How much computing power would we need to evolve computer via Darwinian evolution that can program itself ?

_{Denyse O'Leary

September 29, 2020

Artificial Intelligence, Darwinism, Intelligent Design, Mathematics, Natural selection}

Share: Facebook; Twitter; LinkedIn; Flipboard; Print; Email

You really do not want to think about this much computing power:

Yampolskiy then asks, could an immensely powerful computer succeed where all the others have failed? One way of approaching the question, he suggests, is by asking, what is the computational power of biological evolution? Some truly impressive figures result:
“If all the DNA in the biosphere was being transcribed at these reported rates, taking an estimated transcription rate of 30 bases per second, then the potential computational power of the biosphere would be approximately 10¹⁵ yottaNOPS (yotta = 10²⁴), about 10²² times more processing power than the Tianhe-2 supercomputer . . ., which has a processing power on the order of 10⁵ teraFLOPS (tera = 10¹²)…
“If we were to simulate 10²⁵ neurons over a billion years of evolution (longer than the existence of nervous systems as we know them) in a year’s run time these figures would give us a range of 10³¹-10⁴⁴ FLOPS.³⁹ –Yampolskiy Rv. Why We Do Not Evolve Software? Analysis of Evolutionary Algorithms. Evolutionary Bioinformatics. January 2018. Doi:10.1177/1176934318815906
These are not small numbers, even in the computing world. But even if these numbers were available, we need not assume that anything like human intelligence would necessarily result, via purely Darwinian means, if we ran only one simulation
Denyse O’Leary and Roman Yampolskiy, “Can computers evolve to program themselves without programmers?” at Mind Matters News

But read the fine print: We would need to run many trials of planets in parallel in order to simulate the real conditions in the universe. Yampolskiy concludes, ‘In fact, depending on some assumptions we make regarding multiverse, quantum aspects of biology, and probabilistic nature of Darwinian algorithm such compute may never be available.’”

That raises some questions about Darwinism in nature but we’ve talked about that before…

Further reading:

Can AI really evolve into superintelligence all by itself? We can’t just turn a big computer over to evolution and go away and hope for great things. Perpetual Innovation Machines tend to wind down because there is no universally good search. Computers are powerful because they have limitations.

Comments

Regarding the linked article in which researchers try to evolve an algorithm, I notice one difference between these attempts and what must have been the situation on the ground if Darwinism were true. In the simulation, survival to propagate the next generation is based on relative fitness, and not absolute fitness. For this reason the results of simulations like this have much less to say about the real world. After all, it is entirely possible, and has at times been the case, that none of the population survives an event, or that just about everyone survives.EvilSnack_{September 30, 2020
September
09
Sep
30
30
2020
06:46 AM
6
06
46
AM
PDT}

The basic question of intention comes in strongly here. Computers, like abacuses, are tools to fulfill human purposes. Each new REAL development, as opposed to ACADEMIC development, came in response to a human purpose. Most of the purposes were military or commercial or industrial. Improve the accuracy and speed of a cannon, improve the accuracy and speed of bookkeeping, improve the accuracy and speed of lathes. The self-evolving computer would need to have a purpose beyond simply proving a point, otherwise it wouldn't know which way to evolve. When human organizations outgrow their original purpose, they turn to crime. Parkinson's most important law. We're seeing it right now with the Public Health Officers who were no longer needed to improve sanitation, so they switched to committing a holocaust. In short, the self-evolving computer is guaranteed to do evil.polistra_{September 29, 2020
September
09
Sep
29
29
2020
10:35 PM
10
10
35
PM
PDT}

