Tuesday, July 25, 2017

Zuckerberg's understanding of AI doomsdaylogy is more limited than Musk's

...unfortunately, AI doomsdaylogy is a moronic cult, not a credible field of science...

One week ago, Elon Musk repeated his Luddite demand to regulate and combat the AI existential threat before it's too late.

We're facing an imminent uprising of the robots, we were told, so a new IPCC – the Intelligent People-like-robots Combating Committee – has to be established. Everyone who has some dangerous robot, e.g. a self-driving car, an intelligent thermometer, a smartphone, a fridge, or a mixer, will have to euthanize them or report every usage of them to the IPCC. New production of dangerous robots will be banned to fight against the existentially threatening uprising.

I am not quoting Musk exactly but I am sure that I am accurately conveying the basic spirit. He is a Luddite. An insane one. It's rather amazing for an entrepreneur who is considered high-tech to be a Luddite when it comes to the Artificial Intelligence but it just happens to be the case.

Now, another billionaire who happens to be wealthier and more famous – and who was forecast to become George Soros v2.0, something that I find infinitely more threatening than the uprising of the robots because it could delay the dieout of George Soroses substantially – happens to have a view on the AI existential threat that is closer to mine than Musk's.

We learned that
Zuckerberg blasts Musk warnings against artificial intelligence as 'pretty irresponsible'
While using a low-tech grill on his backyard, Zuckerberg used the term "irresponsible" because in his optics, safer-than-humans self-driving cars and more-competent-than-humans physicians are going to arise from the Artificial Intelligence and Elon Musk wants to rob us of these advances and kill millions of people who would otherwise live.

(Musk's worry about self-driving cars is that lots of them will be hacked and crashed simultaneously. Great, possible, but it's also a threat that will surely be addressed by some features if they're ever sold en masse, right? I don't think that the threat differs much from the threat that someone poisons the tap water in big cities.)

Clearly, the two men disagree. Elon Musk made us sure that an interaction of the two wealthy men dedicated to this topic has taken place:

My main point has already been expressed in the title: I actually think that Elon Musk is right: Musk's understanding of the teaching is deeper than Zuckerberg's. And doomsday prophets can often repeat the list of doomsdays more correctly than those who criticize the doomsday prophesies.

The catch is that the "field" in which Zuckerberg's knowledge is "limited" relatively to Musk's isn't quite a prestigious scientific discipline at all. In fact, it is not a field of science at all. To be more accurate, it is a moronic cult, a pile of feces and stupidities that some overgrown children including Musk have made up and they are scaring each other. They seem to be as excited as children with the masks of vampires who are hiding in the closet to scare their 5-year-old friends.

Dalibor Janda, Everything Onto Mars, 1985. Around 2005-2006, robots will shoot the horses and all of us to Mars. Fail.

I believe that my point is much deeper than anything that is normally discussed in these contexts. There are lots of people and cliques of people who invent their own new "interdisciplinary sciences" and they automatically become world's top experts in these new sciences. According to the rules of these new interdisciplinary sciences, they may be the best experts, indeed – it's because they have written down the rules for themselves. But unfortunately, they don't ask whether their new "interdisciplinary sciences" are any good or whether better answers may be obtained by using the methods of the old sciences in which these prophets are... not so good.

(Let me mention that some tension between Musk and Zuckerberg may exist for AI-unrelated reasons. Last year, Musk's SpaceX was scheduled to deliver Zuckerberg's satellite on the orbit that would guarantee the Internet for much of Africa. Musk's rocket exploded, destroyed the satellite, and Zuckerberg didn't try to suggest that "everything was fine".)

Here's a newer tweet that may be used as an example of the "technical wisdom" that Musk "understands" very well while Zuckerberg's understanding may be "limited":

Musk praises a text by someone else – he must praise him because he is a fellow brother believer in the AI Armageddon cult – but emphasizes that he is even smarter than that author because he also knows that we face a "double exponential rate of improvement". It's because both hardware and software will experience an exponentially accelerating rate of improvements. (I improved the formulations a bit, all these people talk like vague moronic šitheads.)

Mirai (JP-flavored CZ band): The Robotic Baby. Hey, I am the robotic baby, I have the same program as you.

