The Human mind vs. the Intelligent Machine
Last year, as many of you in the field of AI will know, an intelligent machine (i.e. a version of IBM Watson) successfully defeated a GO World Champion over a 5 game series.
This solidified, in the world of AI, if there was any doubt that the AI based on self-learning pattern recognition methodology represents the best course currently openly available to us to deliver intelligent machines.
I don't have a debate with this conclusion, but I do wonder if the "breakthrough" that was represented by the triumph is indeed as "big a deal" as one might on first sight think.
My point here is that for PRS to actually work really well, a declared outcome has to be understood and a way of achieving that outcome already in existence, preferably with both lots of data and lots of experience behind it.
On the experience side, my metaphor is the way that children evolve their ability to speak a particular language or in many cases, many languages up until the age of 7 or 8. It appears that we might be hardwired (as a survival instinct) with a need to communicate and may actually recognize that learning a language is the key to this. Furthermore, good communication (i.e. adopting naturally to rules that allow understanding) as opposed to bad communication achieves the outcome much more readily and thus, via a natural form of PRS, is achieved by a child much more quickly.
As far as I am aware, in relation to the GO challenge, the intelligent machine was shown what winning "looked like" - the outcome, and then was introduced to patterns that led up to that winning being achieved. This was done on a massive scale, but not at an exhaustive one, even though maybe the machine hardware/software could have done this given enough time.
Now, that's think about how this becomes a winning approach against the GO champion and why perhaps it highlights where intelligent machines will really make a true difference and where they may not.
The GO Champion, at least as far as I am concerned, is not an evolutionary multi-generational entity. What I mean here is that the Human Champ does not represent the entire accumulation of the history of winning (ie achieving the outcome) at GO. The person may in fact encapsulate through study, experience, and interaction a huge amount of knowledge, but not its entirety.
This is important to understand because my view on the human mind has always been one where half the battle is to avoid unnecessary process wastage (noise inputs with no correlated value to outcome), and the other to avoid the pitfalls of adapting pre-recorded (but unrecognized) biases that take us down using a System 1 instinct approach way too early in the decision making process.
Now if we imagine that both unnecessary wastage, as well as instinctual bias can be removed from the intelligent machines programming code, leaving only PRS techniques to focus on the outcome (i.e. winning) then in theory, with enough data, and enough experienced reiteration of probabilities to decrease the error margin to win, we will create an intelligent machine capable of beating a human champion. This is clearly easier said than done (by a long shot), but if one further assumes that human endeavor to success is 90% recombinant and 10% breakthrough (i.e. an entirely new system to win at GO) than in actually the reiterative process may not be as mind boggling as the sheer number of possible moves on the board might represent.
The point that arises from this is that as we digitize more and more of the world's knowledge both in the "rear view" mirror, as well as going forward, and as we align this knowledge toward achievement of outcomes (hopefully good ones), we enable more and more intelligence, and advancement on this, to become part of the intelligent machine.
For many situations this will mean that the challenges may be less around the agents at work and more about the data, data management, and contextual elements of the data structure itself that facilitates learning. Clearly this was addressed in the battle to teach GO to an intelligent machine..