Thursday, 16 February 2012

How about a fair fight? Human on Human

Way, way back in Too Old I criticised Jaron Lanier’s concept of hip hop. I stand by that critique. However, as time moves forward and social media continues to be integrated and moreover, normalised in contemporary technolgised societies, much of what Lanier had to say in relation to mob mentality, crowd wisdom and anonymity holds true. What I want to return to specifically in Lanier’s work is the idea of human elegance in programming.

Last month Bonkras, a computer program with some serious processing muscle behind it again defeated legendary shogi player, Kunio Yonenaga in a hyped event not unlike that of IBM’s Big Blue and Kasparov. Though it must be said that Yonenaga was a far more graceful loser. As usual a flood of analytical reporting sluiced through prime time current affairs and variety shows, all under the heading “Man versus machine” (etc). It only took the first such headline to send my mind at lightspeed (or thereabouts) to Lanier’s You Are Not A Gadget. Lanier rightly points out that in spite all of the processing power and depth capable in contemporary software, the all too frequent trope of the machine beating the human is not only flawed but essentially a logical fallacy. After all, it was humans in the first place who created the platform for the software to run on in addition to the software itself. Frequently such misanthropic software takes months if not years in actual man hours and numerous testers, debuggers, graphic designers, engineers and mathematicians to produce. Indeed, if we were to properly frame Yonenaga’s recent rematch defeat it would be more sensible and far less human hating to render it:

“Team of humans building on the work of hundreds of mathematicians, programmers and engineers before them find a way of (a) externalising a number of calculations relating to a highly specific, highly controlled and limited environment via post-mechanical, integrated circuit technology after many months of research, trial and error and then (b) pit this technology against a human who, like the majority of the species is incapable of making calculations at the speed and density of machine mentioned at (a) and then waiting for the (inevitable) outcome.”

Although accurate, a far less attention grabbing headline. What is interesting to me is that we rarely think twice about the mechanical advantages most machines have over humans in most situations (from bulldozers to power saws). However, when it comes to the mind there seems to be a fascination around the binary culture/nature, machine/man, as though they are not in fact one and the same. May I stand accused of sour-graping Bonkras’ victory, but is it not bleedingly obvious that had someone not turned on a power switch and launched the software application, Yonenaga’s victory would have been assured. At least, at this level, the necessity for a human hand can easily be seen.

I believe what would be more interesting and more deserving of praise would be rival Bonkras type programs which would allow a level playing field, where multiple teams put their software, calculating and algorithm creating prowess to the test by facing off against each other. It is neither sporting nor fair to place an external calculating device designed to a maintained by a team of professionals against a single human opponent. After all, let us take a look at some of the mathematics involved. A computer is capable of number crunching at a level far outstripping human capacity and on raw numbers alone, it is thus capable of taking a vast amount of database information probability and correlating it against mathematical possibility within the environmental confines of the game itself.

In other words, shogi-master-crushing-software can draw on an entire recorded history of actual moves at various stages of a single game, it can also reference paths to victory and defeat and then analyse them in terms of frequency. Furthermore, path trends based on frequency of use correlated to actual number of victories/defeats over time can be analysed in order to develop a probability tree which can be further divided according to per turn or event based timing. All of this raw data and analytical data of actual patterns over time forms a probability matrix. This probability matrix can then be indexed against a possibility matrix which is made up of the actual number of possible moves given the environmental (a nine by nine board) and rule based (especially the ability to return pieces to the board) conditions/limitations. The scale of the probability matrix is truly awesome. As I am not a mathematician, my analogies will be crude and probably to some degree incorrect. Astute readers, please correct!