Now, this is a classic example of a moron who doesn't understand the basic things but who uses some would-be advanced technical jargon in order to look intelligent and impressive (to the other morons). If he had a working high school knowledge of mathematics or if he at least spent 30 seconds with Google and Wikipedia, Musk could have known that a double exponential function is a function with the general formula\[

\Large f(x) = a^{b^{x}}.

\] The function \(\exp(\exp(x))\) is a simple example. This is indeed very quickly increasing. For example, if you substitute \(a=b=10\) and \(x=100\), the value \(f(x)\) is equal to the googolplex, a very high number, indeed.

The off-diagonal elements of the density matrix in decoherence are dropping doubly exponentially,\[

\rho_{ij} \sim 1/\exp[A\exp(t/t_0)]

\] because the number of degrees of freedom where the original observable gets imprinted grows exponentially and each of them adds a multiplicative factor of the order \(1/\exp(A)\) from the inner product of states resembling "the traces of dead and alive" to the matrix elements.

But has Musk actually found a double exponential in his AI doomsdaylogy? Well, he hasn't. If software grows like \(\exp(t/t_0)\) and hardware grows like \(\exp(t/t_0)\), then the sum of software and hardware grows like \(2\exp(t/t_0)=\exp(\ln 2 + t/t_0)\). It's still just an exponential, not double exponential, growth.

Even things that depend on the the product (multiplication) of software and hardware will grow like \(\exp(2t/t_0)\) which is still an exponential growth, not a double exponential one, although the "timescale of doubling" is shorter by a factor of two. Nothing qualitative changes about the exponential evolution.

Aside from the sum and the product, you may try to invent some justifiable "combination of hardware and software" in your effort to get a double exponential function. You will fail. The double exponential function accelerates way too quickly. You can't achieve this rate of accelerated progress just by some clever separation of your engineers to hardware and software experts.

In fact, in the real world, even the exponential function seems to be a way too optimistic Ansatz for the rate of progress. If you look at the actual numbers in any exponential-like expansion, you will find out that the actual growth is slower i.e. the timescale of doubling is getting longer.

If Musk meant the "double exponential" in the correct sense, the statement is just pure crackpottery. No accelerated progress in the real world can ever be made double exponential.

Now, he may count himself as a better expert in the AI doomsdaylogy because the statement "the AI progress is double exponential" is, by definition (that he wrote down), one of the characteristics of the spiritual leader of the scientists in the AI doomsdaylogy.

But there exists another, old-fashioned interpretation: Elon Musk is one of the most visible crackpots who is just talking complete nonsense – contradicting basic properties of simple mathematical functions – in order to justify his predetermined anti-AI Luddite clichés. And 10.6 million full-blown imbeciles (based on his number of Twitter followers; I should add the additional 80 million morons who weren't able to create a Twitter account yet) are impressed with his big statements although these statements are seen to be pure bunk by everyone who has at least some clue about the words.

Now, if you think that I am too innovative when I point out the growth of the self-anointed experts in these new pseudosciences and that this growth is an absolutely new phenomenon, you may be disappointed. It's not so new. In an 1981 BBC interview "The Pleasure of Finding Things Out", Feynman made the exact same statement. One must only watch the world and realize that many things and many people perfectly fit certain descriptions. Many of these descriptions were made decades or centuries ago but they remain perfectly relevant and valid.

Just like the climate doomsdaylogists, Elon Musk is a self-anointed expert in another similar field – the AI doomsdaylogy. By its definition, the field studies the Armageddon that will be caused by the Artificial Intelligence. And he can offer lots of deep wisdom to you, e.g. that you may produce a googolplex of chips by exponentially increasing software and hardware industry over 100 units of time. Well, you just can't. And because of the defining thesis, the AI doomsdaylogy is a moronic cult and it will always be one. The most serious problem of the AI doomsdaylogy just can't be fixed because the "fixed" discipline would no longer be AI doomsdaylogy.

We have a growing spectrum of these self-anointed experts in new disciplines that really don't exist according to any rational or scientific evaluation of the situation. We have lots of pompous fools who are full of šit and Elon Musk is sadly one of them.

No comments:

Post a Comment