Like the fruit of our loins, our temporal progeny are often thankless. We toil and sweat to give them just what we thing they will like, and they quit their jobs, grow their hair, move to or from San Francisco, and wonder how we could ever have been stupid enough to think they’d like that.

Why are we so bad at predicting what will make our future selves happy?


All brains make implicit predictions about the immediate, local, personal future - “last time I smelled this smell, a big thing tried to eat me => a big thing is about to…”. But these are short-term, limited, automatic and largely unconscious - author calls this nexting rather than predicting.

When nexting is wrong, it generates surprise which has several involuntary physiological expressions. We can use these to test what kinds of predictions nexting is making. This is how we know that eg babies already have a mental model of kinetics.

Long term predictions are more interesting. Humans seem to be the only animals that can make them. We can not only make plans based on what we want now, but what we predict we will want in the future.

Most of the expansion in brain size in homo sapiens was in the frontal lobe.

People who have frontal lobe damage (eg after icepick lobotomies) are capable of nexting but not of making plans for the future. Similarly, they can solve arithmetic puzzles but not mazes.

Icepick lobotomies have a similarly ‘calming’ effect, hence their brief popularity.

Thread unifying planning and anxiety is the ability to think about the future. Humans spend a lot of time thinking about the future, almost involuntarily. (Mindfulness is essentially practice in not making temporal simulations for some stretch of time.)

Predicting the future has obvious benefits for planning, but there are also emotional side-effects. Spending lots of time simulating good future outcomes is pleasant, but leads to higher predicted likelihoods of those outcomes (availability heuristic) which sets us up for disappointment. On the other side, predicting unpleasant events can provide a vaccination effect - pain that takes us by surprise usually produces stronger reactions.

Humans also have a desire for control - to be the cause of change in their environment. The perception of control is important for mental health. The author cites several experiments which are not particularly convincing in themselves, but it’s probably also worth following up the citation.


The pursuit of happiness is generally reported to be the reason behind most decisions. But how to measure happiness?

“After a day spent killing his parents, Frank was happy”. Philosophers like to distinguish ‘true’ happiness from base pleasure, the former being a product of virtue and a live well lived. But that requires taking the uncomfortable position that, even if Frank claims he is happy, and his brain scans agree, somewhere deep down inside he is not. Better to leave virtue out of it.

Gives the example of conjoined twins who claim to be happily enjoying their lives. One might doubt whether they are as happy as the average person, but there is no way to directly compare the experiences.

“I’m not happy, but I’m happy that you are happy.” Not every use of the word ‘happy’ means emotional happiness. Can also refer to approval of an idea, expectation of happiness etc.

Can’t even compare two experiences over time within the same subject - memory is too unreliable and prone to interference. Worse, new experiences alter our minds and we are unable to see the world as we did before, making objective comparisons impossible.

Language-squishing hypothesis - when the twins rate their happiness as 8, they are feeling what other people feel when they say 4. Their scale is squashed because of the paucity of their experience.

Experience-stretching hypothesis - when the twins experience an event that causes a rating of 8, they feel as happy as other people do when they experience an event that causes a rating of 8, even though other people need much more extreme events to reach the same rating.

Latter hypothesis seems more sensible when we consider changes in reaction to events over the lifetime of an individual:

If impoverished experiential backgrounds squish our language rather than stretch our experience, then children who say they are delighted by peanut butter and jelly sandwich are just plain wrong, and they will admit it later in life when they get their first bite of good liver, at which time they will be right, until they get older and begin to get heartburn from fatty foods, at which time they will realize that they were wrong then too.

Can confuse fear with attraction, apprehension with guilt, shame with anxiety eg bridge experiment. Anxious reappraisal seems to be the deliberate confusion of anxiety with excitement. Implication is that you can be wrong about what you are feeling and why.

Phenomenon of suddenly realizing you have not been paying attention to what you are reading. The text is familiar, your eyes have just scanned it, your brain processed the text - you experienced reading. But you weren’t consciously aware of that experience. The distinction is a little muddy. Computationally, seems like the part of your brain that decodes text was operating fine, but the part that pays attention to the result was elsewhere. Seems similar to subliminal messaging, where there is no conscious experience but the text affects behavior so must have been linguistically decoded.