Let us say that post-Pacific War (1945 onward) there are shogi tournaments in four major urban centres (Tokyo, Nagoya, Osaka, Fukuoka) with eighty participants drawn from the cities and surrounds capable of playing at a semi-pro and professional level. That is a total of 320 players in a national “league”. Then let us presume that there are a total of four elimination tree tournaments per year, plus a final fifth to determine overall champions in each league. At this point alone, in a single urban centre we have a total of: eighty games in a single tournament which includes progress to the end and the determination of a third place winner. Multiply this by four (number of tournaments in a single year) for a total of 320 games played. Let us say that the top twenty from each urban centre are chosen to play in a national tournament, that is another eighty games played for a total of 400 games in a year. For the sake of simplicity, I am not including special events, champion versus champion exhibitions matches and so on. But that 400 is not the end, we have skipped a step, first let us go back to the local leagues. 320 games/league per year is a total of 1280 in addition to the 80 “national” championship rounds for a grand total of 1360 games per year of recorded professional level shogi. Multiply that number again by 67 (difference in years between present and beginning of records) for a total of 91,120 games to provide the base raw data for a program to analyse.

Now let us put this number in context. If a game of shogi takes thirty minutes to play to completion (and this is a conservative, lower limit estimate) then the total time spent to complete 91,120 games is 2,733,600 minutes, or 45,560 hours or again 1,898 days which is equivalent to just over five years of non-stop, shogi with no breaks for eating, drinking, sleeping or living. Simply put, it is unlikely any human in existence is capable of calculating at a speed required in order to effectively beat a software based opponent capable of utilising this data in real time on a consistent basis. Rise of the robots indeed.

Do not fear reader, as I am no technophobe. My ever-suffering partner would attest otherwise simply by starting to count the number of gadgets in my possession. As I have written again elsewhere, a software program is as elegant piece of engineering or art as anything else, more physically based that preceded it. I have written about my return to heavy metal but have only touched ever so briefly on my return to gaming. And that return can be attribute to Lanier’s argument presented above. When games are not conceptualised as “mindless” diversions, “mere” entertainment or painfully legitimised as “art” and instead seen as a conceptual interface between a user and programmer/programming team, it becomes possible to enjoy gaming on a meta level while still enjoying the thrill of shaving precious 100ths of seconds of Mario Kart 7 time attack times. When gaming on the Zen level it becomes possible to locate engineering pragmatics, hardware and software based coding limitations and personal idiosyncrasies of the creators.

Zen level gaming is a term I use to refer to the practice or event of being simultaneously engaged with the problem posed by the software at hand and then attempting to understand how that problem sits in the wider game as a mechanical consideration (in other words an attempt to contextualise the game in time and space). In days past, this was easier than it was now as hardware limitations imposed discipline on colour palettes, sound frequencies and calculations. With today’s processors, truly breathtaking virtual worlds, down to the smallest level of detail are created and finding the “logic” an mechanisms of the games can be somewhat more difficult. Though not always. This is simply another articulation of the Bonkras/Big Blue paradigm. Games offer far richer ways to humans of interacting with complex environments with restrictions reduced for the purpose of creating a sense of freedom. Designers are working ever-harder, in tandem with hardware developers to cover their tracks, to paper over conceptual/perceptual cracks and create mind bending illusions for preventing discovery of deus ex machina. But new generations of children (and curious adults for that matter), weaned on and equally familiar with evolving digital technological frameworks continuously find new ways of simultaneous engagement and reflexive distance. In other words any distances achieved between artificially assigned separations between human and machine are constantly closed through evolving use patterns.

To me then this whole “man vs machine” debate is deeply problematic. It relies on a populist and incorrect separation of human and technology which is at best naive and at worst irredeemably cynical. Humans have learned to see and create the world through calculation a concept they created, just like memory or language, which in a full feedback loop comes to continuously (re)create them. Technological innovation stemming from calculation has allowed us to exceed our limitations in virtually every natural, limiting element. Thus there is no such thing as “man vs machine” since “machine” can only ever be, even when (apparently) fully separate from human biology by way of artificial intelligence, of human. Let us set the correct terms for a fair fight, then?
Post a Comment