As to
Darwinian algorithm is incomplete or wrong—Finally, we have to consider the possibility that the inspiration behind evolutionary computation, the Darwinian algorithm itself is wrong or at least partially incomplete. If that was true, computer simulations of such algorithm would fail to produce results comparable with observations we see in nature and a search for an alternative algorithm would need to take place. This would be an extraordinary claim and would require that we discard all the other possible explanations from this list." - YAMPOLSKIY RV. WHY WE DO NOT EVOLVE SOFTWARE? ANALYSIS OF EVOLUTIONARY ALGORITHMS. EVOLUTIONARY BIOINFORMATICS. JANUARY 2018
He is correct that the Darwinian algorithm itself is wrong.
Top Ten Questions and Objections to 'Introduction to Evolutionary Informatics' - Robert J. Marks II - June 12, 2017 Excerpt: There exists no (computer) model successfully describing undirected Darwinian evolution. Period. By “model,” we mean definitive simulations or foundational mathematics required of a hard science.,,, We show that no meaningful information can arise from an evolutionary process unless that process is guided. Even when guided, the degree of evolution’s accomplishment is limited by the expertise of the guiding information source — a limit we call Basener’s ceiling. An evolutionary program whose goal is to master chess will never evolve further and offer investment advice.,,, There exists no model successfully describing undirected Darwinian evolution. Hard sciences are built on foundations of mathematics or definitive simulations. Examples include electromagnetics, Newtonian mechanics, geophysics, relativity, thermodynamics, quantum mechanics, optics, and many areas in biology. Those hoping to establish Darwinian evolution as a hard science with a model have either failed or inadvertently cheated. These models contain guidance mechanisms to land the airplane squarely on the target runway despite stochastic wind gusts. Not only can the guiding assistance be specifically identified in each proposed evolution model, its contribution to the success can be measured, in bits, as active information.,,, Models of Darwinian evolution, Avida and EV included, are searches with a fixed goal. For EV, the goal is finding specified nucleotide binding sites. Avida’s goal is to generate an EQU logic function. Other evolution models that we examine in Introduction to Evolutionary Informatics likewise seek a prespecified goal.,,, The most celebrated attempt of an evolution model without a goal of which we’re aware is TIERRA. In an attempt to recreate something like the Cambrian explosion on a computer, the programmer created what was thought to be an information-rich environment where digital organisms would flourish and evolve. According to TIERRA’s ingenious creator, Thomas Ray, the project failed and was abandoned. There has to date been no success in open-ended evolution in the field of artificial life.5,,, We show that the probability resources of the universe and even string theory’s hypothetical multiverse are insufficient to explain the specified complexity surrounding us.,,, If a successful search requires equaling or exceeding some degree of active information, what is the chance of finding any search with as good or better performance? We call this a search-for-the-search. In Introduction to Evolutionary Informatics, we show that the search-for-the-search is exponentially more difficult than the search itself!,,, ,,,we use information theory to measure meaningful information and show there exists no model successfully describing undirected Darwinian evolution.,,, ,,, if the fitness continues to change, it is argued, the evolved entity can achieve greater and greater specified complexity,,, ,,, We,, dub the overall search structure 'stair step active information'. Not only is guidance required on each stair, but the next step must be carefully chosen to guide the process to the higher fitness landscape and therefore ever increasing complexity.,,, Such fine tuning is the case of any fortuitous shift in fitness landscapes and increases, not decreases, the difficulty of evolution of ever-increasing specified complexity. It supports the case there exists no model successfully describing undirected Darwinian evolution.,,, Turing’s landmark work has allowed researchers, most notably Roger Penrose,26 to make the case that certain of man’s attributes including creativity and understanding are beyond the capability of the computer.,,, ,,, there exists no model successfully describing undirected Darwinian evolution. According to our current understanding, there never will be.,,, https://evolutionnews.org/2017/06/top-ten-questions-and-objections-to-introduction-to-evolutionary-informatics/
And while Yampolskiy is correct in his assumption that the Darwinian algorithm is wrong, Yampolskiy is incorrect in his belief that a 'alternative algorithm' will be able to the mimic the 'human level' intelligence that is necessary to create computer algorithms in the first place. The primary reason for this failure of algorithms to ever be able to mimic human level intelligence is obvious. Humans possess free will, algorithms do not. And it takes free will to be truly creative.
Algorithmic Information Theory, Free Will and the Turing Test – Douglas S. Robertson Excerpt: Chaitin’s Algorithmic Information Theory shows that information is conserved under formal mathematical operations and, equivalently, under computer operations. This conservation law puts a new perspective on many familiar problems related to artificial intelligence. For example, the famous “Turing test” for artificial intelligence could be defeated by simply asking for a new axiom in mathematics. Human mathematicians are able to create axioms, but a computer program cannot do this without violating information conservation. Creating new axioms and free will are shown to be different aspects of the same phenomena: the creation of new information.,,, The basic problem concerning the relation between AIT and free will can be stated succinctly: Since the theorems of mathematics cannot contain more information than is contained in the axioms used to derive those theorems, it follows that no formal operation in mathematics (and equivalently, no operation performed by a computer) can create new information. http://cires.colorado.edu/~doug/philosophy/info8.pdf The mathematical world – James Franklin – 7 April 2014 Excerpt: the intellect (is) immaterial and immortal. If today’s naturalists do not wish to agree with that, there is a challenge for them. ‘Don’t tell me, show me’: build an artificial intelligence system that imitates genuine mathematical insight. There seem to be no promising plans on the drawing board.,,, – James Franklin is professor of mathematics at the University of New South Wales in Sydney. http://aeon.co/magazine/world-views/what-is-left-for-mathematics-to-be-about/ The danger of artificial stupidity – Saturday, 28 February 2015 “Computers lack mathematical insight: in his book The Emperor’s New Mind, the Oxford mathematical physicist Sir Roger Penrose deployed Gödel’s first incompleteness theorem to argue that, in general, the way mathematicians provide their “unassailable demonstrations” of the truth of certain mathematical assertions is fundamentally non-algorithmic and non-computational” http://machineslikeus.com/news/danger-artificial-stupidity Robert Marks: Some Things Computers Will Never Do: Nonalgorithmic Creativity and Unknowability – video https://www.youtube.com/watch?v=Cm0s7ag3SEc Observation of Unbounded Novelty in Evolutionary Algorithms is Unknowable - 2018 Eric Holloway and Robert Marks Abstract. Open ended evolution seeks computational structures whereby creation of unbounded diversity and novelty are possible. However, research has run into a problem known as the “novelty plateau” where further creation of novelty is not observed. Using standard algorithmic information theory and Chaitin’s Incompleteness Theorem, we prove no algorithm can detect unlimited novelty. Therefore observation of unbounded novelty in computer evolutionary programs is nonalgorithmic and, in this sense, unknowable. http://robertmarks.org/REPRINTS/2018_Observation-of-Unbounded-Novelty.pdf
bornagain77_{September 29, 2020
September
09
Sep
29
29
2020
04:37 PM
4
04
37
PM
PDT}

You must be logged in to post a comment.

Leave a Reply