Blindsight - sufferers have vision but not conscious awareness of vision - they cannot see but they can catch a ball thrown to them. Also the inspiration for one of my favourite sf books.

Alexithymia - inability to describe emotional state. Patients have normal physiological reactions to eg disturbing pictures, but are unable to distinguish them from normal pictures.

So, how can we do science to happiness if we can’t measure it?

After all, what are the odds that everyone misremembers banana-cream pie as better and coconut-cream pie as worse than they really were?

This is a book on cognitive science - that’s exactly the kind of weird effect I would expect to read about! Maybe the coconut-cream pie had a lingering aftertaste that soured the memory of the actual eating experience. I don’t think the author would miss that in practice, it just seems to be an unfortunate choice of example.

Measuring happiness is important, because it’s pretty much the terminal value.


“What would it feel like if…” is subject to systematic error.

Memories are very lossily compressed down to key facts. When recalled, they are decompressed, filled in and recompressed - that is, recollection changes memories. Plenty of experiments on eg eyewitness recall that demonstrate this. The illusion is completely opaque - subjects can be given vivid memories of things that they did not in fact experience, even if they are warned beforehand about the trick, and they will vigorously defend those memories.

Similar filling-in-the-details illusions happen for vision and hearing too. Eg replace a letter with a cough in “the [cough]eel was on the orange” and “the [cough]eel was on the shoe” and subjects will hear the cough but also hear the missing letter, even though their brain can’t even know what the letter should be until the end of the sentence!

Naive realism.

Children begin as realists. Confusing small objects with far away objects is confusing perception with reality. Similarly for confusing own knowledge with global knowledge (eg smarties experiment). In similar experiments for adults, eye and hand tracking reveal that subjects initially make realist mistakes but then correct/override. Implies that realism doesn’t go away, we just learn to distrust it in certain situations.

Imagination. Can construct complex objects without effort. Eg ‘imagine a plate of spaghetti’ leads to a rich mental image, but that image contains lots of detail that neither the description or your conscious mind provided eg school-cafeteria slop vs high-end restaurant. When I ask you to imagine how happy spaghetti for dinner would make you, you base your assessment off that image without correcting for the fact that it is not a representative imagining. All the extra detail sneaks in unasked for and unquestioned.

To repeat, when making predictions about a family of events we simulate one member of that family, with no insight into how that member was selected, and we don’t notice that we did that. Attribute substitution strikes again!

Cute experiment: Ask subjects to make predictions about future events. One group is asked to describe all the details they imagined and told to assume those details are all perfectly accurate. The other is told not to make any assumptions about the details beyond those specified. Then both groups were asked to rate their confidence in their prediction. Both had the same level of confidence in their predictions.

Does this imply that methods from Superforecasting or similar might help improve personal predictions?

Test subjects are much better at discerning patterns marked by the presence of some attribute than patterns marked by the lack of some attribute. Similarly, when attempting to detect correlations people often fail to look for all combinations of X&Y, X&~Y, ~X&Y, ~X&~Y. __And similarly missing P(X~Y) for Bayes rule.__

When asked to pick pairs of similar countries, subjects are focused on similarities. When asked to pick pairs of dissimilar countries, subjects are focused on dissimilarities. Often leads to both groups picking the same pair of countries. Similarly, people tend to focus on positives when picking options, and focus on negatives when choosing which option to give up.

If asked how they would feel two years after the death of their child, most people only imagine the devastation, and not any of the other details of their lives.

Students who were asked to describe the details of a typical day were then better at predicting how happy they would be after their football team won/lost the match tomorrow. Forced them to focus on the entire day, not just the one moment in their imagining.

Californians no happier than average, yet most Americans believe they would be happier if living in California. Tend to focus on easily-available images like sunshine and palm trees rather than traffic and house prices.

Similarly, tend to underestimate happiness of people with disabilities because we forget that they still have an entire life going on. “Blindness isn’t a full-time job”.

