Organizing the conflicts with Deutsch and rationalism
The conflict
Over the last year, I’ve been trying to better understand Deutsch’s worldview in general and how it intersects with rationalism in particular.
In my piece Deutsch Eats Rationalism, which was an honoroble mention in ACX’s book review contest, I wrestled with the idea that Deutsch’s worldview is caustic to rationalism.
I started it somewhat provocatively:
[Deutsch’s ideas] form a universal acid: They dissolve your entire worldview.
Especially if your worldview happens to contain Bayesianism, forecasting, evolutionary psychology, behavioral genetics, cognitive biases, concerns for animal welfare, and an unease about the existential risks posed from the development of artificial intelligence.Did I miss anything?
It’s a big list of conflicts.
You may be thinking, is Deutsch really worth reading? What’s the likelihood that he’s right and all these smart people are wrong about all these different topics?
(Deutsch would remind us that we’re also wrong for talking in probabilities.)
Admittedly, the idea that Deutsch dissolves every tenet of rationalism may be too strong. In reality, I’m agnostic to how this all plays out. But these ideas are at least seemingly in conflict. I wrote the piece because I’m in the unfortunate situation where the two biggest influences on my worldview seem to be in complete contradiction1.
Further, the two communities talk past each other at times2. Why should rationalists take Deutsch seriously if he denies obvious facts like genetic influence and indispensable tools like Bayesianism? Why should Deutschians take rationalists seriously if they still follow the obviously broken induction?
They’re both making absolute howlers from the other side’s perspective.
Organizing points of disagreement
As noted, I’ve attempted to unpack these conflicts in Deutsch Eats Rationalism. That might be the best place to start if you’re new to these issues. In this piece here, I’m more interested in organizing the conflicts.
My first step was to list out the different influences on Deutsch. Not all these areas of conflict come from the same place. Many come from Deutsch’s main intellectual influence Karl Popper, but other intellectual fingerprints can be found in Deutsch’s work.
There are other aspects to Deutsch’s worldview, but I've picked out three ideas that seem to be doing most of the work in creating all this conflict3:
Induction cannot create knowledge: Knowledge does not come from applying histroic base rates. Nor does it get “verified” from repeated observations.
People are fallible: People are subject to error. Although objective truth exists, we can never be certain we have found it.
People are universal explainers: There are no barriers to what knowledge we can create (via conjecturing explanations)4.
In this piece we won’t argue for or against these ideas. We’ll simply identify which of the rationalist ideas conflict with them respectively.
Like many taxonomies, it might not be cutting things perfectly at the joints.
Hopefully it’s more useful than not.
An example of how to interpret this diagram: Behavioral genetics (as a set of ideas and findings) conflicts with the Deutschian idea that people are universal explainers. Bayesianism (as an epistemology) conflicts with the Deutschian idea that induction doesn’t work to create knowledge.
Not all of the ideas included in the diagram are part of the rationalism canon. Rationalists aren’t running around arguing for (or against) proportional representation voting systems.
Here’s another cut. This time we have the ideas commonly held by rationalists in bold.
As you can see most rationalist ideas conflict with either the orange or the blue category. They must be either implicitly endorsing induction or denying that people are universal explainers.
When attempting to reconcile the conflicts, we can treat them separately.
The conflict with the two in the orange circle, Bayesianism and predicting the future, stem from Karl Popper’s influence on Deutsch. It boils down to Bayesianism being crypto-induction. Yet induction is false (as Hume pointed out) and we don’t use induction (as Popper pointed out but Hume was confused by).
Three of the ones in the blue circle (evolutionary psychology, behavioral genetics and cognitive biases) all present the same issue. Namely, they implicitly posit that there are barriers to creating knowledge and solving problems that cannot be overcome. It’d maybe be better to group them:
Animal welfare is a bit different. The conflict arises from the fact that people are universal explainers and animals are not. Perhaps suffering relies on being a universal explainer. Although, I’m hesitant to attribute arguments against animal welfare to Deutsch as he has not explicitly made them5.
As far as I can tell, Deutsch’s lack of concern about an existential risk from AGI stems from two directions. The first is that many rationalists argue we may stumble into a dangerous AGI by ramping up current machine learning techniques. However, for a Deutschian, these techniques are essentially induction. Deutsch would posit that current techniques will not be able to create new knowledge. If they do, I’d take that as a pretty big counterexample to Deutsch’s worldview.
The second direction is the fact that people are universal explainers. A true AGI, one that can create new knowledge, would also be a universal explainer and would be a person in a relevant moral sense. This implies that we shouldn’t enslave an AGI (slavery is wrong) and that there are no insurmountable barriers to an AGI just as there are no insurmountable barriers for us (so an AGI won’t be stuck on it’s starting goals).
Similarly, for x-risk in general, estimates of our impending doom rely on induction like estimates based off of historical base rates of near misses or meteor frequencies or something. But we can’t apply these base rates because problems are solvable. In principle, there is no reason why we can’t cool the planet, or deflect meteors, or create political or technological knowledge to solve nuclear risk. I wrote more about this particular issue here.
Why are problems solvable? Because we are universal explainers. There are no barriers to the knowledge we can create.
I’ve put Empiricism in the intersection of the three circles.
Empiricism is just the position that our knowledge comes from the outside world via our senses. This runs completely contra to Popper and Deutsch. Rather, our knowledge comes from bold conjectures about the world (people are universal explainers). These conjectures are uncertain (people are fallibile). And they are not sourced from repeated observations (induction cannot create knowledge).
This of course is the core of Popper, and Deutsch spends much of The Beginning of Infinity and The Fabric of Reality dealing with it. If you haven’t read them, you should. Even if you disagree, they’re a delight.
I’ve never actually considered myself a rationalist, per se. But, as Michael Neilsen joked, a rationalist’s membership cry is "I'm not a rationalist, but…".
I wonder how many rationalists have even head of Deutsch. He’s had conversations with rationalist or rationalist-adjacent people like Tyler Cowen, Robin Hanson and Sam Harris and has been cited by Steven Pinker and Scott Aaronson, so there’s a decent chance many know about him.
Other aspects of Deutsch’s work are quantum computation, multiple universes, and constructor theory, among others.
I discuss in more detail in Deutsch Eats Rationalism.
Excellent piece and graphical organisation of the conflicts!
"Empiricism is just the position that our knowledge comes from the outside world via our senses. This runs completely contra to Popper and Deutsch. Rather, our knowledge comes from bold conjectures about the world (people are universal explainers). These conjectures are uncertain (people are fallibile). And they are not sourced from repeated observations (induction cannot create knowledge)."
. The problem with the above is false dichotomy Empiricism is not everything , because you can't generate explanations just by looking. But it is not nothing, because a conjecture that has never been tested is not knowledge.
This has similar problems to Deutsch 's critique of induction. It is true that pure empiricism is not a source of explanations, but it does not follow that empiricism can play no useful role: empirical evidence can even play a role in Popperian science, as a source of refutation.
It is true that empirical data need interpretation. It follows that pure empiricism is useless, but does imply that empiricism has no use.
Kant got it right centuries ago: "Thoughts without content are empty; intuitions [sense data] without conceptions blind’