Thursday, May 23, 2013

Quantum Computing Fast and Slow

I just read two very different science books, Daniel Kahneman's Thinking, Fast and Slow and Scott Aaronson's Quantum Computing since Democritus. Not much to connect the two except both deal to some extent about probability and computation and I want to write a blog post for each chapter, for much I disagree with both authors. But that's what makes them so much fun, so rare to find science-oriented books both worth reading that have the guts to say things that one can disagree with.

In full disclosure, Scott and I agree that he would post about my book if I wrote about his but what a deal. Scott's book is a pleasure to read. He weaves the story of logic, computation and quantum computing into a wonderful tour. You can get an idea of Scott's style by how he explains how he will explain quantum.
The second way to teach quantum mechanics eschews a blow-by-blow account of its discovery, and instead starts directly from the conceptual core - namely, a certain generalization of the laws of probability to allow minu signs (and more generally, complex numbers). Once you understand that core, you can then sprinkle in physics to taste, and calculate the spectrum of whatever atom you want.
He approaches the whole book by this philosophy.  Every now and then he moves into technical details that are best skipped--either you already know it or will get lost trying to follow. But no problem, the story remains. You need to appreciate Scott's sense of humor and his philosophical tendencies, and he does get way too philosophical near the end, particularly a strange attack on Bayesian that involves God flipping a coin. At the end of the book Scott contemplates whether computer science should have been part of a physics department but after one reads this book the real question is whether physics should be part of a CS department.

Kahneman gives a readable tour of behavioral economics with a variety of examples, though I don't agree with his interpretation of many of them. His fast and slow refers to decisions we make instinctively and quickly (like judging a person based on first impressions) versus more slow and deliberative (like multiplying numbers). There is a computer science analogy, in that his fast refers to what we can do with machine learning, simple trained models to make quick judgments that occasionally gets things wrong. I'm not a huge fan of behavioral economics, but it is useful in life to know the probability mistakes people make so you can avoid making them yourself. The wikipedia article has a nice summary of the effects mentioned in the book.

While these two books cover completely different areas, the themes of probability and computation pervade both of them. One simply cannot truly understand physics, economics, psychology and for that matter biology unless one realizes the computational underpinnings of all of them.

10 comments:

  1. "Kahneman gives a readable tour of behavioral economics with a variety of examples, though I don't agree with his interpretation of many of them"

    Lance, what's an example where you disagree with his interpretation?

    ReplyDelete
  2. Yes Lance, for all this griping you ought to provide at least one substantive critique. If you don't like philosophy you can just say so, without complaining that Aaronson gets "way too philosophical." If it's not just a matter of taste, but you actually disagree with something in his argument, why don't you make that clear. And if you are "not a huge fan of behavioral economics," is it because of an actual scientific disagreement or what?

    ReplyDelete
  3. I can see I came out much more negative than I thought, as I do agree with most of what is in both books and I really enjoyed reading the books.

    Why I'm not a big fan of behavioral economics is a topic I'll return to in a future post.

    ReplyDelete
  4. There's a recent email of Kahneman (regarding a few "priming effect" results) that might justify some amount of skepticism.

    ReplyDelete
  5. Let the physicists first give a Nobel prize to Grover / Simon / Shor - or Bennett -- we'll talk about department take-overs after that.

    ReplyDelete
  6. >There is a computer science analogy, in that his fast refers to what we can do with machine learning, simple trained models to make quick judgments that occasionally gets things wrong.

    While that works as an analogy it fall flat when taken literally. The fast thinking is common sense, sarcasm/humor detection, object categorization, intuition, pattern recognition on high dimensional data; exactly those things which ML and AI have found very difficult to replicate. The Slow system based on probabilistic and logical reasoning, consistency, rationality + convex preferences, and calculation are much more of a breeze for ML.

    ReplyDelete
  7. Theory of computer science is naturally a part of mathematics. Not physics. Both in practice and in principle (many CS theoreticians are in math departments; especially in Europe).

    ReplyDelete
  8. There has been strong relations between mathematicians and computer scientists, but mainly with theoretical computer scientists. Computer science is more of an algorithmic perspective. These days you can find people with graduate degrees in not just mathematics but almost any field working in computer science departments, from health sciences to biology to economics to psychology. And often mathematicians suffer more as they start working in computer science than others because of leaving in their Platonic world. Mathematics lacks the pragmatic computational perspective of computer science which is arguably the unifying principle of what we refer to computer science. In that sense, it is kind of misleading to call computer science part of mathematics. A mathematician may work in economics or physics or biology or whatever as mathematics is the language of the science. The sciences has become so dependent on mathematics that doing any advanced science without mathematics is completely impossible. But if we look at the trend we see that a similar trend in adoption of algorithms in other sciences, even in mathematics to lesser extend. We see a increasing trend in use of computers and mathematics. In a few decades the use of computers and knowledge of algorithms will become the norm in mathematics. A mathematician may think any person who uses mathematics or proves theorems is a mathematician. I would consider any person who has the pragmatic computational perspective and a sound understanding of algorithms a computer scientist. Other scientist have similar encompassing definitions of their fields. There is no point in arguing about someone being a mathematician or computer scientist or physicist or economist or ... unless one has a huge ego.

    ReplyDelete
  9. Uhm I'd actually say I learned a lot from the details in Scott's notes, when doing undergrad...

    ReplyDelete
  10. I've been calling "Quantum Computing Since Democritus" a 'pop-science' book for very smart people. The technical details are possible to follow and some of the earlier ones can be pretty helpful (QCSD was my first exposure to the Cook–Levin theorem and I rather liked the explanation), but for some of the later complexity results, you'd better have a textbook with you if it's the first time you're hearing about interactive proofs and the like.

    ReplyDelete