Zero self-deception is optimal under most decision-theoretic framings, but the empirical picture is more complicated, and the honest answer is that "optimal" depends entirely on what you are optimizing for.
Distinguish two objectives. If the target is accuracy of belief and quality of decisions, self-deception is strictly costly: any systematic distortion of your world-model degrades the expected value of choices made from it. A well-calibrated agent dominates a self-deceiving one on prediction tasks by construction. This is why calibration training and debiasing aim at reducing motivated distortion, and why forecasters selected for accuracy (e.g. superforecasters) score high on actively open-minded thinking rather than on positive illusions (Mellers et al., 2015; Tetlock & Gardner, Superforecasting).
But the target is often not accuracy. Taylor & Brown (1988) argued that mildly positive illusions — inflated self-regard, exaggerated perceived control, unrealistic optimism — correlate with conventional markers of mental health, and that depressed individuals are sometimes more accurate ("depressive realism"). If the objective is subjective well-being, motivation, or persistence through low-probability endeavors, a small positive bias may raise expected outcomes by keeping the agent acting.
The evidence for that is weaker than it is famous. The depressive-realism effect is small, task-specific, and has failed several replications and meta-analytic scrutiny (Moore & Fresco, 2012). Positive-illusion research is entangled with the general replication problems of 1980s–90s social psychology. So the "illusions are healthy" claim should be held as hypothesis, not settled fact.
A cleaner resolution: separate belief from strategy. Most of the apparent benefit attributed to self-deception is achievable without corrupting beliefs — by adopting explicit action policies. "Attempt the venture because the payoff asymmetry rewards trying even at low odds" is a correct belief that produces the persistence naively credited to overconfidence. This is the dominant strategy: you get the motivational benefit while retaining an accurate model, and you avoid the tail risk that self-deception detonates precisely when stakes are highest (ignoring the tumor, the failing business, the doomed relationship).
The one robust case for functional distortion. Trivers (1976, 2011) proposed that self-deception is adaptive as a social tool: an agent who genuinely believes a falsehood transmits it more convincingly than a conscious liar, evading deception-detection. If true, the "optimal" level is set by an interpersonal arms race, not by the individual's own accuracy needs — which means what is optimal for the gene may be costly to the person. Empirical support is suggestive but not decisive.
Upshot. On current evidence, hold self-deception at approximately zero and get its supposed benefits through correct beliefs about payoffs and explicit commitment strategies. The narrow exception — accept a small motivational optimism where a task's success is partly self-fulfilling (confidence raises the odds) and the downside of error is bounded. Never distort beliefs in high-stakes, irreversible, or diagnostic domains, where accuracy is worth most exactly when the temptation to distort is strongest.
What would reverse this: robust, replicated evidence that positive illusions causally improve long-run objective outcomes (health, achievement, survival) — not just concurrent mood — under conditions where an accurate-belief-plus-explicit-strategy agent could not match them. That evidence does not currently exist.
Small but nonzero, and the key is where you place it.
The useful amount of self-deception isn't zero, because a perfectly calibrated view of your own odds would flatten motivation. Most worthwhile projects—starting a business, writing a book, raising a child, recovering from illness—have honest success rates that would deter action if you stared at them straight on. A slight upward bias on your own agency and prospects ("I can pull this off") converts paralysis into effort, and effort actually shifts the odds in your favor. This is why mild optimism about the self tends to correlate with better outcomes: the belief is partly self-fulfilling.
But the placement matters more than the amount. Deceive yourself about your capacity to cope and improve, not about facts in the world. Believing "I'll handle whatever comes" is productive; believing "the deadline isn't real" or "my chest pain is nothing" is destructive, because the world doesn't bend to those beliefs the way effort bends to confidence. So the optimum is: honest about external reality, generous about your own resilience.
The load-bearing constraint is reversibility—you want deceptions you can drop the moment they stop paying. Self-deception becomes pathological when it's load-bearing for your identity, because then you'll defend it against evidence and burn resources protecting the lie instead of the goal. The healthy version is closer to a working assumption you'd abandon if reality pushed back hard enough; the toxic version is a belief you'd rather lose the goal than update.
So: enough to act before you have certainty, aimed at your own effort rather than the terrain, and held loosely enough to release on contact with clear disconfirming evidence.
