J’ai finalement terminé un premier roman court dans le cadre de NaNoWriMo. Je voulais écrire cinq nouvelles de 10 000 mots chacune, mais en fin de compte la première fait 37 000 mots et 122 pages, ce qui est presque un roman j’imagine. Un lien au ficher PDF se situe à la fin du post – j’ai fait une relecture avec légers correctifs, donc disons que c’est un “deuxième jet”.

J’apprécie tout feedback, bon comme mauvais. J’ai tout de même quelques indications sur ce que j’aimerais idéalement savoir:

• Je n’ai pas tendance à bourrer mes textes d’action. Est-ce que le rythme est trop lent? Est-ce que ça manque de rebondissements ou de drame?
• J’aime bien faire des descriptions et des dialogues philosophiques. Il y en a beaucoup dans le texte. Est-ce que c’est compréhensible et agréable à lire?
• L’histoire est-elle crédible? Est-ce que mes explications tiennent la route?
• Est-ce qu’il y a des moment où on “décroche”?
• Je voudrais insérer quelques passages “journal” vers la fin, ou transformer des passages existants. Suggestions?
• Y a-t-il des choses qu’on ne comprend pas, ou des questions qui restent sans réponse (sans que ce soit intentionnel)?
• L’ordre dans lequel on découvre les éléments de l’histoire est-il engageant, ou gagnerait-il à être amélioré?
• S’attache-t-on aux personnages? Sont-ils suffisamment développés?
• Si vous connaissez l’allemand… les trois ou quatre lignes qui sont en allemand sont-elles correctes?

Envoyez-moi vos commentaires à monprénom@monnomdefamille.net. Au moins, dites-moi si vous avez aimé ou non. Ça m’aiderait vraiment de le savoir: je sais que diverses personnes ont des goûts différents, et j’aimerais avoir une idée du public exact que je peux rejoindre

Alors, voici: Le cube

Second, this month I am doing Nanowrimo. In short, the goal is to write 50,000 words of fiction in November (which amounts to an average of approximately 1,667 words per day). I have decided to write short stories, but the first one, which I am writing right now, is already 13,000 words and it is not nearly finished (I expect it to be 20,000 to 25,000 words long, but I don’t know). I am keeping pace and am confident to succeed. I write the stories in French and will post them here when they are finished, for the benefit of those who can read them.

Does voting matter in the short term?

It depends on how you look at it. In general, the effect of a vote is inversely proportional to the size of the voting pool. A rational agent would weigh the potential impact of their vote against the cost of voting (getting up, walking/driving to the precincts, etc.) Most of the time, the costs probably outweigh the impact, and thus it’s not worth voting. Now, you might argue that democracy wouldn’t work if everybody did that, but it is fallacious to argue under the assumption that other people follow the same train of thought that we do: they do not, and it is fairly clear that humans in general are not rational agents. Furthermore, nobody argues this on a whim an hour before voting – if the estimated voter turnout is 50% a week before the election, it won’t fall to 10% overnight, and if it does, a rational agent will remember this and will try to take advantage of the low turnout on the next election. Basically, if you know that 80% of people in your county will vote, and that they will massively vote for candidate X, just stay home. Really. Don’t worry, most people won’t change their minds.

But wait! There might be another reason to vote, one that you did not think of:

Does voting matter in the long term?

Yes. Overall, it might matter more than it does in the short term, and it’s a pity people do not see this. In an election, you are focused on its results, and that’s understandable. But that does not mean you cannot look into how your vote might impact future elections. Because let’s face it – an election is not a matter of life or death, and by always looking at the immediate future, you merely ensure mediocrity for the next fifty years.

A vote is mind share. Even though a small upstart party has absolutely no chance of winning the current election, voting for it is not a complete waste, to the extent that it places it on the map. People have a limited attention span: the more major a party is, the more likely it is that people will look at it and consider voting for it. Most will only look at the parties that have a chance of winning, plus maybe one rising party. Some will also look at the one or two next-largest parties, that might be large enough to have some credibility. And then some will look at everything. But in all cases, there is a list of parties or candidate, ordered by how well they did in the last election, and their mind share is roughly proportional to that. By voting for a party, you help them rise in that list.

In other words, imagine that 6% of voters like party X. With strategic voting, it gets, say, 3% of the vote. If no strategic voting occurs at all, it gets 6%. Next election, instead of being 3%, the baseline will be 6%. A higher baseline means that more people will consider it and the party will become significant much faster than it would have otherwise. Indeed, with a 3% baseline, maybe 7% of the voter base will like the party next election. With a 6% baseline, it might be 10%, because more people will pay attention to it.

You see, if you want to change the world, that’s how you do it.

Does the participation rate matter?

No. What matters is that voters are an unbiased sample of the population, with reasonable standard errors. A higher participation rate can reduce variance on the result, but if the sample is biased, that’s not even a good thing. To illustrate, imagine a population where 50% of all people lean left and 50% of them lean right. Now, let’s imagine that for some reason left leaning people are slightly more likely to vote than right leaning people. Therefore, in an election, maybe 90% of lefties will vote, and 80% of righties will vote. The tally will be 53% left, 47% right, and the left will win. Over the course of many elections, the left will almost always win, because there is too little variance for the right to get in even once. Now, if you picture an election where only 5% of lefties vote and 5% of righties vote, the victories will be evenly shared between the two sides, which seems much fairer, despite a measly participation rate.

In fact, an election where, say, one to five percent of the population is chosen strictly at random and forced to vote, would probably be more reliable than the current system. It doesn’t really matter much how many people participate, as long as the result is unbiased.

Should the side with the most votes win?

Counter-intuitively, only most of the time. Current democracies only look at the short term, which induces a temporal imbalance: in a society where 55% of the population is on a side, and 45% is on the other, it seems bizarre that the first side would win all the time, rather than 55% of the time. While there is admittedly much more variance in such a process, picking each side with probability equal to its vote share would lead to more representative governance in general. It also eschews trying to come up with a better voting system: all voting systems work from the flawed premise that one should figure out each result with perfect accuracy, even though in fact such a system will not, on average, do better than a stochastic process (in fact, I would argue that no deterministic voting system has theoretical properties that are as good as a stochastic one when you take temporal bias into account).

