic math doesn't work — the energy cost of short willpower tasks is negligible relative to blood glucose levels. Worse, merely rinsing your mouth with glucose (without swallowing) produces the same "repletion" effect, which mirrors findings in exercise science where mouth-rinsing with sugar improves endurance without providing any actual energy. And Carol Dweck found that ego depletion only occurs in people who believe willpower is depletable — a finding Alexander treats with appropriate skepticism.⁵

The most telling objection is Kurzban's reductio: if your willpower is truly "depleted," why can you still force yourself to work for a million-dollar reward? Money, Alexander notes, shares this property — you'll claim poverty and skip the restaurant, but if your mother is dying you'll find the money for a flight to California. Both willpower and money are budgeted, not truly exhaustible. The brain is running an opportunity-cost calculation, not draining a tank.

Alexander's own model — a stupid default system running on short-term reinforcement learning, plus an evolutionarily novel executive system that can overrule it at metabolic cost — has the virtue of explaining drugs. Adderall and modafinil don't increase glucose. They change the default system's behavior so the executive override isn't needed. This is consistent with Ainslie's bargaining framework: the drugs don't give you more willpower, they reduce the temptation your willpower has to fight.⁵

The Practical Upshot

The Ainslie framework doesn't make willpower easy. But it does make it legible. If self-control is an intertemporal bargaining game, then the levers are:

Bright lines. The clearer and more categorical the rule, the harder it is for your impulsive self to argue that "this case is different." "No sugar ever" is a stronger rule than "sugar in moderation" because there are no edge cases to exploit.

Precedent awareness. Treat each decision as setting a precedent your future self will observe. The donut matters less for its calories than for what it teaches your future self about your reliability.

Identity-level commitments. Rules that have been absorbed into your self-concept are nearly indestructible. The goal is to move commitments from "things I do" to "who I am."

Environmental design. Don't rely on willpower where you can rely on structure. Not stocking junk food in the kitchen is an external commitment device — it removes the bargaining situation entirely. This is where Extended Mind Thesis meets self-control: offloading the work of willpower onto the environment.

The self, on this view, is not a unified agent exercising control over recalcitrant desires. It's a coalition — "permanent roommates" competing for the right to embody you, held together by the shifting influence of available rewards. Disturbing? Perhaps. But also liberating: you don't need to be a single, perfectly rational agent to live well. You need your internal coalition to develop institutions good enough to keep the peace.