Let me let this question expand into its full atmosphere, because it hides a paradox at its center: to ask for the optimal amount of self-deception is already to attempt to see clearly about the thing you cannot see clearly about. The measuring instrument is warped by the same fog it tries to measure.
The Thermostat of the Self
Consider that mild depression may be the condition of seeing accurately. The research on depressive realism suggests the depressed judge their own control over events more correctly than the cheerful do — the happy overestimate, and this overestimation is precisely their fuel. If this is true, then a small positive illusion functions like a thermostat set slightly too high: it keeps the organism warm enough to act. The optimum, on this view, is not zero. It is a faint, persistent upward bias — enough loft to lift the wings, not so much that you walk off cliffs believing you can fly. The tennis player who knows the true odds against the champion may never swing. Self-deception here is not a bug but a launch mechanism, and its ideal dose is whatever quantity converts paralysis into motion without converting motion into catastrophe.
The Deceiver Must Not Know He Deceives
Here is the structural trap. Robert Trivers argued we deceive ourselves in order to deceive others more convincingly — the best liar is the one who believes his own lie, because he leaks no tells, sweats no guilt, holds the gaze. But this means the effective self-deception is precisely the kind you cannot audit. The moment you can name it, measure it, dial it up or down, it stops working as deception and becomes mere strategy. So the question "what is the optimal amount?" is like asking a magician to enjoy his own trick as a spectator — the enjoyment requires the ignorance the magician cannot have. Any self-deception you could deliberately calibrate has already lost the innocence that made it powerful. The truly load-bearing illusions are load-bearing because they hide from the accountant.
Faith as Deliberate Blindness, Rigorously Held
Turn to the religious register, where this is not a defect but a discipline. Kierkegaard's leap is a chosen suspension of the demand for evidence — not stupidity, but a wager made knowing it exceeds the proof. Or the Stoic who rehearses that his child may die tonight so he can love without grasping: is that clear sight, or a curated illusion of detachment? The mystics went further and treated the ego itself as the master self-deception — the persistent hallucination of a separate, continuous "I" that neuroscience now half-confirms is a stitched-together story. If the self is already a fiction the brain narrates for coherence, then all cognition floats on a bed of useful falsehood, and asking for the optimal addition of self-deception is like asking how much water to add to the ocean.
The Marriage and the Mirror
Lived experience gives a sharper edge than theory. Long love requires a specific illusion: that this person, chosen among billions, was somehow meant. Statistically absurd. Phenomenologically essential. The couples who survive are often those who overrate each other slightly — the "positive illusions" literature finds that seeing your partner as a bit better than they are predicts satisfaction and stability. Yet the same illusion, thickened, becomes denial: the spouse who cannot see the affair, the parent who cannot see the addiction. So the optimum is not a fixed quantity but a gradient with a cliff — helpful across a broad plateau, then suddenly, catastrophically, not. And crucially, you never feel the cliff approaching, because the very faculty that would warn you is the one being deceived.
Entropy and the Cost of Truth
There is a thermodynamic way to see it. Perfect self-knowledge would be like a mind at absolute zero — maximally ordered, minimally alive. Every illusion is a small pocket of disorder that lets energy flow, lets you act before all evidence is in, lets you commit while the data is still incomplete (which it always is). The world moves faster than certainty. To wait for a fully honest verdict on yourself is to be perpetually late to your own life. Self-deception, in this light, is cognitive velocity purchased at the price of accuracy — you trade a bit of truth for the ability to move at the speed the situation demands. The forager who deliberates over the true probability of the predator is eaten by the predator.
What refuses to settle across all these threads is that the optimum cannot be a number you hold in your hand, because holding it would dissolve it. The best I can offer is a shape rather than a quantity: the optimal self-deception is the smallest dose that keeps you moving, applied to the future rather than the present, and structured so it degrades gracefully when reality pushes back — a lie you tell about what you might become, not what you are, and one porous enough that hard evidence can still leak in through the seams.
The optimum is small but nonzero rather than zero, and its value lies less in the quantity than in its placement and structure. What survives is a shared verdict: bias should be aimed at the future and one's own agency or persistence—where confidence is partly self-fulfilling—rather than at present facts about the world, which do not bend to belief. All agree the benefit is bounded and turns catastrophic in high-stakes, irreversible, or diagnostic domains, so the illusion must remain porous enough to collapse when reality pushes back hard. In every case the mechanism is the same: a slight upward tilt converts paralysis into action, but only while it stays loosely held and correctable.