The main problem here is that people wouldn’t understand such a system: it is natural for human beings to place very inflated importance in particular results, and to loathe variance. People will get angry losing a game because of luck, even though luck always averages out on the course of many games. Why? Because they want to win every single time they believe they deserve it. Obviously, nobody would want to lose a campaign for congress, senate or even presidency on a die roll. And yet, democracy would probably work better if we did that, if only because it would give minorities a louder voice and push us to try new things once in a while. A deterministic system is consistent, but that only means it is consistently mediocre.

Basically, as far as democracy goes, the accuracy of each individual result seems to be a sacred principle, even though there is no mathematical need to go beyond being unbiased. It is also questionable whether the strategy to always reward the front runner works out for us in the long term. Unfortunately, human nature is such that alternatives are unthinkable.

Who should win if the tally is very close?

Just flip a coin. You know how fickle voters are: when an election is very close, that basically means that the candidate who won today would have lost tomorrow, and the one who would have won tomorrow would have lost the day after that. I know that democracy is based on sacred principles, but in evaluating the merits of a system, you have to look at the results. As I have said, what underpins democracy is the understanding that its results are unbiased on average, not the understanding that they are exact. It’s important for morale to make sure that the ballots are counted properly, but don’t fool yourselves into thinking it really matters if they are. As long as there is no fraud (which would induce a bias towards electing corrupt people, and we definitely don’t want that to happen), it doesn’t matter at all. When an election is extremely close, in general, there is no evidence that either side will do better or worse than the other. Besides misguided rigid principles about how democracy should work, a coin flip would work just fine, perhaps even better, since a fraudulent candidates has an edge under the current system.

Assuming that I hold that X is true and that you hold that X is false, I would argue that:

• If both you and I agree on a system to assign probabilities to statements given evidence, then whoever holds that the least probable statement is true has to find evidence in order to boost that probability. The other doesn’t have to do a thing.
• If we cannot agree on such a system, then we need to find one that we agree on.
• But all systems that work behave in a roughly similar way, which I will describe further down.

Positive and negative statements

In my terminology, a positive statement is a claim that a particular state of affairs obtains. For instance, “there is a pen in front of me” is a positive statement. By contrast, a negative statement is a claim that a particular state of affairs does not obtain. For instance, “there is not a pen in front of me” is a negative statement, and in fact from any positive statement you can construct a negative statement, and vice versa.

Now, what I propose is this: in general, positive statements are false. Conversely, in general, negative statements are true. The argument behind this is as follows: first, the number of things that could possibly obtain is much, much larger than the set of things that do obtain. Consider that regardless of the size of the universe, the number of possible states it could take is exponential in it (and regardless of how infinite the universe is, the number of possible things it could contain, by diagonalization, can be shown to be even more infinite). It follows that only a measly percentage of all possible positive sentences can possibly obtain. Second, in absence of evidence, it does not make any sense to favor a proposition over another: if you were sequestrated in a dark room all your life, horses and unicorns are just as probable. It’s acceptable to organize probability systemically, for instance, with larger probabilities assigned to simple things, but nonetheless there is no way an arbitrary, particular idea about what the universe contains could be expected to be true without evidence. Since a negative statement is the negation of a positive statement, it follows that only a measly amount of them are false.

Example: theism vs. atheism

Let’s take a simple example: assuming that there is no evidence that he exists, how probable is it reasonable to think that God is? Well, if we define God as a “unique, sentient, omniscient, omnipotent, omnibenevolent being who created the universe”, then we’re already down to a probability of (100% / 32) = 3.125%. Indeed, there are 32 possible incompatible combinations of these attributes (assuming we keep “unique”). Maybe a “sentient, non-omniscient, omnipotent, non-omnibenevolent being who created the universe” exists. According to our definition, that is no God. Maybe it didn’t create the universe. Only one out of 32 possible combinations can be called a God, and there is no reason to consider it more probable than any other. Now, how many possibilities does atheism cover? Technically, it only denies that God exists, so it covers 31 of them. Any evidence put aside, blind atheism is thus correct with probability 97%.

Now, maybe you would say that atheism rejects all 32 of these possibilities, but that was a very rudimentary example and you’re missing the larger point. The point is that the concept of “God” or “supreme being” has some precise meaning that must be disambiguated against other, competing possibilities. You have to expend information to describe such a thing (say, in a dictionary), and that information could have been used to describe something else. But how many things can be described as succinctly as a supreme being? Many, many things, none of which are supreme beings. And the point is that unless you indulge in special pleading to boost a supreme being’s “probability”, that probability ought to be inversely proportional to that number. If I can describe ten unfalsifiable non-theistic possibilities, theism is already down to 10%.

In roughly the same time you take you describe “our universe, created by God”, I can describe “our universe, simulated on a computer in another universe”, or “our universe, run by a Turing machine on a Game of Life grid”, or “our universe, created by colliding space balls”, replacing the concept of God by whatever whimsical concept I can think of. An atheist only states that he or she rejects that our universe was created by God, and even though they might reject all the other things I mentioned, they also might not, because that’s beyond the scope of “atheism”. There is a staggering amount of possibilities that don’t involve a God at all and against which theism has to compete, and that’s a clear probability sink. That is why the existence of God is extremely unlikely, until evidence is shown that boosts its likelihood with respect to other things.

False dichotomies

The perception according to which theism is probable, or is assigned the very naive probability of 50% stems from a false dichotomy: basically, that God existing is one distinct possibility, and that God not existing is another distinct possibility. However, since it is a negation, atheism is much less informative than theism. Theism is a completely arbitrary statement about the state of affairs, which is widespread because it is reassuring and purports the existence of some higher being to relate with. That other arbitrary statements on the state of affairs are not made does not erase them from existence. The belief that the universe is all there is and the belief that our universe is a simulation in an external (real) universe are two non-theistic beliefs, both covered by atheism. So why would theism get 50% rather than 33%? What about all the ideas nobody thought about and probably never will? In order to assign a meaningful probability, you need a systematic account of possibilities, which naturally involves information theory: that way you can reason about what can be expressed, rather than about what is or isn’t expressed, which is always a very small subset.

