A Response to the Imitation Game – Types of Learning
A few things have interested me as of late, one is an idea presented by Noam Chomsky regarding language and the other is Alan Turing’s reference to learning machines in his paper “Computing Machinery and Intelligence”. In the ancient Chinese text, the Chuang Tzu, it says that words are for communicating meaning, however once one has gotten the meaning there is no further use for words. It goes so far as to say that one longs for someone to cease the words with and to sit in silence perfectly understood. Chomsky, a father of linguistics as we know it today, said something in an interview which really caught my attention. He spoke of a similar principle of language. The enormous diversity of language and multitude of languages around the world comes along with the fact that language development is an intrinsic part of every human being. This means that before a language is created, there is some thought which precedes the language. How else do you form words to express a thought? The thought exists before the words, the words are simply the chosen mode of communication. This remains mysteriously intangible and brings me to Turing. The idea of learning and what it means to learn has a similar mystery to it. How is it that those abstract thoughts in our heads (I call it thoughts, but since we can’t place our finger on what quite they are, we may even call them bologna), turn to complex sentences and languages? We somehow learn to express ourselves, develop a system, yet do not remain confined to the system. The system of communication evolves everyday, hence, language evolution. So what is the nature of this learning? What are different ways in which we may learn? I attempt to outline a few of my ideas below. I am aware that there is extensive research on these subjects and I am likely heavily under-informed, however I believe it valuable to experiment with ones own ideas as much as research others — that in of itself is a seminal aspect of learning, not to mention a fundamental tenant of the scientific process. Hence, I do not find the writing of this any less valuable than a thoroughly researched expert paper.
Step-Wise Learning
This is probably one of the most systematic forms of learning in which there lies a clear schedule and set stages of learning. Imagine the belt and form system of Kung Fu or Tae Kwon Do in which each stage requires mastery of certain forms or techniques. Done incorrectly and you may get a wooden stick to the hand. Conversely, done right and you may pass to the oh so shiny green belt. As described a key component to this is a system of rewards and punishments.
Recursive Learning
Why? What? How? When? Questions are fundamental to this type of learning. You use what you know to learn what you do not and for this you must have a keen ability to question. A shark can swim in water, but what if we put it on a beach? How would the shark move then? Could I then eat that tasty fried fish I wanted to bring into the sea water and not be afraid of the shark taking a bite at me? This type of learning is perhaps what may bring us closest to developing what we have learnt through primary, secondary, and higher education to be the scientific process.
This type of learning is particularly interesting and mysterious to me. It brings me closest to Ramanujan’s idea of “infinity in the palm of your hand” or the idea of a single atom creating the universe, for that single atom has in it the possibility of the infinite. Or even the idea of the void in which there is nothing, absolute nothingness, yet somehow in that nothingness we have ideas of everything (a black hole). While perhaps getting more theological, the idea of a single point of creation is one which captures our minds to this day even in more scientific senses. The idea of a master algorithm, a single line of code which can give rise to every possible line of code. The search for a theory of unification — a single equation to describe the entirety of the universe. Aum, to return to the more theological. Perhaps put more in Zen terms “reality is a rotten tomato”. The importance of that statement being that it is not important what reality is described as, rather simply that reality is. It is not important how we approach it: aum, unification, an algorithm, what is important is that the concept is valid. Because of the diversity of this idea cropping up in all aspects of our thoughts of the universe, it cannot be discounted. Therefore I believe that this concept of everything from nothing is one which is a valid application to recursive learning.
One Foot Forward — One Foot Back
In my mind, I find this to be a particularly difficult type of learning to program. There are only so many steps you can take forward on the same path. In rock climbing for example, you must often take a step down, around, and then back up in order to get past a block in your ascent. It will do no good to simply continue climbing up, unless your goal in rock climbing is to fall off the rock. This requires the setting of smaller goals to achieve larger goals. The meeting of these smaller goals collectively lead to the accomplishment of the higher goal, however it must be remembered to take a step back in order to take two steps forward. The challenge I believe lies more in knowing how many steps back to take and which possibilities to pursue from that point. In a particularly complex maze, you can hit a dead end and then choose to back up to the last decision you made to make you hit that dead end. Or you can choose to go back one step further and reconsider the decision which led you to that decision. And so on. It may simply be a matter of intuition in order to determine which step to stop at and reconsider. These intangibles or as Alan Turing mentions, extrasensory perceptions, are ones which are the most baffling to program into machines today. Perhaps my reader can inform me as to the state of this type of programming today as I know it is a broadly researched problem at this point — it no longer being 1950, the time of Turing’s paper.
