[Day 135] A penny of my thoughts: Artificial inteligence

  So, I've decided to try out a new style of content - "compact essays" (I will try to keep them in 1k - 1.5k words range) in which I'll succinctly express my opinion on one topic of my choice. I'll usually go with something along the lines of psychology, futurism or occasional conspiracies (stick to your guns they said xd). I don't really want to poke into fields I have very limited knowledge of because the article itself would be likely factually incorrect, draw an incorrect conclusion or commit various fallacies due to my inability to grasp the complexity of the topic I would be covering. In this pilot, I've chosen a topic I am a big fan of - AI -  to make it easier to successfully manoeuvre in this newly developed format. I reckon that the quality of this type of posts will slowly improve over-time as I gain a better understanding of how to lay out a convincing formal argument in an understandable form. Most of the info I'll be using as the source material is easily accessible in digital (or sometimes analogue form even though nowadays, you can buy almost any book through the internet) form. Anyways, without further ado, I give you: my thoughts on AI

Disclaimer: My "thoughts" on the topic I'll be covering are simply accretion of opinions of people that are much proficient and eloquent than myself, however, anyone that has any sense at all knows that gathering info this way is much more beneficial (in terms of resources spent) and easier than drawing conclusions from the first principles each time you want to think something through (needless to say that I would likely arrive at flawed conclusion given the disparity in overall wisdom between me and some of the greatest minds on the Earth).

  So, what is artificial intelligence (AI)? The simplest definition I could find goes something like this: AI is a computer system able to perform tasks normally requiring human intelligence. In spite of the fact that there are many fields which interests overlap with this phenomenon (and each of these fields define the term itself slightly differently), the "basic" version of its definition is pretty simple. The more interesting question is: Why do we want AI?/What do we gain from building it?

  I would presume that the most logical answer is this: we want to facilitate and speed up our progress as a species as much as possible - in my mind, there is no doubt about the fact that any rationally thinking member of our species would likely affirm to this assumption. Of course, there are other auxiliary factors that may cause people to be interested in AI such as its cool, interesting or provides significant monetary gain for ones that get to manufacture, distribute and operate it.
But how might AI help us to facilitate our progress as a species?

  Standard computational devices have sufficient computational "brute" power, but they have very limited operational flexibility. On the contrary, humans have the ultimate apparatus to flexibly approach any given task but they lack sufficient computational power to reliably arrive at the best possible solution, especially when dealing with complex problems (we are prone to biases, and the more complex the process is, the more probable is that our thinking pattern will deviate on the "fallacious path" leading us astray from the correct conclusion). Knowing our weaknesses and weaknesses of our current technology, we would like to create some advanced form of AI would be able to combine the computational power of a computer with the cognitive flexibility of the human brain which would result in us being able to solve problems are now simply beyond our reach.

  On the surface, the idea sounds nice but it's not without its shortcomings - and in this case, their shortcomings are so massive that when left unaddressed, they could potentially eradicate our entire species as it is but why is that so? Off top of my head, I can think of two major problems we will encounter when trying to design some advanced form of AI. Both of them are connected through the notion that AI we want to create should ultimately be useful to us. I think we can all agree that spending money at something that is not ultimately useful to us is not a particularly smart resource management strategy - this principle is obvious and the development of AI is by no means excluded from it.

  The issues we will face when trying to construct AI that is ultimately useful for us:

a) it needs to be autonomous - if AI we construct and it will not be autonomous it will require an obscene amount of maintenance to keep it running which would probably render it barely usable - plus it will not be able to problem-solve as efficiently which is directly tied to the next point:

b) it needs to be "smarter" than anything we have up to this point - the main reason why governments, as well as companies, want to produce an autonomous AI with superior problem-solving skill is to - obviously - solve tons of problems we have currently no answer to and we pay massive consequences because of that (ways to revert global warming, pull the 3rd world countries out of poverty, effectively cure cancer etc.).

  When you think about the fact that we want to construct something with these properties, you inevitably come to the conclusion that, with our current understanding, we put on our shoulders more than we can currently manage. The fatality of trying to construct superintelligent AI might not be evident at the first glance but consider this: given that this AI will be much smarter than us, we will have a hard time catching up to it to understand its findings - the perfect analogy to this would be being in a class with this smart kid who understands something let's say 2x faster than you - you would have a pretty hard time keeping up with him right? The law of compound interest says that the disparity of knowledge between you and the smart kid would grow larger every day and after a while, you wouldn't be able to catch up to him even if you would work 5x as hard as him for the rest of your school year. That does not sound very pleasant. Now consider this process amplified 1 trillion times (I'll be considerate, so an average human does not look that stupid in comparison to advanced AI).

  If we assume that we are not the most intelligent species/beings that will ever exist - assuming that we are would be pretty arrogant because there is no evidence of that (apart from 6000 years old collection of gruesome stories) and conversely, there is an overwhelming evidence that we are nowhere near the peak of absolute possible intelligence/wisdom attainable (yes, our progress sped up quite a bit last 150 years, but we might as well still be at the bottom of  a "possible intelligence curve"). In short: if humanity does not eradicate oneself due to WW III breaking out, or some colossal natural disaster happening, it's basically assured that our computational power will advance to the point at which we are able to construct an autonomous agent (AI) that is as flexible in terms of cognition as human but processes information 1 trillion times faster. What would be such an agent would be capable of? 

  If you were given an assignment you are required to work on which would take approximately 31 866 years to complete (assuming that you work 24/7/365), the AI would be able to complete it in mere one second. That kind of computational power with the cognitive flexibility of the human brain is terrifying. One could say that is doesn't matter if we can't catch up to it if it provides us with some useful information but that is simply not true. We cannot possibly imagine what would happen if we turned on AI with such computational and problem-solving power for even one second (it's unimaginable how far ahead would you be if you could think about data presented to you for ~32 000 years). After realising this crazy disparity between your intelligence and possible intelligence of AI, it goes without saying that even if we did our best to word a problem on which  AI is supposed to work with utter precision, and with the best possible intention, we would still have simply no clue what would happen if we turned it on to do its thing (see paperclip-maximizer thought experiment).

  Many of the greatest minds have already realised the severity of the situation (standards solution such as having AI to explain everything to us so we can catch up to it or simply letting it operate in a closed environment to ensure that its safe would hardly be an obstacle for a superintelligent AI if it decided a course of action that would prove fatal to the human race) and are publicly calling for something like a Manhattan project that would bring together the greatest authorities on the topic to figure out the best way out of this rabbit hole. One simple solution would be to stop a construction of an autonomous agent completely, but that would dwindle out potential progress hence it should only be our last resort.

  Another problem we are currently facing that hardly anyone is aware that AI as an issue that needs to be addressed. It's not as apparent as some other issues (famine, illnesses, global warming) and to the majority of people, it does not seem like that big of a deal because it's too abstract and rather counter-intuitive (this very issue is addressed in greater detail in brilliant Sam Harris's TED-talk). As of now, the best thing we can do in regards to the problem of AI is to get more young, ambitious people interested in the topic and hope for the best.

  That would be it for today, hope you are having a beautiful day.

  Cheers.