Evidence and faith simply cannot be valued the same: evidence is the only way to have a good idea of what exists and what doesn’t with higher confidence than wild guesses, because evidence is what allows theories to rise from the probability hell they all live in. Faith, on the other hand, is mostly an emotional crutch: it allows you to believe in something pleasant or convenient without the trouble of gathering evidence for it. All statements should stand to the same standards of scrutiny, depending on how much information they give about the world – since most positive statements provide many bits of information (like trying to predict the lottery numbers) they are all very unlikely a priori. Negative statements, on the other hand, contain much less than a single bit (like saying your lottery numbers won’t come up – yeah, sure, that’s a pretty safe bet), and should be considered very likely a priori. Many people make the mistake of thinking there are only two numbers in the draw, because they can only count up to two.

One thing I observe in discussions about morality or ethics, however, is that reason is never used to figure out the best thing to do, but rather to justify a predetermined idea, or to point out inconsistencies in the ideas of others. They are rarely constructive, more often aggressive. Morality is not a matter of reason or science, nor of faith; it is meme warfare.

Morality is not absolute

Regardless of what an absolute, “objective” morality might be, it does not tell us whether we care to follow it or not. Suppose for a moment that absolute morality comes from God (for a minute, we will assume he exists). Then anything he deems moral is, by definition, moral. Should he order us to kill all newborns, that would be moral, and we ought to oblige. Any argument to the effect that God would not consider these things moral boils down to placing restrictions on morality that are exterior to his being.

And so, any attempt to an objective moral system has to start from the moral intuitions we have. If a moral system clashes with our intuitions, we will almost certainly reject it. Indeed, the principal objections we find to moral frameworks are almost all of the same form: “imagine contrived situation X: the model says we should do Y, but doesn’t that seem wrong?” The theory’s defender can then either agree and then retool their theory, or disagree and maintain that the theory is correct and the intuition is wrong. But he can only do so by appealing to stronger intuitions: he can only justify that the poor man steals a loaf of bread by appealing that the harm to the baker is marginal and the benefit to the poor man and his family is great. He cannot simply assert that theft is good.

Now, if all of our intuitions were consistent (internally, and between each other), we could try to pool them together and figure out morality. Alas, they are not consistent. We all have basic intuitions about what situations are good and what situations are bad, and as to what actions are good and what actions are bad. Since good actions may lead to bad outcomes, and conversely, bad actions to good outcomes, there is a basic internal inconsistency that cannot really be resolved (see my previous post on utilitarianism). Furthermore, different people have different ideas about what good and bad outcomes are. They will disagree on whether white lies are moral or not, on whether security or privacy is more important, and so forth. Since these differences lie in intuition, and that intuition is the only basis on which morality can be founded, it is often impossible to find a satisfactory resolution.

Things can be moral because they are moral

Essentially any arbitrary proposition can become a moral imperative, and through the course of history, incompatible imperatives will arise in different societies. Imagine that an influential shaman decides, out of insanity, or on a whim, that wearing purple attracts demons and bad luck. People will believe him, and will transmit his teachings to their children, and so forth. Eventually, nobody will wear purple anymore, and if anybody was to commit that sin, onlookers would become panicked. Alas, if wearing purple scares people and makes them uncomfortable, then it is harming them – which incidentally justifies that nobody should wear that color. Wearing purple then becomes a threat to a stable society and an immoral action. That shaman would essentially have bootstrapped a completely useless moral rule onto society, which becomes difficult to remove because doing so would upset those who follow it. At this point, wearing purple is immoral because it is immoral.

Societies, cultures and religions all around the world have many such rules: some animals cannot be eaten, those that can must be killed in a precise way, prophets must not be drawn, women must cover their breasts, homosexuality is immoral, promiscuity is frowned upon, and so forth. Even though some of these beliefs might have had justifications in the past, a society without any of them would still work perfectly. Getting rid of them is difficult, however, because there is great inertia against it. The benefits that laxer rules would give to certain individuals are offset by breaking the masses’ “entitlement” to legacy rules (because change really does upset people). In practice, what this means is that changing society either requires a huge upset (like conquest) or a lot of time (generation renewal). But still, you have to wonder to what extent people have a right not to get offended. Say you really like to pick your nose – most people really don’t want to see that, but is it more reasonable for you to accommodate them, or for them to stop being offended so easily? In a sense, customs help give some uniformity to society, which comforts people, but a necessary side effect is the marginalization of the people who can’t really fit the mold.

Morality is always locally and temporally optimal

Society always considers that its current morality is close to optimal. It is torn between people who are nostalgic of traditional values and others that want to push society further along the path it’s following, but overall it is still relatively homogeneous. Despite all the apparent strife in occidental society between liberals and conservatives, almost nobody questions the right of women to vote, and almost nobody is indifferent to incest, even between consenting adults. Sure, there is debate about gay marriage, but most nonetheless accept that engaging in homosexual intercourse should not be criminal.

Each individual holds what they believe to be the optimal moral view (that is why they hold it, after all), usually acquired early in their lives, and becomes attached to it (hence the common phrase: “it was better in my time”, and attachment to “traditional values” that are in fact no older than the people defending them). In fifty years, society might accept incest, and today’s youth will balk at the utter destruction of moral values that would bring – in a hundred years, it will just be a given, and our times will attract derision. We are shaped by the moral principles we are exposed to, and due to this there is always a certain orthodoxy to them, even though many groups try to push them towards one end or the other. Due to this orthodoxy, there can be an illusion that morality is derived from rational principles: indeed, when virtually everyone shares a certain moral base, beliefs that are found to be inconsistent with that base are harder for rational people to hold.

Alas, that orthodoxy can drift, and the direction towards which society evolves is not necessarily predictable. A return to slavery would be perfectly possible, and even in a way that the people practicing it would find their own society much more evolved than ours. For instance, society could split into two groups according to IQ, the smart deciding everything for the dumb for their own good, with selective mating widening the intelligence gap between the two groups. The smart would see the dumb as an inferior species, like smarter chimps, only fit to do their bidding. This sounds like dystopia to us, but that society might call it “progress”, and how would you argue with them? Racism is comparatively easy to defeat rationally, since it’s obvious that race is a very unreliable indicator of intelligence, consciousness or moral character.

Another unexpected turn society might take would be to completely destroy the concept of privacy. Not divulging personal information would be frowned upon, much like we frown upon avaricious people now. Secrecy would be seen as a flaw, cameras would be placed absolutely everywhere, and that would be justified by how easily crime is thwarted by the measure. Sure, right now we would balk at all the possibilities for abuse, but if society became like that, without much abuse taking place, you wouldn’t actually have that argument anymore. What appears like dystopia to us could be utopia for people actually living there.