Tend to imagine more details for events in the near future rather than far future. Eg asked to imagine a good day tomorrow, subjects come up with more variety than if asked to imagine a good day next year. The good day tomorrow contains a more realistic mix of good and bad moments, but both groups of subjects rate the realism of their vision the same.

Vividness of the near future leads to weird time discounting eg subjects choose $20 in 365 days over $19 in 364 days, but choose $19 today over $20 tomorrow. This doesn’t seem irrational to me - of all four options, $19 today is the only one that doesn’t require making an extra trip back to the lab.


Historical depictions of the future are always adorable because they predict that things will be more or less the same, just shinier and with more hovering.

Wilbur Wright once predicted that “man will not fly for 50 years”, 2 years before his first successful flight.

As it turns out, when brains plug holes in their conceptualizations of yesterday and tomorrow, they tend to use a material called today.

Predictions of the future and memories of the past are heavily filtered through current mood. Subjects remember that they always had similar political opinions, feelings about their romantic interests and levels of head pain as they do right now. People who are currently full buy less food when shopping for the coming week. Subjects before a quiz predict they will choose a candy reward over the answers, but are consumed by curiosity after the quiz.

When trying to picture a penguin or remember a song, the visual/auditory areas of the brain are used to simulate the experience. Similarly, we make predictions about emotions by simulating the event. Someone asked how they would feel about a tragic event displays muted versions of the same physiological signs they would show if it actually happened.

We close our eyes or plug our ears when imaging because current sensations have higher priority than simulation. When imagining feelings, we can easily confuse current feelings for the results of the simulation.

We never confuse imagined scenes with actual vision, or imagined songs with actual sound. The same is not true for emotions - we happily confuse current emotions with future predictions. So depressed individuals find it hard to believe anything is worth doing because whenever they simulate something they feel depressed and attach that feeling to the simulation rather than their current state.’s a lot like trying to imagine the taste of marshmallow while chewing liver.

The direction of the timeline changes across cultures, but all cultures reason and talk about time using spatial metaphors.

Variety (eg in meals) is preferable across short time periods because it defeats habituation. Across long time periods (eg choosing Sunday meals for several months in advance) subjects tend to also vote for variety, but in-the-moment reports show that subjects who get their favorite every time are happier on average, because there is enough time between each event to avoid habituation. Author argues that this mistake is made because subjects render the scenario spatially as a series of meals. This seems like gross speculation.

Imagination is generally atemporal. We construct images, sounds, smells, tastes but not times. When asked to imagine an event at a specific time, eg 50 years in the future, subjects initial physiological reaction is the same as if asked to imagine it now, and only shortly afterwards do they correct the image to their 50-year-older self.

That first reaction appears to anchor their final prediction. Subjects ask to predict how much they would enjoy spaghetti in the morning vs in the afternoon made reasonable predictions. But subjects asked the same question whilst simultaneously performing another tasks answered much lower if they were full or much higher if they were hungry - anchoring predictions against their current reaction.

We judge situations by comparisons to present or past states, rather than by absolute value.

Imagine that you have a $20 bill and a $20 concert ticket in your wallet, but when you arrive at the concert you realize that you’ve lost the ticket en route. Would you buy a new one? Most people say no. Now imagine that instead of a $20 bill and a $20 concert ticket, you have two $20 bills in your wallet, and when you arrive at the concert you realize you’ve lost one of the bills en route. Would you buy a new one? Most people say yes.

I’m unable to imagine getting all the way to the concert and deciding to give up and go home, even for much more expensive tickets.

We similarly mistake the range of presented options for the range of possible options. Our choices can be distorted by adding or removing options that we wouldn’t have picked anyway, because they affect the comparisons we make.

So to accurately predict our future experience, we have to predict the kinds of comparisons we will be making in the future. My 10-year-old self would probably predict that having enough money to buy all the video games I want would be the cause of never-ending happiness, not realizing that future me would be comparing any gaming time to a much wider range of activities and finding that it comes up short. Eg buying massive speakers because they sound better compared to other speakers in the shop, and then getting home and comparing the massive ugly speakers to your elegant, subtle apartment.

Loss aversion.


We disambiguate stimuli using signals such as context, frequency and recency.