Pattern Recognition
I connect most closely to this type of learning. As a tabla player, I am constantly exposed to patterns. Tabla is a North Indian percussion instrument in which the percussion is passed down orally. There is therefore, a vocal percussion language associated with the drumming. In learning orally I learn to pick up on patterns in phrasing, much as a baby you pick up various patterns in words and sentences. These patterns combine further and further to make more and more complex compositions of which I can, after sufficient experience, learn within one to two listens of a “new” composition. I say “new” because in a way, the composition may only be considered new in the sense that it is an amalgamation of previous patterns or a development off of them. In other words it may only be a new composition in the sense that it is a growth of a seed planted from earlier teachings (going back to the idea of everything from nothing presented in recursive learning). The idea of this is simply that pattern recognition allows for faster learning and a level of what we may call mastery. Malcolm Gladwell mentions in his book Blink that what we think of as mastery is none other than someone who can see patterns faster than a less experienced person. A chess master is so because in one glance, the master is able to see more patterns than a less seasoned amateur player.
Emotional Learning
This is topic has a wide berth. Although seemingly only applicable to humans in this case and not universally to machines and humans alike (or to put it more generally, not applicable to intelligence itself),Turing argues in his seminal paper “Computing Machinery and Intelligence” that the question of whether machines have emotion or feeling is one basically of consciousness. And it is simply not pertinent to question if another person can think or you are the only one that can think. It is far more productive to simply assume that A and B can both think and therefore continue to communicate based on the question of the imitation game rather than that of “can machines think” or in this case feel. If one does not truly know whether a machine can feel unless one is the machine, then it is as Turing puts it, “polite convention” to assume that both human and machine feel and move on with investigating the question at hand. For this reason I do not believe that emotional learning pertains only to human intelligence.
Now to proceed with describing learning through emotional learning — the most ready at hand example comes through music. One does not usually find musicians bereft of emotion as successful in their field. As Bruce Lee puts it, “don’t think, feel”. It is hard to suppose Beethoven or Bach became proficient composers from thinking mechanically of a C major followed by an E major in triplets written in fortissimo of which the conductor serves no role but to make sure the orchestra stays on task and slap the musicians if they do not. Rather these composers likely see the colors, as the famous Quincy Jones quite literally does, in their music and it is those feelings which drive their creations. How does this chord make me feel? What if I speed this section up? Wow, I feel like a drugged up sloth! Emotional reactions like these drive learning and creation and they are also key drivers in what may be more widely known as experiential learning.
Taking a Stab in the Dark
Randomness seems to be a key element for us to both learn about the world and learn in the world. If we could completely predict every next occurrence, there would be nothing left to learn. Learning as we know it would simply be the certain knowledge of what comes next. At this point all is known and learning ceases to be a valuable construction in the world. There is no conceivable method that I can imagine, even with the ever increasing amounts of storage capacity and interconnected computation we see in today’s world, which would allow for enough storage capacity to contain every bit of information present in a single moment of the universe. This problem is present in economics via Kenneth French’s Efficient Market Hypothesis, stating that prices affect all available information. Hence, no financier can beat the market due to requiring knowledge of everything about the economic system to accurately predict what will happen next. For these reasons, randomness cannot be avoided in learning, and therefore is essential to any intelligent learning system
A last note to mention is that along with these types of learning, there are tools of learning which enable learning to take place. One can draw the parallel between Turing’s description of the three most important parts of a computer system and these tools. The three main components of a computing system are store, an executive unit, and control. All of these involve various parts of the human psychology — the prefrontal cortex, cerebrum, memory. However these tools lie separate from the methods of learning presented above and require separate discussion.
You may have been able to tell a very heavy yin-yang effect present in my description of each of these types of learning. There is no mutually exclusive type of learning, all are interrelated and each method must contain an element of the other. We as intelligent beings are not a single one of these systems, but instead are a combination of all the above mentioned and I am sure more which I have yet to uncover. Yet we have many more years ahead and as Bruce Lee famously stated, “All knowledge, ultimately means self-knowledge”. Truly the mystery within is one of the most profound mysteries among the history of our species and to be a self-conscious drop in that ocean is the beginning of a wave.
Citations
“The Stupidest Thing You Can Do With Your Money.” Interview. http:/freakonomics.com/ (audio blog), July 26, 2017. Accessed September 22, 2017. http://freakonomics.com/podcaststupidest-money/.
Turing, A. M. “Computing Machinery and Intelligence.” Mind 59, no. 236 (1950): 433–60. http:/www.jstor.org/stable/2251299.