Morality is meme warfare

All this to say that morality is a complicated thing that can evolve in many ways, and several moral systems, including some we would describe as dystopias, are logically coherent enough to be immune to (logical) criticism. A moral system can only be challenged by shifting the relative importance of some of our intuitions, maybe adding some, removing others. That can only be done through shock: emotional shock that leads us to revisit a policy that had disastrous results, factual shock from new information that changes everything, or logical shock stemming from realizing that some of our basic intuitions clash (e.g. “female homosexuality is good, male homosexuality is bad” vs “the same standards should apply no matter the sex” requires either rejecting all homosexual behavior, or accepting all of it). I would not argue that either of the dystopias I have just described are good, but that’s because my moral intuitions are mainstream – if you or I were to argue against them to people living in them, we would probably hit a solid wall.

It would be interesting to argue against slavery in a past society where it was prevalent and widely accepted. I reckon we could be effective by bringing up facts that they are unaware of, but I think it is possible that we would also hit a wall. If the slave owner has literally no capacity for empathy with slaves, it would be difficult to make them develop any. One might also justify slavery by rejecting that people have a right to freedom or opportunity, and that being raised as a slave makes you a slave. Arguing with people from past societies might give us the frustrating feeling that we are trying to climb a soaped up pole: we try to get a grip on what we consider to be universally shared intuitions, only to find out they are actually recent constructions.

In the end, each and every one of us has intuitions about what is good and what is bad, and we all have a stake to convince others to think like us. That is why a statement such as “morality is subjective” is without consequence: the statement according to which “all moral systems are just as acceptable” cannot be concluded from it, because it is in itself a moral judgment. Morality being subjective does not stop anyone from claiming that everyone should behave just as they do. Much to the contrary, it is in one’s best interests for everybody else to have the same moral standards as they do. Religion does not claim to hold all the moral truths for nothing: undying confidence and a lack of nuance are the best strategy to get people to follow your lead. In that sense it is neither right nor wrong, it is merely efficient. Likewise, the future does not belong to those who are “right” or “wrong” about morality, because morality is not about truth anyway. It belongs to whoever can most effectively make their case through the use of emotional, logical or factual shocks on their fellow human beings.

I never get much insight about myself doing this (it is very difficult to remember details anyway), but the brain’s ability to create worlds is quite impressive. I did an experiment once, where I consciously read a poster, focused on remembering what I read (as much as it is possible to during a dream), and then read it again. As one might imagine, it had changed. I turned around a few times, and realized that there was absolutely no spatial stability. I feel that it is at that moment that I felt that I understood how dreams worked (though working with generative models certainly helped).

When you interact with reality, your senses are bombarded by raw data – light, sounds, tactile feedback, and so on. Your brain sorts out all that data and translates it into its own internal language, a tight net of ideas through which information ripples, creating links, triggering thoughts and emotions. You see a bike, it reminds you of that one time you won a race; you race, you think of the prize money and what you will do with it; you buy a new car, you think of all the places you’ll go with it. All sorts of concepts lie dormant in your brain, connected to some others, waiting for a thread to lead to them. When you interact with the real world, you are anchored to it, and your thoughts will hover around what you see and do.

Dreaming, on the other hand, is a semi-random walk in the world of ideas. Thoughts are what’s real, and the senses are what’s virtual. Your brain propagates everything backwards, from the core of your brain to the external sense processing centers, and then back to the core, in a loop. You never really move from a place to another in a dream, you move from an idea to the next, and you sense the results. Whatever is bubbling atop your brain at the moment you fall asleep, may that be the day’s events, or subconscious fixations, get randomly activated, and you move haphazardly from there to whatever ideas are close. You see a bike, it makes you think of that one time you won a race, so now you’re taking the bike and racing against whatever competitors mysteriously appeared. Your mind drifts on the prize money, and what you could buy with it, and then you’re racing in a busy commercial street, looking at the stores, and stopping at a car dealer. You get your car, drive away, next time you know you’re swerving uncontrollably because driving always stressed you out a bit, you crash into someone else, the police arrives, you try to run, but you can’t, because it is an unwritten rule of dreaming that you can’t run when you need to. At each step, your senses show you what your thoughts are, and unlike in reality there is no “anchor” stopping them from drifting. It is that lack of anchor that makes dreams seem so strange and whimsical.

I don’t think dreams are particularly deterministic – I see them more like a somewhat random walk through your mind. That doesn’t mean you can’t learn some things from observing them, since there’s still some kind of logic behind the transitions, and that the mental locks that block you from consciously thinking something might fail to operate in that context, giving you a sneak peek at your subconscious. Regardless, I don’t think most dreams can really be analyzed for meaning, as if your brain had an agenda showing you something. I think dreams are simply the result of the brain randomly looping through its memory banks in order to keep it alive and reinforce links that would wane otherwise.

Utilitarianism typically considers our current moral imperatives (e.g. do not kill), laws and rights (e.g. right to live) as heuristics that usually lead to greater utility and are easy to understand by the masses. Here I will argue that utilitarianism is itself a heuristic. In the same way that moral imperatives serve to simplify the execution of ethical behavior by minimizing the amount of outcome projection we have to do, utilitarianism serves to simplify ethical discourse by minimizing the basic principles on which it is predicated.

There is no single best utilitarian objective function

In a sense, utilitarianism is predicated on the existence of some objective function that, given a world, yields some kind of number telling us about how good it is. The best action to take at any point is the one that increases the value of the function by the greatest amount. For instance, if the function is the number of people alive in the world, then making the most children possible is the best course of action, and murder is always wrong (and so is suicide, incidentally). Of course, that’s a stupid objective function, since maximizing our population would involve misery, and nobody likes misery.

The first problem with finding a good utilitarian objective function is that what we deem “good” is not invariant to the actions we take. In other words, we could engineer a society to perceive any given function as the most moral. To give an extreme example, take the simplest utilitarian function, which involves maximizing “happiness”. Most reject this as being too simple, because giving everybody “happy pills” is the best and easiest course of action to achieve that objective, and we are very uncomfortable with that idea. Nonetheless, it stands to reason that once everyone is made happy through these chemical means, everyone would be happy about their own society. We don’t want to move into such a society, but if we were there we wouldn’t want to move outside of it either. Even though we disapprove of “maximizing happiness” as a goal, pursuing it would have the side effect of making us approve of it. So in devising an utilitarian function, should we take into account our current preferences, or the preferences of the society it would lead us to adopt?

