The map of the methods of optimisation (types of intelligence)

post by turchin · 2016-09-15T15:04:31.478Z · LW · GW · Legacy · 10 comments

Contents

10 comments
Optimisation process  is an ability to quickly search space of possible solution based on some criteria. 
We live in the Universe full of different optimisation processes, but we take many of them for granted. The map is aimed to show full spectrum of all known and possible optimisation processes. It may be useful in our attempts to create AI.

The interesting thing about different optimisation processes if that they come to similar solution (bird and plane) using completely different paths. The main consequences of it is that dialog between different optimisation processes is not possible. They could interact but they could not understand each other. 

The one thing which is clear from the map is that we don’t live in the empty world where only one type intelligence is slowly evolving. We live in the world which resulted from complex interaction of many optimisation processes. It also lowers chances of intelligence explosion, as it will have to compete with many different and very strong optimisation processes or results of their work. 

But most of optimisation processes are evolving in synergy from the beginning of the universe and in general it looks like that many of them  are experiencing hyperbolic acceleration with fixed date of singularity around 2030-2040. (See my post and also ideas of J.Smart and Schmidhuber) 

While both model are centred around creation of AI and assume radical changes resulting from it in short time frame, the nature of them is different. In first case it is one-time phase transition starting in one point, and in second it is evolution of distributed net.

I add in red hypothetical optimisation processes which doesn’t exist or proved, but may be interesting to consider.  I mark in green my ideas. 

The pdf of the map is here: 






10 comments

Comments sorted by top scores.

comment by g_pepper · 2016-09-15T16:04:29.035Z · LW(p) · GW(p)

Interesting list, thanks for posting it.

One question: You describe a quantum computer as "hypothetically more powerful than total calculation power of the universe". What does that mean?

On the one hand, even a deterministic Turing machine is hypothetically more powerful than the total calculation power of the universe, since a hypothetical Turing machine has an infinite tape. So your statement would appear to be trivially true (because you said "hypothetical").

On the other hand, it seems that no actual quantum computer can be more powerful than the total calculation power of the universe, since any actual quantum computer that we were to build would be part of the universe.

So, what does this statement really say regarding the power of a quantum computer?

Replies from: turchin
↑ comment by turchin · 2016-09-15T16:40:19.813Z · LW(p) · GW(p)

My statement mostly repeat the claim which I read somewhere that computational power of QC of several thousand qubits will be stronger then computational power of classical computer in the size of all Universe.

I can't find the link now, but maybe will find it later.

Replies from: g_pepper
↑ comment by g_pepper · 2016-09-16T02:59:55.819Z · LW(p) · GW(p)

I read somewhere that computational power of QC of several thousand qubits will be stronger then computational power of classical computer in the size of all Universe.

I have read this sort of claim as well. However, I recommend skepticism; there has been a lot of hyperbole in the popular press regarding the potential power of quantum computing. A good source of objective information about the potential and limitations of quantum computing is Scott Aaronson's blog, Shtetl-Optimized. Scott was a professor at MIT and is now at the University of Texas at Austin - he is very knowledgeable about QC and, equally important, he is an engaging writer that can make complex topics (reasonably) clear to non-specialists. He has spent a bit of time on his blog debunking some of the wilder claims regarding the power of QC.

comment by g_pepper · 2016-09-15T16:17:55.126Z · LW(p) · GW(p)

You say of the Internet (search engines, database engines, exchange media, distributed calculations) that it is "self-aware". That seems like a strong claim to make with no further explanation. How did you arrive at that conclusion?

Replies from: turchin
↑ comment by turchin · 2016-09-15T16:32:12.451Z · LW(p) · GW(p)

I meant that Internet knows about its own existence in form of scientific research which studies Internet and is published in the Internet. It is also self-aware by the mind of people who use it, nothing mystical here.

I think that idea of "consciousness" should be broken in several separate ideas:

  1. Self-aware - something simply including model of itself.

  2. Have qualia - ability to have own qualitative subjective experiences

  3. Field of consciousness - united perception field which integrate different modalities in one group of things actual now. (For example mainstream mass media play such role for civilization, but they don't have their own qualia).

