Posts Tagged ‘evolution’

The Optimality of Morals

Posted in Philosophy on May 4th, 2009 by Noldorin – 8 Comments

This post essentially follows on from the Notes on Kant post by David, which having prompted rather a lot of comments and one or two conversations, led to a few interesting conclusions on the subject of morality. Here, however, I mainly intend to express my own views and conclusions on the nature of morals (though David seems to be of much accordance, at least in his end point). I’ll leave it to anyone who wishes to comment to counter my points.

To start, I should mention that my own philosophy on morals seems to be largely in accordance with rule utilitarianism (or at a minor variant of it). What follows is pretty much the set of ideas that guided me to the eventual conclusion regarding optimality. Specifically, I argue that an action is moral if it is beneficial to either oneself or society (or both) and not detrimental to the well-being and continuation of the society as a whole. The intentions of the individual performing the action must also satisfy these conditions if the action is to be deemed moral, else the action must be morally neutral at best. Importantly, this specification implies that choices made with self-interest in mind can be moral so long as the communal benefit is non-negative. This becomes quite obvious given the assumption that the well-being of individuals in general contributes to the well-being of society (at least in an indirect way). Note that the arbiter in all these cases must be hypothetical as well as purely objective (nature itself, if you will), meaning that even though a certain action may be considered immoral as a consensus of society, it may nevertheless be neutral or even moral in actuality. Saying this, in a well-functioning and successful society, there would seem to be a general requirement that the judgement of moral worth of actions is reasonably accurate in a high proportion of situations.

Considering these points and their commonalities is primarily what led me to believe that when you boil everything down, morals are nothing but an approximation to optimality of society. Now as soon one mentions optimality, the question of a measure automatically follows. Of course, most people probably have some vague notion of what an “optimal society” is, but since the aim here is to be as formal and specific as possible, I really need to define a cost function, at least in loose terms. At this point, I would imagine that the opinions of most people would tend to diverge rapidly. Some would reason that the cost function is purely dependant on well-being/happiness/pleasure (whether more for the individual or society separates the hedonists from the utilitarians), while others would state the straightforward biological (yet to many cold and unpleasing) view that optimality is but a measure of the size of the population and thus the continued ability to self-replicate. Finally, the more religious among us might contend that optimality is simply the perfectly obedient following of teachings passed down to us by God. In essence, this cost function is nothing other than the “meaning of life” (in the widest possible sense) – something that may never be defined, and certainly not a discussion I’m going to include in this post! Whatever view people wish to take, I believe that the basic statement that “morals are an approximation for optimality” holds well in all cases. Like the nature of this optimality, the mechanism by which the concept of morality has been instilled in us (evolution, creationism, Spinoza’s God, or whatever) is also open to debate, but nonetheless is not able invalidate this theory in itself, at least in the way I see it. Yet all these concepts of optimality surely do have similarities. Another important feature of the cost function is that parameters should be not only the current state of humanity and the world, but also the states at points in the future (perhaps stretching infinitely far ahead in time). In the end, I think I can say that I do personally feel reasonably content with this definition of morals (albeit most likely an incomplete one). In my mind one cannot proceed any further in a formal definition without invoking reasons akin to the “meaning of life” such as those just mentioned – all very contentious or speculative and therefore not terribly helpful as bases for any fundamental theories, in my view.