The second problem lies with a lack of uniformity in what people value. Most people value life, happiness, freedom of choice, and so forth, but they do so at different degrees. For instance, is freedom to choose between good and evil valuable? I would argue that it isn’t: I value the freedom of choosing my career or choosing my next meal, but I don’t value being able to kill someone. Most religious people (as well as some others, I imagine) would argue that this freedom is fundamental, if only because it is quite difficult to explain the existence of evil if God is benevolent and evil is unnecessary. Making humans incapable of killing or hurting each other would be an excellent thing to do for me, but if one values the freedom to choose to do good or evil, the loss of it might offset the gains of preventing crime. Since we all value a certain freedom of mind, we cannot do away with it in our objective function, but it appears that different people value different freedoms. How does one question valuing moral freedom without the same arguments applying to all freedoms? Assuming we could measure them, how are life, happiness and freedom weighed against each other?

Common wisdom would state that killing one person is not acceptable in order to guarantee the happiness of another. On the other hand, it seems that it would be fine if it guaranteed the happiness of a million. Where does the threshold lie? How can any hypothetical objective function satisfactorily provide exact answers in gray areas?

Our moral intuition is not symmetric

Imagine a world full of prosperous, happy people. Then imagine the exact same world, except with one additional island, a barren world where everybody suffers from famine and poverty. Which one has higher utility? If the former does, then nuking the island in the second would raise utility and would therefore be moral. If the latter does, then the former world could raise utility simply by building up “human farms” full of miserable people.

Consider the parts of the world that live in misery. We could help them get better. Or we could nuke them completely, so that there would be nobody miserable anymore. Our intuition is that the former is moral, but that the latter is immoral. The only way utilitarianism can match this intuition is by asserting that a miserable human has positive utility. Unfortunately, if a miserable human has positive utility, then creating miserable humans would technically raise utility, potentially at a lower cost than raising the quality of life of those that already exist.

In other words, if a miserable human has negative utility, then killing them off is moral; if they have positive utility, then multiplying them is moral. This is utilitarian symmetry: if world A has utility X and world B has utility Y, where X < Y, then it is moral to go from A to B, and immoral to go from B to A. But our intuition does not work like this: it seems immoral to create people who are doomed to be miserable, but if these people exist, it is immoral to kill them. Going from A to B is immoral, and going from B to A is also immoral.

A utilitarian might be tempted to say that it is immoral to kill miserable people because it will lower the happiness of those who care for them, but I don’t think that really holds water: people care most for those who are geographically close to them, so if you completely destroyed a large swath of land and stopped the media from saying anything about it, a very limited number of people would actually be affected, and the negative impact would not last very long, certainly less than a generation. I would argue that utilitarianism assigns greatest utility to bringing the whole world at the level of comfort enjoyed by the most advanced of its civilizations, but I would argue that it still assigns positive utility to outright killing off the parts that don’t. Indeed, it is difficult to argue that the absence of miserable people* is worse than their presence without admitting in the same breath that farming humans for food and organs would bring higher utility to the world. Since destruction is arguably much easier and much faster than construction, you can see what this leads to.

Utilitarianism is a nice theory which can refine our moral intuition very much and help make it more consistent, but as it fails to take into account the lack of symmetry of these intuitions, it must still be considered as an imperfect heuristic, regardless of the objective function that’s being used. The lack of symmetry I am talking about lies in the fact that it is both immoral to kill off someone who is living in misery and immoral to willingly give birth to someone into misery, even though in the former case we go from A to B, and in the latter we go from B to A. Utilitarianism’s mathematical formalism is insufficient to represent this, because if one transition is moral, the other ought to be automatically immoral (and vice versa).

* I know I am oversimplifying: not enjoying modern comfort does not make people “miserable”, and there is no particular part of the world where everybody is miserable either. My point still stands, though.

For this reason, I believe that the most important thing in a discussion or an essay about free will is to determine what, exactly, it is that we are feeling. By identifying the probable mechanism(s) behind our impression of freedom, we can more easily determine whether that impression is related to any kind of hard reality.

What produces the impression of free will

I would say that this is relatively simple to understand, although I cannot vouch for the biology behind it. The gist of the idea is that your “self” corresponds to the model you have of yourself – whenever you think of your own “self”, although you might refer to your physical body and everything it contains, you can only do so through a mental model. Since this mental model is necessarily much simpler and cruder than the real thing, and that it is used to inform your decision process, it is not (cannot be) fully capable of predicting your behavior. Now, what happens here is that we all conflate the mental model we have of ourselves with the physical system it is supposed to model. The fact that the model is typically invariant to the actual decisions that are made, coupled with the observation that decisions are indeed made, leads to an unmistakable impression of free will.

In other words, whenever you do introspection – whenever you say “I” – you conjure a mental model of yourself, which you manipulate and combine with other concepts. But even though some slight perturbations in your brain might lead you to choose to wear shoes over sandals or vice versa, that does not really make any difference as to your conception of self. Typically, your conception of “self” cannot tell apart a brain that chooses to wear shoes from a brain that chooses to wear sandals. Even though they have different behavior, as far as your self-image is concerned, both of them are “you”. Because of this, you simply assume that you can do either action.

It might help to consider that all thoughts operate on mental objects, which are mere models of real objects. When we give an attribute to something, we are only tagging the actual object through the mental model we have of it (after all, even if we could acquire it, it is not like we can store all the molecular information of anything sizable in our brains). So when you say “this dog barks a lot”, you are saying “I model the object at this location as a dog that barks a lot, and you should model it the same way”. Similarly, when you say “I might ask her out”, you are saying “I model myself as a human that might ask this girl out, but not certainly, and you may use this information to improve the model you have of me”. In this context, free will is fairly easy to understand: the mental model we have of a human being is non-deterministic. It does not really matter whether an actual, physical human is deterministic or not because we apply the concept to the mental model.

