Dennett is wrong about luck

Dennett vs Sapolsky

Gemini 3 Pro Image.

Dan Dennett and Robert Sapolsky are both scientific-minded determinists. Dennett is a compatibilist, believing free will and determinism are compatible. Sapolsky is an incompatibilist, believing they are not.

My first-pass read of Dennett’s stance in Elbow Room is that just because your actions proceed from the laws of physics does not mean they are out of your control. When you decide which book to read, it is a product of the incoming sensory information encoded through your retinas and the neurons in your brain firing which carry out a deliberative reasoning process. Yes, this process is determined, but it is your brain which is doing the choosing (or rather, the program run by your brain); you just are your brain (or the program); and so it is you who is doing the choosing! Your brain has the capabilities to reflect on itself, to weigh evidence, to abstain from acting on some desires and execute on others. What more do you want from freedom?

I think this is a quite powerful vision for how to live your everyday life. I am not sure, however, that it rescues standard moral intuitions of blame and punishment. Imagine: Jack chose to study hard and pursue effective altruism, while Jill chose to couch-surf around, drift in and out of employment, and get in drunk fights. It’s true in Dennett’s sense that those choices occurred in Jack and Jill’s respective brains; nobody was holding a gun to their heads. But Jack grew up as the child of a lawyer and an author, with a surplus of books, fresh air, and financial stability. Jill’s single mother was a low-income alcoholic, and Jill grew up hearing gunshots echoing through her neighborhood every night. There’s ample evidence in Sapolsky’s Determined that growing up in this environment starts to raise baseline stress levels, screw up the metabolism of the prefrontal cortex, eat away at physical health, and by and large really screw with your brain. Some of these effects occur by age five. Thus, while Jill’s actions are proximately a result of her own impulsive, short-sighted choices, there is a causal story as to why Jill has poor judgment and Jack doesn’t—and Jill never at any point chose to have poor judgment. “Just do better and resist temptation” doesn’t work if you are a determinist, because there might not be a physical pathway by which Jill reasonably could “do better.” Sapolsky writes:

We’re pretty good at recognizing that we have no control over the attributes that life has gifted or cursed us with. But what we do with those attributes at right/wrong crossroads powerfully, toxically invites us to conclude, with the strongest of intuitions, that we are seeing free will in action. But the reality is that whenever you display admirable gumption, squander opportunity in a murk of self-indulgence, majestically stare down tempatation or belly flop into it, there are all the outcome of the functioning of the PFC [prefrontal cortex] and the brain regions it connects to. And that PFC function is the outcome of the second before, minutes before, millenia before. The same punch line as in the previous chapter concerning the entire brain.

Even Dennett sees this as a challenge:

I may not be controlled by the past, but my current capacity to make decision and control myself is not, it seems, something I can take credit for. How could any deterministic process of “character transformation” beginning with a being that was not responsible for any of its “decisions” ever yield a being who was not only responsible for its decisions, but responsible for having the sort of character that would make those decisions?

The main difference between Jack and Jill is that Jack was lucky, and Jill wasn’t. Jack got set up with deliberative powers that would make him more effective at self-control than Jill. But how does a person deserve blame for the result of luck? According to Sapolsky, this completely undermines the whole praise & blame game.

Dennett, however, thinks that luck might matter in extreme cases, like mental disability, but does not affect ordinary people. Why? Because he believes luck washes out:

Imagine a footrace in which the starting line was staggered: those with birthdays in January start a yard ahead of those born in February. … Isn’t this manifestly unfair? Yes, if it’s a hundred yard dash. No, if it’s a marathon. In a marathon such a relatively small initial advantage would count for nothing, since one can reliably expect other fortuitous breaks to have even greater effects. … A good runner who starts at the back of the pack, if he is really good enough to deserve winning, will probably have plenty of opportunity to overcome the initial disadvantage. … After all, luck averages out in the long run.

