Disagreements with Deutsch

A long time ago I wrote an article called “Disagreements with Deutsch: The Afterlife.” It was a terrible article and, in my opinion, contains a significant mistake. Though the general concern being expressed is worthy of some consideration, this article does a poor job expressing it.

The way I recall the story, I decided to not publish it. Months later, I realized it had somehow been published and was now my top viewed article on Medium. I didn’t feel like I could take it back, so I left it up.

However, the general idea of having disagreements with Deutsch is a sound one. What are the ideas that Deutsch introduced in his books that I’m not sure I agree with? I’ve spent years researching Deutsch’s Fabric of Reality after I read it and eventually came to accept all four strands (it took me a long time to accept many-worlds quantum physics) because I failed to refute them via criticism.

But there are other arguments Deutsch makes where I found some pretty good criticisms and so don’t yet accept his arguments. I wanted to make a list of these ‘disagreements’ I have with David Deutsch.

This seemed like a good idea for a Critical Rationalist site because it’s important that we don’t accept anyone as an authority. So if I found what I thought might be good criticisms of something Deutsch said, then those should be raised and discussed. And I will do so in due time. Obviously, once I raise the criticisms, I may find that I am wrong and Deutsch is correct, of course. Or my criticisms might resist criticism. We’ll see.

But the idea that we can and should feel free to criticize Deutsch is an important one that I want to establish right away as an appropriate area of discussion for the blog.

Now each of my ‘disagreements with Deutsch’ obviously needs a serious attempt on my part to present the criticism. Deutsch isn’t someone you just casually criticize in my opinion using ad hoc criticisms. One needs to really take the time to present one’s case. Because of that, in this post, I’m merely going to list (without argument or criticism yet) what my disagreements are. You may expect me to do future posts where I lay out the criticism in detail.

Here is my list:

  1. Minds must be nearly blank slates because of the theory of universal explainers and all genetic pre-disposition is swept away early on. Deutsch does claim something like this in his interview with Christofer on the Do Explain podcast, but he isn’t strong in the claim. I feel like it’s easy to think of counterexamples. Or maybe I’m misunderstanding the idea.
  2. Animals must be automatons because of the theory of universal explainers. I’m not actually sure Deutsch argues this. It’s a bit unclear. But many of his follows do argue this using quotes from him. I’m uncertain if this is true or not (maybe animals are automatons), but I’m not at all convinced this represents an implication of universality alone.
  3. Artificial Evolution creates no knowledge (As he argues in BoI p. 160-161) Certainly, artificial evolution is a non-universal knowledge creation process and creates very little knowledge. I originally thought Deutsch must simply be mistaken when he said it creates no “knowledge.” But since then it’s occurred to me that I may even agree with his underlying point but that he’s using a specialized definition for “knowledge” I’m not familiar with. If that is the case, then my actual disagreement isn’t over the claim that artificial evolution creates no knowledge, but over the fact that he’s violating word nominalism.
  4. The Turing Test is just the empiricist mistake. (BoI, p. 155). Well, it is the empiricist mistake, but quite a bit more than that. We can and do regularly determine if someone is a universal explainer or not by talking to them, and we’re quite adept (though fallibly so) at detecting creativity in someone else via such conversation. The interesting question is “why does the ‘Turing Test’ work so well despite making the empiricists mistake?”
  5. Deutsch’s reason in FoR for the principle of optimism (FoR p. 350-351) Yes, I know, he doesn’t argue the principle of optimism in FoR. He does that in BoI. But really it’s a bit more complicated than that. It’s specifically his argument in FoR related to what he later calls the principle of optimism that I find criticizable. Not the one in BoI.
  6. His BoI reason for dismissing the Omega Point (compare FoR p. 364 to BoI p. 450-451). Actually, that doesn’t mean I accept the Omega Point (I don’t because of other reasons) but specifically Deutsch’s argument for not accepting it simply seems incorrect to me by his own standards.
  7. An infinite expanding universe is the same as an infinite contracting universe in terms of implications for computation. (BoI chapter 18) They don’t seem like they have the same implications to me. One seems better than the other.
  8. That the hard-to-vary nature of beauty means it has to be non-parochial. I don’t believe beauty to be simply parochial. But I think that much that we experience as beauty is in fact subjectively a matter of taste even though it’s hard-to-vary — thus it must be parochial. This seems to follow from Deutsch’s own arguments that there are two kinds of beauty.
  9. That the hard-to-vary nature of morality means it has to be non-parochial. This one I actually agree with him on, but I think his argument must of necessity be incomplete and there is more important philosophical work waiting to be done here to fill the explanatory gap.
  10. All interesting problems are soluble. This isn’t so much a disagreement as I’m unclear what he means by this. It’s trivially true in some senses but less obviously true in others. The less obviously true version is far more exciting, however.
  11. Free will requires the multiverse. In FoR Deutsch claimed Free Will required the multiverse. But in his Constructor Theory paper and the forward for the audio version of FoR he walks that back a bit. So perhaps this isn’t a disagreement as of today. But in any case, I don’t believe the multiverse has any relevance to what we call “free will.”

What are your areas that you disagree with David Deutsch on? I know I’m not the only one because I’ve talked to many of you that have your own disagreements with him.

Oh, and this isn’t even my list of Disagreements with Popper.

2 Replies to “Disagreements with Deutsch”

  1. This is great. I haven’t read all of “Fabric of Reality” but I’ve read BoI twice.

    I question Deutsch’s assertion that people are “universal explainers”, because cognitive closure seems a possibility.

    I was too wondering about point #3 (“Artificial Evolution creates no new knowledge”). I think what he was saying is that current systems have not demonstrated that, not that it’s impossible. However it was curious. There are deep learning systems and evolutionary algorithms that have discovered new drug molecules. However in one of the most high profile cases of this it was found the signature molecule that was discovered was very similar to existing : https://www.nature.com/articles/s41587-020-0418-2

    His critique of AI is a bit outdated – obviously we have deep learning systems that learn from data, and not everything is hard coded in. However I think the amount of knowledge inserted into deep learning from things like architecture choice and how to process the data (which gets lumped under the term “inductive biases”) is greatly underappreciated. He is essentially correct that AI are not very creative, though. However, if you just define creativity in a combinatoric sense, which I think conforms with the common usage of the term, deep learning can be used for that (for instance GANs can be used to overlay the face of one person on another, and also change the voice as well.) The stuff created by or with AI so far is pretty shallow and uninspiring though.

    1. Your example of AI creating a new molecule is apt in that it’s ambiguous which is how most such examples of knowledge creation in AI are.

      I believe, however, that there are just too many examples of AI clearly creating knowledge that we’d not have but for the AI. One of the best examples is Alpha Go coming up with entirely creative new moves that no human had ever seen before. It strains credulity that this knowledge ‘came from the programmers’ (who, btw, were in a panic when the AI made that move because they thought it was a bug.)

      And I can easily list example after example of ML/AI creating new functions that no human knows how to create (face recognition, for example.) But in each of these cases the function is heuristics-based (as you point out in your paper) and therefore usually has no reach. So ML does create knowledge, but very little and it’s got no reach for the most part. (As you said, it’s generally ‘uninspiring’.)

      I also think it’s obvious that ML/AI has HUGE amounts of creative knowledge inserted into it by humans. The human does nearly all the work in creating the ML program in the first place. They spend considerable time trying out by trial and error which architectures work the best (related to your point about inductive biases), creating the right features, and creating training labels. The amount of knowledge coming for the human is overwhelming compared to the amount coming for the ML program.

Leave a Reply

Your email address will not be published. Required fields are marked *