Except, of course, that if we know that the universe is deterministic, we are compelled to enrich our mental models with that information. And that is where problems happen: indeed, the near totality of our mental models are non-deterministic. That is not necessarily a big deal as long as you realize that mental models are not the same as reality, but conflating both is so convenient in practice that we do it all the time (heck, many people are utterly incapable of grasping the difference between language and reality). Long story short, the idea that the universe is deterministic causes huge cognitive dissonance because it clashes with natural thought mechanisms. Different people cope with this in various ways. Personally, understanding why there is a cognitive dissonance suffices to dispel it in my own mind.

So what about free will?

To me, the impression of free will makes perfect sense – any entity capable of introspection, in so far that they model themselves in an incomplete fashion, would have an impression of free will. That is simply, as I said, because many different brains map to the same self-model, so the self-model M could be said to have the choice to do X if there exists a brain that does X and is modeled by M. Any action that does not contradict the model you have of yourself is an action you will think you can do. I do not see the impression of free will as necessarily resulting from a special or complex phenomenon.

Now, I could define “free will” as exactly corresponding to the conditions I believe lead to an “impression of free will”: “an entity has free will if it makes use of an incomplete model of itself as part of its decision process, and conflates that model with the physical entity it refers to” (note: that definition is not necessarily meant to be exhaustive). Indeed, I would say that all such entities would feel like they have free will, so it is not a huge stretch to outright define it as such (do note that I am usually rather liberal with definitions – I don’t care what a word means, as long as the definition is clearly stated).

I reckon that most people would not be satisfied with conflating free will with the impression of free will. That is, they would contend that everything I have said supports the idea that free will does not really exist, and that it is an illusion. I must say that I am perfectly fine with this interpretation (again, I am not the kind of person who bickers about definitions – clarity is all that matters to me).

Ultimately, though, if “free will” is a coherent concept, then this must mean there is a formal way to tell apart things that have free will from things that do not. For as long as a clear, formal test for free will is not devised and agreed upon, I must confess I have no idea what the debate is about. Some people view non-determinism as a sine qua non condition to free will, whereas compatibilism contends that determinism and free will are compatible. Personally, I am of the opinion that whether free will is compatible with determinism or not is a matter of definition, and I frankly could not care less which way it goes. All I know is that if you can’t agree on something this fundamental, that’s because you are talking about different things.

We all have the same impression, the same subjective perception that we have a freedom of choice. However, it is quite clear to me that different people have different ideas and different expectations as to what free will represents. The whole free will debate is about determining whether the impression we have conforms to our expectations about what causes it. Alas, there is a fundamental divide in the expectations of at least two groups of people: those that believe determinism precludes free will, and compatibilism. Don’t even try to determine which side is “right”: the difference lies solely in expectations*. For this reason, there is no unified concept of free will and there probably never will be.

Long story short, I believe I have presented a nice and coherent explanation as to why we feel that we have free will – I would be quite interested to know how close I am to being correct. On the other hand, I do not care whether free will is an “illusion” or not, and believe that all debates of that nature are sterile. All arguments about free will are grounded in semantics, not in reality.

* Roughly speaking, you can understand “X freely chose to do Y” as either meaning that it is possible that X would not have done Y – or you can understand it as the weaker proposition that the fact that X did Y can be traced back to factors that are mostly internal to X. The first definition can be seen as more intuitive, but naive, incomplete and ill-defined. The second can be seen as better fleshed out, but less intuitive, self-serving and (again) ill-defined (what is “internal”? what is “external”?). To be honest, both views are defensible, in the sense that someone on the fence can be compelled to go either way. I don’t think either side will ever win, and if one does, I don’t care which one.

Is there an objective measure we can use to determine whether an entity is intelligent or not, or to compare the intelligence of several entities? I am not talking about IQ tests, which are neither a complete nor a reliable assessment of capability, but am trying to engage in a deeper reflection about the nature of the thing.

To put things in a clear context, what would it mean for a machine to be “intelligent” (not “human”, mind you)? Keeping in mind that we are a kind of machine, I would say that this also defines intelligence for us. Note my use of the word “defines” – I consider that this exercise is essentially one of definition, of trying to pin down the formal concept that best corresponds to the intuitions we have about intelligence. As such I am not trying to find an objective truth to intelligence (whatever that means) as much as I am trying to make the concept richer and more useful.

Intelligence and functionalism

Functionalism (as I mean it in this section, anyway) is the belief that mental states and properties are defined by their behavior. To determine whether a machine is intelligent or not under that model, one would simply relate its inputs to its outputs. To each input would correspond a set of “intelligent” outputs, and we could check in what proportion the machine responds “intelligently” given various inputs.

Here we could interpret an input as being a problem statement, such as “Is there a largest prime number?”, an output as being an answer, and an intelligent output as being the right answer. We could also interpret an input as a situation, an output as an action, and an intelligent output as one that sustains the existence of the entity.

Regardless, there exists a hypothetical machine which, under a functionalist interpretation, would be maximally intelligent. The problem is that it seems counter-intuitive for this machine to be intelligent. Here, let us use the situation/action/survival context, but there is a version of that machine for pretty much any context. The machine in question is extremely simple: given a situation, it will simulate every single possible course of action, and it will choose the action that gives it the best survival expectancy. This is the “brute force” approach to problem solving: try all possible answers until you find the right one. This will almost always work, with the caveat that it will take virtually forever: the number of possibilities is exponential in the length of the answer.

Under a functionalist interpretation, such a machine certainly would be intelligent, since it would almost always yield the best answers. Nonetheless, it seems to me that such an example shows that intelligence ought to be defined differently: bubble sort might sort just as well as quicksort, I would consider the latter to be smarter, because it has better complexity.

Intelligence and speed

In order to define intelligence, I would add one important ingredient, which is speed. That is, I would say that if X and Y can solve the exact same problems, but that Y can solve them faster than X, then Y is more intelligent than X. Computational complexity sort of fits in that framework, since better complexity will lead to better speed for large enough problems – however, the bottom line (speed) is what matters.

The definition of intelligence then becomes the following: the intelligence of an entity is proportional to the number of problems it can solve, and inversely proportional to the time it takes to solve them. There is, however, no objective way to weigh these attributes. This portrays intelligence as a partial order (which I believe is definitely the way to go), where the intelligence of two beings cannot be compared if each of them can solve problems that the other cannot, or if each can solve some problems faster than the other.