Now to properly round off these theories, I ought to explain in more detail what I mean by an “approximation” to optimality. In my conclusion, I came to realise that moral principles (stressing the fact these principles are what are percdeived by men to be moral) may not necessarily lead to optimality in all cases, however you want to define the term. There can clearly exist an action performed at a certain time that to the best of everyone’s knowledge appears moral, yet has long-term ramifications that are generally negative – an unlikely case perhaps, but a quite possible one irrespective. It then follows that either a) the action cannot actually be considered absolutely moral because of these consequences in the (distant) future, or b) the motivation/action is perfectly moral (given the limitations in the nature of the actor) but not necessarily optimal under whatever cost function you choose. I would think option a) would appear immediately quite wrong, since it would contradict the idea that moral actions can be knowingly performed, which just silliness under any definition. We are then forced to accept option b), in other words that morals are only approximation guides to optimal behaviour and therefore optimal results (though most likely very good approximations). The next question is: does there exist any better approximation to optimality than morality? Of course, omniscience combined with perfect reasoning might be considered the ideal way to produce an optimal society and would seem to appear more “useful” than morals, but this is something which we as humans fall short of by an effectively infinite margin. Let’s suppose the evolutionary viewpoint for a moment here, simply because it leads to a curious hypothetical case. Is there a point at which we as a species may become intelligent enough to produce a more optimal society purely by reasoning? Is there a threshold at which it suddenly becomes more sensible to follow pure reason than moral instincts, or will both always be required to varying degrees? I’ll leave those questions unanswered, since they are largely side points to my cse, though it does at least highlight the issue in relation to current and past societies. Now surely no-one would argue that high-level reasoning can’t be used alongside (augmenting?) instinctual/inherent morals (indeed it is arguably a more “intelligent” form of morals that makes mankind particularly moral). Nevertheless it should be quite clear in looking around ourselves that there are dangers in the outcomes of limited reasoning overriding recognisably moral behaviour. Perhaps we can even attribute immoral behaviour at its root to to the arrogance or egotism (by this I really mean selfishness) of humans – whether in valuing their own well-being over that of society as a whole or their own powers of reasoning over moral principles. The latter is perhaps a more unintentional form, due to the failure of limited consciousness to realise its own limitations in forseeing complex (or at times even relatively simple) consequences of actions. To explain what might appear to be the widespread existance of the dominance of egotism in individuals’ personal cost functions, we may attribute this to the imperfection of our nature or the fact that evolution has taken an imperfect shortcut. In either case, it is certain that placing a significant weight on self-interest is highly beneficial to both the individual and the society, yet just a bit too much can have hugely negative effects. For me, what the commonness of egotism implies is nothing but the presence of something other than morality in people’s own cost functions – whereas morality has its benefits and imperfections, egotism simply has less of the former and more of the latter, and is grouped outside of morality for this reason (while a modicum of self-interest being on the moral side). Clearly, there is some sort of spectrum in judging the moral value of traits. Drawing all the previously mentioned things together, I feel I can now justify my definition of morality as an “approximation” or “shortcut” to optimal behaviour for the species as a whole.

It is without doubt important to stress that morals have their own imperfections and limitations, like analytical reasoning, and depend on the individuals (or society thereof) in which they have formed. Yet depending on how you look at it, morals have  evolved or been designed specifically for the purpose of optimal society. Although morality may be less adaptable than intelligence (at least over the timespans ranging from days to maybe lifetimes), it assuredly has a more “tailored” purpose, and therefore has its place alongside, and arguably ahead of, analytical reasoning.

If I were to now summarise how I believe optimal behaviour should be guided, I would say that it’s necessary to be somewhat careful not to propose something too uncompromisable. In reality, it’s almost always the case that reasons are more intricate and subtle than immediately apparent. In stating an emphasis on paying due attention to intrinsic morals (loosely, which can be recognised as principles and codes that typically “feel right” and are “seen to be right” by consensus of society), and contrarily wariness in ignoring these morals in favour of some sort of pure reasoning. “Reasoning”, after all, when performed by humans, cannot help but be intruded by egotistical motives, among other notable imperfections. Do we not after all have a fear of so-called purely “rational” or “logical” machines not hesitating to perform tasks that are undeniably immoral in the eyes of man (if not only founded in science fiction and our imaginations)?

As a quick final note, I ought to mention that nowhere in my musings have I required morals to be static in nature. Equally, there would not seem to be any issue with them being unchangeable. At this point I’m further tempted to divide morals into two categories: intrinsic and social. Again, this is a matter on which I’m only going to lightly touch. The latter is the one of particular interest in that it could imply a varying cost function for optimality. It also suggests a mutual feedback cycle between the will of society and contemporary moral values (with analytical reasoning somewhere in the process, potentially acting both positively or negatively).

Right, so this post has wound on long enough by now, and is only getting increasingly vague and leaving more loose ends. Still, I hope that I’ve at least partially conveyed my theories and general impressions on the subject. I’m not sure how everything appears to others who haven’t followed the continuous discussion on the wider topic (largely originating with David’s post). I’d certainly be keen to hear what anyone else thinks on the subject and the ideas presented here. I would not at all be surprised to receive opinions that this relation of morals to optimality seems distasteful or even incomplete to many people. Indeed, I am not sure that I am wholly satisfied with this explanation as the “root” of morals. (How can I, having already cynically accepted the fallibility of human rationalisation?) Maybe it is as a student of physics that I realise our theories of the nature, physical or human, are always but approximations to a more profound reality.

