I think that Scott Aaronson is an intelligent man but over the years, I have pointed out that he's staggeringly irrational – and therefore not very intelligent – in a wide variety of issues. Those include most of the political questions which is unsurprising but Aaronson also holds strong beliefs that directly touch his expertise but they're utterly irrational, too.
Undergraduate student Ewin Tang
One of his beliefs is – it's almost as general as what I am going to say – that whenever some fast (polynomial) algorithms haven't been found so far, they will never be found. The most explicit "substatement" of this type is Aaronson's statement that \(P\neq NP\) must be treated basically as a fact even though neither \(P\neq NP\) nor \(P=NP\) has been proven.
In numerous blog posts, I pointed out that every pair of "qualitative propositions" in pure (especially) discrete mathematics that haven't been proven to be strictly equivalent must be considered uncorrelated. In other words, no partial, "rather good", evidence may ever exist. It means that when you formulate a proposition about discrete portions of mathematics, fair and rational mathematicians simply must remain open-minded and allow the research into "possible better evidence that \(P=NP\) as well as possible better evidence that \(P\neq NP\)".
Aaronson disagrees and claims that he can behave as if \(P\neq NP\) has been proven even though it obviously hasn't been proven. In some sense, he is an arrogant aßhole who just isn't willing to see his own limitations. He wants to believe that he may guess the truth values of propositions in computer science without doing any actual work.
Let me tell you something. Aaronson doesn't have this supernatural ability. No mathematician or another human being has it, either! Everyone who believes in similar supernatural abilities of some men like Musk, Aaronson, or other "gurus" is a complete and hopeless idiot.
\(P=NP\) hasn't been proven yet but the Quanta Magazine's Kevin Hartnett wrote a story about some cute smaller yet analogous advances
The 18-year-old Ewin Tang (who should only start his graduate studies in a few months) was assigned a task by Scott Aaronson in 2017:
Prove there is no fast classical recommendation algorithm, and thereby confirm Kerenidis and Prakash’s quantum speedup is real.I must say what the fast classical recommendation algorithm is. There is a matrix \(M\times N\) remembering which of the \(M\) Netflix users have watched which of the \(N\) Netflix videos. Now, a new user is added with his list of watched videos and there is some task – probably with much more refined rules that I won't reproduce here – to quickly find similar Netflix videos and similar users and try to figure out whether it's natural for the particular user to watch a particular movie.
There must be some rules that decide whether the recommendation is good. I don't claim to understand these rules now.
However, Kerenidis and Prakash have decided to sell this mundane, practically useful recommendation problem as something that "quantum computers are good for". (There has been a lot of prejudice in that work – they just decided to construct a result of a certain type.) Well, they wrote down some quantum algorithm. But the quantum computers are only useful if you really need them. Because classical computers are so much cheaper and easy to produce today, the quantum computers are only very valuable in practice in this context if the classical computers can't solve the problem comparably quickly.
Here, Scott Aaronson inserted a similar belief as his belief that \(P\neq NP\) – the only relevant difference is that this particular Netflix recommendation problem is less famous, less general, and less far-reaching than \(P=NP\). But Aaronson knows what the answer is before a proof appears in one way or another, doesn't he? And as I have mentioned, Aaronson's answer is always that no new much faster algorithms ever exist if they hadn't already been found. So he could directly ask Ewin Tang to prove a no-go theorem, Aaronson believed.
Instead, Ewin Tang has proven a yes-go theorem (arXiv) and proved that Aaronson is full of šit. Tang has found a fast classical algorithm that Aaronson was not only incapable of constructing. Aaronson was incapable of imagining that such an algorithm could exist. I think that in this sense, it's fair to say that Ewin Tang has proven to be better than Scott Aaronson in this type of problems not just by one level but at least by two levels. Aaronson can't even see how something like that could exist. Tang can not only envision how it could exist – he can actually build it!
Well, the difference is actually at least three levels because Tang could not only envision and write the algorithm; it was also the simplest task among those that Aaronson has listed for him!