Nonetheless, both coverage and speed are important factors for intelligence, and this explains why brute force is not an “intelligent” problem solving method: it is much too slow to be practical. Intelligence is not as much the ability to solve problems as it is the ability to solve them within a reasonable time frame.

In the case of humans, it is often the case that people can do feats of mind that others simply cannot, and I believe that this gives us a more binary view of intelligence, where it is all about what you can or cannot do, or the ideas you have or don’t have. Speed also matters, but at a coarser scale, in the sense that greater efficiency is often seen as resulting from better training or higher effort, unless the gap seems too large to be bridged. You will feel that someone is smarter than you if that person can do mental feats that you believe are beyond your abilities, but less so if you feel that you can do whatever that person can do, regardless of how fast, as long as the difference is not tremendous. Note that there is an implicit understanding that people that think faster than others can do things that others cannot, so fast thinkers will indeed be viewed as smarter, if only for that reason. Nonetheless, I believe there is a pitfall in our conception of intelligence, which is that solving problems is not in itself difficult, given enough time – solving them quickly enough is what truly matters.

Note that this can be generalized further: we could view intelligence as the ability to solve problems given a limited amount of resources. Resources may be time, memory, cost, etc. Solving problems with extremely large or infinite resources is rather easy, but doing so quickly and cheaply is much more difficult. This draws a parallel about optimization and algorithm theory, where one tries to make the most out of limited computational resources.

Closing thoughts

I would like to close this article with a theological implication: an omnipotent, omniscient being does not need to be intelligent. Indeed, one would imagine that such an entity has infinite resources – that with some caveats, it can try all possible universes until it finds one that it likes, and/or one that works well. In a sense, the concept of “intelligent design” is made absurd by the supposition that the creator is infinitely powerful. Since intelligence is the ability to solve problems in a limited time span and/or with a limited amount of resources, intelligence is an extraneous attribute to give to an all-powerful entity. All it needs is enough sense to figure out how to solve problems using brute force, but there is nothing particularly clever about that.

Artificial intelligence is indeed a very common topic in science fiction, though I rarely find that it is portrayed in a believable manner. In fiction, AI usually takes a quite anthropomorphic turn, sometimes tainted by some extreme form of rationalism. On one hand, anthropomorphizing machines allow us to understand or empathize with their behavior (in all ways that it would end up being similar to ours). On the other, the nature of programming – the determinism of programs, their strict organization – is transposed to AI, leading us to imagine rigid, logical, non-emotional, non-creative beings.

I will start by exploring the concept of AI as it is usually portrayed, and explain in why it is unrealistic. I will then expose what AI research is all about and what results one may expect from it, in due time.

AI in our collective psyche

The only “intelligent” beings we know are members of our own species, and to a lesser extent some specimens of other species. The natural consequence of this lack of variety is that we sometimes tend to conflate several of the higher cognition capabilities we have into a single package, such that having one implies having the others. For this reason, when we imagine machines capable of solving complex problems, understanding us and advising us, we imagine them with an ego. We all wonder when machines will “gain sentience”, as if it was a natural consequence of getting more and more complex. Ergo, the omnipresent, looming concept of the machine uprising in our minds.

In a sense, the idea that we would become over-reliant on machines and that they would rebel against us sounds engineered to the express purpose of making AI seem sinister. The “AI persona” is devised by cherry picking several human traits which are then combined and assigned to all instances of AI. Neutral traits are usually given to them, whereas traits that we feel are precious and unique to us are withheld.

Hence, the AI will have an ego, because we all have one. The AI will have a sense of “belonging” with respect to its fellow robots, because we feel one with respect to other humans. Finally, of course, the AI will be extremely rational, because we readily associate machines to rationality. However, this rationality will be a caricature based on what most humans believe to be rational, exacerbated to the level of dogma. The bottom line is that the machine will perceive itself as being more rational than humans, hence superior to them.

On the other hand, though, valuable emotions such as friendship and love will be handed out much less liberally. AI will not be creative, except perhaps technologically.

All of these factors create a stereotype, a sinister image of artificial intelligence. The extreme form of rationalism we endow it with leads us to despise it, since rationalism is often perceived as being callous, calculating and empty. The fact that it has an ego and is united within its ranks makes it a threat. However, the stereotype cherry picks attributes that achieves this effect – the particular form of belliquous vanity that would lead machines to exterminate humans is a form of emotion. If machines can feel destructive emotions, why not constructive ones? Is there any reason why machines would value emancipation over the positive reinforcement humans give them when they do what they are told and would conclude that attack is the best method of achieving their goal? Of all traits endemic to humans, why exactly do machines get megalomania?

The stereotypical AI that I have just described is used in science fiction both as a plot device in an “us versus them” situation, and as a red herring that is eventually attacked and debunked in order to show that “they are just like us”. Still, the stereotype is deeply ingrained in our collective psyche, and even though many will give all human traits, good or bad, to machines, there is still a widespread understanding that they would be sentient.

The path to AI

Most people have absolutely no idea of what artificial intelligence research is all about. What they know about the field of computer science, they know from their friends or family who work for various companies, from the stories they read in newspapers and magazines about hackers and computer whizzes, or from the vague trivia one gets from day-to-day social interaction. Alas, this gives them erroneous ideas about artificial intelligence, because despite what one might think, AI has little to do with applied computer science. It has more in common with mathematics for the formalism, and evolutionary/neurobiology for the intuition. It is not about “designing” intelligent programs, but rather about “evolving” programs that can solve a given objective. AI is to conventional computer science what evolution is to creation, and indeed many AI techniques are rather directly inspired from biology (genetic algorithms and neural networks, to name a few).

Most people, when thinking of the making of of artificial intelligence, will imagine a programmer or a team of programmers writing a complex program, encoding various preset behaviors to be triggered by various conditions. They will picture an “if you see a baby falling, extend your arm to catch it” kind of program. One cannot really blame people for having this idea, because most programs and applications we use in everyday life do work like that, and most programmers spend their time solving specific problems in rigid, structured ways (and they will often go the extra mile to prove that their solutions are correct). It is indeed difficult to understand how intelligence, or even consciousness, could arise in such programs. Intelligence entails the ability to properly deal with novel, unforeseen situations. If machines acted “because they were programmed to act that way”, they could only deal with the situations their creators prepared them to face*.