RCSU Science Challenge 2009

Posted in Maths & Science, Personal on March 3rd, 2009 by Noldorin – Be the first to comment

The Royal College of Science Challenge is an annual competition open to all students of Imperial College London, and something I’ve decided to enter this year. At the present time, the midnight deadline has just passed and I’ve just taken a huge sigh of relief… As is my habit with most things (writings or otherwise), I ended up not even starting work on the submission until the past Friday, which has indeed turned out to be a very unwise decision. Anyway, the good news for me is that I did finally manage to get the essay completed and uploaded at a panicky 11:53 pm. (At this point I half-expected MS Word/Firefox/Vista to crash spectacularly, but my fortune in fact held out for a crucial 15 minutes!)

Now to the contest itself. I should note that everyone, irrespective of year and department, was able to pick any one of the four proposed questions and then proceed to write a 800 word essay on the subject. A disheartening thought if considered for too, but something told me it was worth a shot anyway. My choice was the following (third) question:

Will Homo Sapiens continue to evolve? If so, how?

For those who don’t know, I am currently studying for a Physics here, and haven’t taken biology for several years now – so it may initially seem rather silly for me to ignore the others and tackle the least physics-based topic of them all. (In actual fact, it figured that I had far too much to say and the awfully low word limit was the source of much exasperation for me!) An interesting short article in the January issue of this year’s Scientific American had gotten my pondering the issue lightly before I even new of the contest, which is probably half the reason. Although I’m sure I could have done a decent job of one or two of the others (certainly the one on the LHC, I would think), I guess the open-ended scope of the evolution one caught my fancy at the moment. Enough said – I do at least feel pretty satisfied with the end result. In getting there, I must also mention the kind assistance of my friend David in repeatedly proof-reading it and making some quite insightful suggestions.

Without further preamble, here is my entry.

Each year, the infamous Darwin awards are given out in order to “salute the improvement of the human genome by honoring those who accidentally remove themselves from it (1). This “award” is of course proclaiming to recognise those people who supposedly perform supreme acts of stupidity and thus assist the process of natural selection in the human race, first described by Charles Darwin in the 19th century. Behind this purely ironic prize is however a much deeper question: have we, as members of humanity, changed significantly over the recent past, and will we perhaps evolve into something quite different in the near or distant future?

It is commonly believed that our bodies have remained static from at least the time of the birth of civilization around 7,000 years ago. The greatest period of “recent” alteration occurred as the species branched off from those of other primates hundreds of thousands to millions of years ago. A recent study by American two universities[1], however, threatens to entirely reshape this perception of our evolution. It contends that a minimum of 7% of the human genome has changed over the past 5,000 years or so. These scientists even go as far as asserting that “humans have evolved as much as 100 times faster than any other time” (2).

One thing is agreed upon pretty much for sure: human evolution, however significant, will follow a radically path in our future than it has for most of our past. Firstly, it is generally believed that our brain size is not likely to increase anything like it has done in our early history. This is not to say that human behaviour will not be transformed – rather, it is considered one of the most probable of man’s characteristics to do so, thanks to the rapidly changing nature of society and growth of technology. Conceivably the most extraordinary course for our evolution to take is some form of symbiosis with machines. Even presently, human dependence on machines is immense. Economies, societies, and even many individuals require them merely to survive. Perhaps the merging of mind and machine is but the next step in a natural progression? Symbiosis, if chosen by most people, would likely add selection pressure towards purely functional behaviour. The thought of abandoning many of the pleasures of human life may be enough to prevent us from following this route, though it does not remove the prospect that one day our species might seriously consider it. The ultimate human stimulus for evolution may in fact be space travel: such a scattering of the species would be liable to produce great diversification and force adaptation.

Perhaps the most astounding change in the manner of our evolution will come as a result of human intelligence. We are now beginning to understand properly the process of evolution on both macroscopic and microscopic levels. The very fact that we are aware of such a process working upon ourselves is exceedingly likely to alter the “natural” course of evolution, maybe even unintentionally. Our efforts to perform artificial selection have already begun and are currently highly controversial issues (often in such guises of genetic engineering or “designer babies”). These ideas and associated fears are certainly not new; science fiction has long speculated on possible paths that we might take in our search for a “better world”. Aldous Huxley’s classic novel Brave New World, written in 1931, describes a future dystopian society where people are engineered to fulfil certain roles within society, reproduction being an entirely artificial, state-controlled operation. Somewhat differently, Frank Herbert’s Dune series explores a universe in which a powerful organization of women operates a long-term secretive breeding program that aims to produce a man with supreme mental capabilities. The common theme in these works is an important one; namely, that there must exist some relatively small group of individuals who possess the authority to direct the course of evolution and the workings of society. Indeed, we all know to where eugenics can lead. Now let us ask ourselves whether, as inherently fallible human beings, we can be trusted with this sort of power?