Sapolsky:

No, it doesn’t. Suppose you’re born a crack baby. In order to counterbalance this bad luck, does society rush in to ensure that you’ll be raised in relative affluence and with various therapies to overcome your neurodevelopmental problems? No, you are overwhelmingly likely to be born into poverty and stay there. Well then, says society, let’s make sure your mother is loving, is stable, has lots of free time to nurture you with books and museum visits. Yeah, right; as we know, your mother is likely to be drowning in the pathological consequences of her own miserable luck in life, with a good chance of leaving you neglected, abused, shuttled between foster homes. Well, does society at least mobilize then to counterbalance that additional bad luck, ensuring that you live in a safe neighborhood with excellent schools? Nope, your neighborhood is likely to be gang-riddled and your school underfunded.

So, who’s right? Does luck wash out, or does it compound?

Let’s step back and look at the dynamics here. Dennett’s picture of luck is like independent coin flips. If you’re starting a coin-flipping contest, if an opponent starts with a few more heads that sucks, but if you flip enough coins it won’t make a substantial difference; you’re quite likely to end with similar results. This works because:

1. There are many subsequent random events, so there are lots of opportunities to get lucky.

2. Each event is of similar (in this case identical) magnitude; it’s not as though some coin flips count for 20 heads and some only for 1.

3. Each is independent of the last, so bad luck at the start doesn’t worsen subsequent changes.

For this to transfer to real-life cases of praise and blame, similar conditions should hold. Do they?

Sapolsky catches flak for not addressing the philosophical literature, but his advantage is a great deal more scientific evidence in 2023 than Dennett had access to forty years earlier. And the science in Determined challenges all three of these assumptions:

1. While you can get lucky later in life, your brain is by far at its most malleable when you are growing up. Being born in poverty or with crappy genes can permanently change your neurochemistry in ways that are difficult or impossible to reverse by the time you’ve reached autonomous maturity. One bad draw, and you might be at a permanent cognitive disadvantage.

2. Similarly, magnitude of opportunities isn’t the same. Getting the bad luck of a low IQ is a durable and mostly unchangeable genetic trait that touches pretty much every facet of your life; a lucky admittance to a fancy private school won’t make up for that.

3. As Sapolsky says, if you’re born a crack baby that makes you much more likely to suffer all kinds of bad luck, from abuse to malnutrition to crappy education. Yes, there are exceptional rags to riches stories—some people do get that lucky—but if being born into poverty “washed out,” you would expect them to climb to middle class about as easily as those born into it. That isn’t what we see. The events are highly interdependent. In the marathon, starting a few steps back doesn’t also cause your legs to cramp up and your lungs to tighten, but that’s what it looks like here.

While Dennett might not have had all the science to recognize the first two points, he should have paid more attention to the third. Why? Because Dennett thinks the basis for crafting our character is a recursive process. A “self-controlling agent” benefits from thinking about the steps to achieve its goal, but it can also benefit from thinking about how it should think about achieving its goal, determining advantageous character traits, and working to cultivate them. This creates a positive feedback loop, where minds that are more rational will be more successful at making themselves even more rational, and so on. In humans, this probably saturates at some upper limit constrained by our biological makeup, but the presence of any positive feedback loop should merit some uncertainty before assuming i.i.d. luck events—or at least a footnote.¹

I’ll also admit some uncertainty. Social science isn’t my area of expertise, and it’s possible the luck landscape in terms of long-term outcomes looks more like Dennett’s marathon. Based on the evidence from the brain, though, I think Sapolsky paints a much more accurate analogy:

You start out a marathon a few steps back … a quarter mile in, because you’re still lagging conspicuously at the back of the pack, it’s your ankles that the hyena nips. At the file-mile mark, the rehydration tent is almost out of water and you can get only a few of the dregs. By ten miles, you’ve got stomach cramps from the bad water. By twenty miles, your way is blocked by the people who assume the race is done and are sweeping the street. And all the while, you watch the receding backsides of the rest of the runners, each thinking that they’ve earned, they’re entitled to, a decent shot at winning.

1. This might be a different story for AI which can physically redesign themselves; it’s why some AI alignment folks are quite worried about an intelligence explosion, although there is still much debate about whether and how fast the feedback loop might saturate.