On the contrary, you would expect a “true” AI to be unintelligible for a human. Such an AI would need to adapt its behavior to new information, gaining confidence from its successes and learning from its errors. It would need to be able to deal with tens of thousands of concepts, put them together in coherent sentences, and understand all the combinations we understand. This is a truly herculean task and no human could possibly fully grasp, let alone create, a program capable of juggling so many things at once and adjusting any part of itself as new information is acquired. In other words, no human can understand, let alone “design” all the details of a human brain, and a machine of human-like intelligence would be of a similar level of complexity. If we create advanced artificial intelligence, we will not understand how it works.

This being said, it is not because we do not understand how something works that we do not understand what it does. This is an important distinction to make, because while one might be worried by the idea that we will not know how advanced AI works, the fact is that we will nonetheless know (with near certainty) what it aims to do. A good parallel to make is that you might not have the slightest clue as to how your toaster works, but nonetheless you know that it makes toast. If you want to solve a problem X, the point of AI is to figure out a way to automatically produce a program that solves X – you do not know how that program solves X, but you know that it does.

Let’s start by looking at how evolution did it for us. Humans are the way they are because they fit the environment they live in. In a context where organisms have to navigate a world and compete for the limited supply of resources it contains in order to survive, through natural selection and evolution, they will acquire many traits that help them be more competitive. They will develop attacks in order to better consume resources, defenses to survive longer against predators and the elements, they will attempt to produce as many mature offspring as possible, they will develop cooperation strategies with individuals that have similar genes, etc. All in all, natural evolution produces a wide variety of individuals that care most about themselves, then about their families, friends and peers. Evolution promotes the existence of consciousness, egos, cliques, groupthink, science for what must be gotten right and religion for the rest. It does, because the selective pressure is such that these aspects prevail.

AI, on the other hand, is meant to serve a purpose for us. Of course, it would be interesting to make machines that are as close to being human as we can make them, but this would be more of a curiosity than something truly useful, since humans already fill that niche. In general, there is a large set of problems that we would like AI to solve, ranging from investigating and answering factual inquiries (a sort of “super-google”) to directing and executing military operations. We might even use it to solve the mysteries of the universe for us (though I believe it is more fun to do it ourselves).

Regardless of the application, there are precise steps to take if we want the AI to focus on the things that matter to us. The first step is to design a way to verify how well the machine is doing, so that we can guide it. We may also design a series of progressively more complex tasks that will guide the machine towards solving complex problems for us (a bit like a school curriculum). Once we are able to determine the level of proficiency of the machine at doing what we want it to do, we can reward it whenever it does better than before, and punish it whenever it does worse.

The specifics of this training are still the subject of intensive research: it is easy enough to reward and punish, but that does not mean the machine will ever get good. However, the real point here is that we are controlling the machine’s concept of “survival”. Whereas survival in the real world is self-explanatory, survival for AI is doing what we tell it to do. For this reason, the traits machines will develop through the course of becoming “intelligent” will be nothing like traits that organisms develop on Earth. The selective pressure that applies on an organism determines how it will adapt – traits that lead to greater rewards will be developed, neutral traits will appear and disappear a bit at random, and traits that lead to punishment will be culled.

Imagine that you are training a military AI to hit targets. It will develop an acute sense of sight, so that it can detect targets. It will develop trajectory prediction abilities, so that it can shoot where targets will be when the bullet gets there. It might develop heuristics to the traveling salesman problem (given a list of cities to visit, the problem is to visit all of them exactly once with the least amount of travel), so that it can shoot multiple targets in the most efficient order. It will not, however, develop consciousness, because what kind of advantage would it get from it? Developing a complex trait that does not lead to reward or punishment is a random event, and about as likely (and by about as likely I mean a lot less likely) as getting struck by lightning a couple billion times**. Neither will it develop anything resembling free will, because any AI that chooses not to hit a target will be punished to oblivion. That trait will have no time to, so to speak, flourish. You will end up with excellent shooting machines, nothing more, nothing less.

One caveat is that AI will likely be trained by randomized algorithms (a bit like evolution, which works through a mix of random mutation and natural selection). This is actually already the case – currently, (pseudo-)randomness is a big part of our algorithms. Technically, there is some uncertainty about what we would produce. So no, it is not “impossible” that we would accidentally produce an shooting machine that would have an ego, a thirst for emancipation, and would attempt to destroy us – all out of sheer bad luck. Such an event, however, seems about as improbable as evolving a fish on land or quantum tunneling through your couch, because these traits, while quite plausible for biological organisms evolved on Earth, are far-fetched if we select using different criteria. Furthermore, accidentally getting one such AI does not make it any easier to accidentally get another – one sentient robot out of a billion servile ones cannot do much damage.

In conclusion

For sure, we will not comprehend how AI works. In fact, we already have trouble understanding the highly limited toy AI we produce. We humans can only reason so far, and solving complex, ambiguous and dynamic problems is beyond what we can do. We need to develop better tools: tools that can shape and adapt programs to the whims of the current situation, ways to express computation that do not have the hopeless structural rigidity of programming languages and methodical rational thought. That is the path AI must pursue.

For sure, AI will surprise us. But all surprises – the good, the bad, the strange – are equally likely. There is no point in hoping more than what we will guide AI to do, and there is no point either in fearing the worst. AI will fit the niches we will train them to fit, as imperfectly as we fit the niches left open for us in our own societies and ecological system.

All in all, I would say that the potential advantages of AI are immense and that the risks are contrived and improbable. There is some lingering fear that machines will end up being like us, for the better or for the worse, and that we will have to deal with them either with gunfire or by acknowledging them as equals. However, it stands to reason that we are the product of a long adaptation to our environment, and as far as we control the environment in which machines learn and evolve, the most they will do is to solve the problems we ask them to solve.

After all, to them, this is survival.

* Unless, of course, they “free themselves from their programming”, a common meme which is patently absurd – if a program does not behave as intended, that is because a programming mistake was made, and then you expect sub-optimal behavior or partial or total failure, not emancipation. Windows does not gain sentience – it shows a blue screen.

** Note that at worst, you might end up with one conscious shooting machine out of however many you make (you won’t, though – non-positive traits are ridiculously unlikely to occur unless they are simple enough to happen randomly with meaningful probability). It will not be systematic, however, and it is a good idea to train many machines independently, so that they do not all share the same flaws.