The future path of human evolution is an uncertain thing, to say the least. If modern research hints anything at what is to come, we should undoubtedly expect huge change. That we may soon have the opportunity, at least in part, to direct our own evolution, will make it especially unpredictable. Yet if we can make sensible decisions and learn to accept what it outside of our control, we may find ourselves transformed into something rather new and astonishing. After all, why should our species, one so devoted to expansion and improvement, suddenly remain static?

Bibliography

1. Darwin Awards. The Darwin Awards. [Online] [Cited: 28 2 2009.] http://www.darwinawards.com/darwin/.

2. What will become of Homo Sapiens? Ward, Peter. January 2009, Scientific American, Vol. 300, pp. 68-73.

3. Huxley, Aldous. Brave New World. London : Chatto and Windus, 1932. ISBN 0-06-080983-3.

4. Herbert, Frank. Dune. s.l. : Chilton Books, 1965.


[1] Teams headed by Henry C. Harpending at the University of Utah and John Hawks at the University of Wisconsin-Madison.

I now only await the news of the shortlist, with some little hope that few enough students of science would bothered to voluntarily enter an essay competition. (Yes, this is a purely unscientific hypothesis.) Regardless, it will soon be seen whether such expectation is in vain.

Evolutionary Algorithms

Posted in Maths & Science, Programming, Projects, Software on January 31st, 2009 by Noldorin – Be the first to comment

Genetic algorithms (or more generally evolutionary algorithms) is an aspect of programming that has interested me for quite a while now. The concept of using natural selection and simulating (in an abstract sense) the process of evolution of biological species with computational algorithms may not seem to useful upon first thought, but has in fact created a whole field of research in recent years. It turns out that genetic algorithms (GAs for short) are extremely useful and relatively efficient to throw at a problem about which you typically know quite little. (However. they are not terribly good at finding perfect solutions, which is why they are often used along with another late-stage optimisation algorithm.) They can be summarised as being essentially optimisation techniques that work in virtually any search space (though with varying degrees of success). Just to list a few examples of problems at which GAs tend to do well:

  • Travelling Salesman Problem
  • Model fitting and prediction (This is used with some degree of success to forecast stock markets.)
  • Evolving artificial neural networks (These two nature-inspired AI algorithms work together quite well indeed.)
  • Parameter/weight optimisation (in any system where there are large number of free parameters and complex inter-relationships)

I will point out the last one in particular, as it could potentially be used rather effectively with a game AI such as the Stratego one I am currently writing – more to come in a future post.

Unsurprisingly, many online articles have been written about evolutionary programming, ranging from basic introductions to scientific papers. If you’re curious about the topic and fancy learning a few things about it, I can recommend these articles, all written in plain understandable language:

  1. Genetic Algorithms Overview by Michael Skinner on Genetic Algorithms Warehouse/AI Depot
  2. Genetic Algorithms by Marek Obitko
  3. Genetic Algorithms in Plain English by Mat Buckland on AI-Junkie

Finally, to the main purpose of this post: I have recently finished coding the beta version of my Darwin.NET project and released it on Launchpad. It is a library for generic evolutionary algorithms, with direct support for genetic algorithms and also an extension for gene expression programming (GEP). The ideas presented for GEP are what initially inspired me to create this library. A comparatively recent idea (traditional GAs were first designed in the 1950s), it was originally proposed in a 2001 paper that can be found here, and is well worth the read. Despite being published for a scientific journal, it is surprisingly straightforward to comprehend and should offer anyone a good understanding of why GEP is so special (and a huge improvement over traditional GAs). The end part clearly details how it can be used to solve several complex problems – according to the author’s statistics, significantly (orders of magnitude) quicker than GAs.

Now, the library that I have just released provides reasonably complete implementations of both GAs and GEP, though I must point out that it has not been extensively tested. (There are currently only two samples included with the source code, though they ought to at least help you get started. Before I attempt to write crazy extensions like a GEP-based algorithm to evolve neural network structure, my priority is to write a few more samples as I gradually improve upon the library. Oh, and I’ll begin to write up some proper documentation too.) I would also be very glad to hear feedback of any sort about the library (here or on Launchpad), or even a simple note that you are using it for a project! Any overlooked bugs are the first things I would like to get resolved of course, but design and feature suggestions are equally welcome.