You know, if you read at least the title of Tang's paper, he has found something more general. His algorithm is "quantum-inspired" so he was capable of seeing that the tricks that were previously presented as "really demanding a quantum computer" actually didn't demand it, and these memes actually distorted the difference between the quantum and classical computers, and he could replicate those "clever ideas presented as dependent on a quantum computer" on a classical computer. Analogous algorithms "inspired by something faster" could very well exist even to prove \(P=NP\). One probably can't get there from Tang's result very directly – but on the other hand, Tang's method to "get inspired by some faster machines" and replicate their functions by simpler ones could have many more implications than the particular Netflix recommendation problem. I wrote this paragraph in order to convey my feeling that this is probably more than an isolated technical result – it's the kind of paradigm shift that is normally enough for someone's whole career, and may be more important and far-reaching than what Scott Aaronson has ever found.
I think that Ewin Tang won't be celebrated by this computer science and journalistic establishment too much because he has shown that people may defeat group think and prove that some people are more arrogant than intelligent along the way. And that's so inconvenient! Some people could start to heretically think that self-anointed experts who know it all – and whom the media describe as omniscient beings – actually don't know almost anything they claim to know. And it must be prevented! Meanwhile, the people who have some ethical decency in science know that Aaronson's task
to prove a no-go theoremwas a failure of Aaronson's scientific integrity. One simply shouldn't push the students in one direction if the right direction is unknown. The only acceptable task for the student is
to prove that the fast classical algorithm is possible, or to prove that it is impossible.If someone pushes the students in a direction picked by the adviser's pure prejudices (and perhaps a long-term agenda because "results of one type only" make Aaronson look better), it's bad. Thankfully, some of this scientifically dishonest bullying happens to lead to a happy end – a proof that the likes of Aaronson are mostly prejudiced, arrogant bullies who are full of šit.
And that's the memo.
P.S.: I am annoyed not only by Hartnett blindness to the key moral lesson of this story. I am also annoyed by the general negative tone of the article. This negative tone really indicates that Tang's achievement is considered an inconvenient truth. You know, Hartnett spins Tang's result as a negative result that "weakens the quantum computers". But this is really a demagogy. Tang's work deals with classical algorithms so, just like Bell's theorem, it doesn't really say anything about quantum computers at all.
Both theorems are only relevant within classical physics. Bell's theorem was not new at all – it re-showed the well-known fact that local classical theories couldn't be compatible with the observed (often very strong) correlations in the microscopic world. On the other hand, Tang's result is a very positive and totally new classical algorithm – Tang has shown that classical computers are more powerful than previously thought. So it is a totally obviously "positive" result, as every particular construction of a program that "does something useful". The result also shows that "the ability of quantum minus classical computers" is smaller than previously thought – but the "absolute" properties of the quantum computers haven't been changed. It's only the classical term that was changed by Tang – and he made the classical computers stronger than before!
The spinning of the result as a negative one if not a disappointment is demagogy and I think that this demagogy is driven by certain filthy and dishonest political goals.
Another P.S.: I checked Aaronson's blog and he wrote something about this paper ("Customers who liked this quantum recommendation engine might also like its dequantization"). There are some extra facts and stories. One thing that I couldn't overlook was that not only Aaronson asked the student to prove the wrong assertion. When the student proved the correct one – after wasting a year with the incorrect, Aaronson's strategy – Aaronson worked hard to delay the publication (even) on the arXiv. The student had to undergo some critical 4-hour defense among some other experts who ultimately said it looked solid. Just like the biased formulation of the problems for students is wrong, the selective pressure against the publication of results of a certain type is wrong, too. In comments #30, #32, Aaronson makes it clear that he would try to suppress a proof of \(P=NP\) even more than that (more than four hours of fire for the student). Sorry but in that case, you're a biased, self-serving ideologue suppressing the scientific progress, Scott, who should pay to the universities and the mankind for the damage you are causing.