Necker cube. If one orientation is rewarded by researchers, subjects start to see that orientation more often and hold on to it for longer. Or claim to?

Most subjects think of themselves as talented, but give definitions of talent that match their own abilities more than others.

After committing to an option, subjects view it more favorably.

In the wake of negative events, a healthy psychological immune system must strike a balance between living in a fantasy world and being depressed about the real world. Defended rather than defenseless or defensive.

Result is we won’t outright lie to ourselves, but we will distort or selectively sample the facts. Confirmation bias, selective recall etc.

Bar for evidence is set higher for unfavorable facts.

Clever Hans. Easy to fool ourselves when we want to believe.

Conscious self-delusion appears to backfire, perhaps by being too obvious. Need to feel like we discovered some new perspective rather than creating it.

Subjects asked to imagine pain of being rejected by judge vs unanimous jury did not correctly predict that the jury would be worse (as judged by subjects asked after the fact). They failed to predict being rejected by an individual would be easier to rationalize away.

Replicated across a range of subjects - people expect that having someone to blame rather than dumb luck should not have any effect on the sting of the event, but in fact it gives their psychological immune system something to work with.

Similarly, people expect to regret foolish action more than foolish inaction but the reverse is true across a wide range of studies. !!! Much harder to rationalize inaction. Why?

Immune system has an activation threshold. Mildly bad experiences can actually cause more damage than awful experiences, because they don’t trigger any defense. (This makes predicting future emotional experiences difficult, because you don’t know whether something will be bad enough to trigger rationalization.)

Similarly, committed choice is more likely to trigger defenses than choices that can be changed. Given the choice, people often prefer to leave their options open, but this can actually lead to less happiness.

We are not purely associative creatures - we can explain the causes of negative/positive events and use those explanations to guide future choices in novel scenarios eg getting motion sickness on a ride might lead to a basic aversion to rides, but also to a conscious aversion to bungee jumping and sailing, but not music and clowns. We can isolate the essential causes rather than gaining aversion to all related signals.

Writing an explanation of a negative event can reduce its impact. Can similarly ruin positive experiences if eg we learn that a compliment had an ulterior motive instead of being spontaneous. Unexplained events linger longer in our minds as we keep poking at the mystery.


Why don’t we get better at this over time?

Availability bias. Leads to poor predictions based on outliers that generated strong emotions.

Emotional memories are highly biased by the ending of events. Good reason to end events on a high note rather than letting them drag on till they dissolve. But this bias disappears when events are compared side by side.

Memories of emotions can be distorted by current beliefs/experiences/theories. Eg priming subjects to think about gender before answering questions causes men to remember feeling less emotional and women to remember feeling more emotional. Eg subjects memories of their own emotions of an event were more stereotypical for their gender than their recorded responses at the time.

It’s hard to learn from experience when we can’t correctly recall that experience.

Memetics - beliefs don’t have to be accurate to propagate. Beliefs like ‘money leads to happiness’ and ‘children are the greatest joy in life’ are good for society but bad for individuals. People are bad at disbelieving them despite the evidence that eg raising children leads to lower life satisfaction. It seems like most of the biases discusses vanish during side-by-side comparisons, which individuals can’t easily make for such major decisions. We’re reliant on statistics instead, which have much less emotional impact.

Reference class forecasting. We aren’t unique snowflakes, and our prediction abilities are so bad that basing predictions on other people who are in the same circumstance can be more useful than relying on our own self-knowledge. Eg in many cirucmstances peoples predictions of how random strangers will feel are more accurate than their predictions of how they themselves will feel.

But it’s really hard to get people to use this. Most people think of themselves as substantially less biased than the average person, so obviously they don’t need it.


Memories of emotional state are subject to systematic distortions, caused by the reconstruction of incomplete memories and by self-serving biases. Prefer contemporary records whenever possible. (I should start using a mood tracker).

Predictions of emotional state are subject to systematic distortions, caused by inaccurate simulations of the future and incomplete understanding of our own minds. Use reference-class forecasting if possible - prefer actual records of how other people felt in similar situations over your own imagination. Otherwise:

Regrets over inaction are much harder to rationalize than regrets over actions, and will sting for longer.