Replies from: g_pepper
↑ comment by g_pepper · 2016-09-15T17:43:19.618Z · LW(p) · GW(p)

I meant that Internet knows about its own existence in form of scientific research which studies Internet and is published in the Internet. It is also self-aware by the mind of people who use it, nothing mystical here.

I suspect that this idea of self awareness differs considerably from what most people think of by self awareness. For example, your criteria would seem to classify the following things as self-aware:

  1. A printed edition of the Encyclopaedia Britannica, if that edition contains an article about the Encyclopaedia Britannica.
  2. A corporate IT department's Asset Management System (which is a computer-based repository of information about applications that the department manages), if it contains information about the Asset Management System.
  3. Any literary work which refers to itself (e.g. John Gay's The Beggar's Opera, in which two of the characters discuss the opera itself at various points in the opera, or Denis Diderot's novel Jacques The Fatalist and His Master, which contains numerous asides in which the narrator discusses Diderot's novel).

Do you consider those things to be self aware?

Replies from: turchin
↑ comment by turchin · 2016-09-15T21:51:30.432Z · LW(p) · GW(p)

Self-aware is about agent which has its own model and able to construct it. Wiki said: "Self-awareness is the capacity for introspection and the ability to recognize oneself as an individual separate from the environment and other individuals." https://en.wikipedia.org/wiki/Self-awareness

Britannica's printed edition is not agent, but its authors were agents and they knew that they were writing an encyclopaedia and that they are its authors.

Internet include agents which thinks about internet and plan its development.

But evolution (until human appeared) didn't include or needed any idea about evolution and nicely worked without it.

What is your understanding of the idea of "self-awareness"?

Replies from: g_pepper
↑ comment by g_pepper · 2016-09-15T23:55:24.909Z · LW(p) · GW(p)

What is your understanding of the idea of "self-awareness"?

The definition that you supplied: "Self-awareness is the capacity for introspection and the ability to recognize oneself as an individual separate from the environment and other individuals" sounds about right to me. By that definition, the internet would only be self-aware if you include the internet's architects and users as part of the internet (as you did above). It is not surprising that a system that contains rational agents as a part of itself is self-aware; many an inanimate object could be considered self-aware if we consider its builders and users as a part of it - for example the three examples I listed previously would be self-aware given that consideration. But, in the case of the internet and in the cases of my examples, it is the intelligent agents that provide the self-awareness; so self awareness is an attribute of the intelligent agents rather than of the internet per se (or the encyclopaedia per se, etc.).

comment by tukabel · 2016-09-16T19:43:08.831Z · LW(p) · GW(p)

Seeing us humanimals and our limitations, it should be quite obvious we are just another stage (last biological) in the Evolution of Intelligence - and we are here just to create our non biological succesor before we manage to destroy ourselves (and maybe the whole Earth).

Because what concerns the organisation of the society, nothing much changed over the millenia, still the same good old animalistic forces (politico-olicharchical predators "eating" the herd of mental herbivores, not slaves anymore, but voters in the demogarchical system - demos votes, but candidates are selected and paid by oligarchy). No surprise, as the brain haven't changed much either.

Just don't be fooled by the memetic supercivilization of less than tenth of percent on top of that humanimal "noise" - which gives humanimals everything it creates - for free and without any control ( science and ideas in general, and the resulting inventions and powerful technologies) - only to see everything regularly misused by the ruling predators (already Nukes were over the limit and nanobot stuff will be clearly too much to handle - imagine DYI nuclear grenade for a dollar that any teenager can assemble at home... nanobots will be orders of magnitude worse... and there are certainly many more risks one can think of).

Will Singularity manage to make it in time?

That's a good question!

Forget "friendly AI", it's the other way round - humanimals are the problem.

Replies from: turchin
↑ comment by turchin · 2016-09-17T09:59:11.631Z · LW(p) · GW(p)

Technically where could be two types of successors:

1) More complex and interesting, like post-humans and some types of strong AIs

2) Boring, and just enough complex to kill us. Examples are Grey goo, paperclip maximizers, or SETI-attack AI.

The first type may be still non-friendly and dangerous, like Homo sapiens were for Neanderthals, but it will continue evolution of intelligence on Earth. The second type is the End.