A Hitchhiker’s Guide to AI – Decisions, Actions, Outcomes
Of all the buzzwords we have had thrown our way over the past few years, few have had as strong a staying power as “AI” (Artificial Intelligence). We have barely digested social media and ASMR, and now we are faced with another buzzword which promises to “change everything”. Can it not all get a bit tiresome? I want to break down AI for the slow mover; the regular non-geek; enough that the prospect of being led into a world “led by” AI and the thought of participating in it as a productive member appears less daunting or tiresome.
The world we have created around us for us, by which I mean the basic aspects of civilization - the buildings, houses, parks, roads, offices, sanitary systems (very important!), even family and friends - are the results of discrete decisions we all make in our individual capacity.
In making these decisions, we are affected by a variety of objects (data points) forming part of our environment which we may or may not choose to take cognizance of - the natural environment, the people we hang around with, the pervading ideas, values and emotions of the day etc. Decisions leads to actions or non-actions; and actions or non-actions, either discretely or accumulated over the course of time lead to outcomes, one of which happens to be the state of civilization we enjoy today. These outcomes affect us individually; and they may have an impact on others too.
The outcomes of one such cycle may feed into one of the stages of another cycle too; so that the outcome of one cycle may be a decision or action of another cycle. String enough of these cycles together, and you have a value chain. String enough of these value chains together and you have an economy, which over time gets called a civilization.
It works in a mathematical fashion. I do not think there is anything particularly profound in saying this: something as mundane as the act of writing this article followed from a decision to do so; and I was encouraged to do so by (awesome!) people who I happened to be connected to quite indirectly.
Often times, there exists a feedback loop between outcomes and the decisions which affected the action/non-action which led to those outcomes. This can be called learning. The process of learning allows one to take cognizance of the outcomes of particular decisions. An implication of the process of learning is that it enables decisions to be taken with particular outcomes desired for in advance.
It would be so simple were this chain to be an enclosed truth. If the outcomes were known to all; and the decisions to be taken to reach that outcome were clear; then I wouldn’t need to be writing this article; I’d have long ago been a centi-billionaire chilling on a beach with Jennifer Lopez! Not just me, but each one of us (Actually no, you don’t get to chill with Jennifer Lopez!).
In practice, each stage of the chain above is faced with its own set of uncertainties, encapsulated as risk. The vast majority of the entire gamut of considerations which should go into the objectives pursued may elude us. Our decisions may not be sufficiently informed. The actions which follow from this may not, therefore, be optimal. We may not know of all possible outcomes. We may not all want the same outcome; not for ourselves and not for others. We sometimes forget, that is, no feedback loop. And not a few times, we could be just lazy. We simply do not know.
Some of it is due to the natural limitations of the processing power of the human mind. There are simply too many variables out there and to take cognizance of even more than a few, leave alone all of them, for any defined bracket of time is simply out of reach for most given their natural faculties.
Because of this imperfect knowledge and imperfect learning, we make imperfect decisions based on probabilities (which are sometimes considered decisions; or which could equally be called guesswork) and our actions may not therefore be optimal.
The result could be that our objectives are not met; the outcomes may not be attained; the outcomes attained may not be what we had desired when we set out, so on and so forth. The inefficiencies which are the product to these imperfections cause a lot of grief and, in preventing objectives from being attained, prevent humanity at all levels from attaining their full potential.
It follows, that, with a greater ability to account for ever greater sets of variables to inform our probability calculations affecting our decisions to act, and more efficient and intense learning mechanisms, the quality of decisions made and the efficiency with which commands are enacted would improve sufficiently that the risk of undesired outcomes may be minimized to a significant degree.
If only there could be a way…
It is these gaps in the chain which may be attempted to be filled, efficiencies achieved and risk of undesired outcomes subsequently minimized, through the use of computers. Being dumb devices, computers can do just that - compute - to the maximum of its ability upon precise advance instruction (programming) and in a repetitive manner so long as it is powered. The computations made can be used to either inform decisions to act or not act that are taken; or the same or a different computer could be used to execute the actions following on from the decisions. These actions lead to pre-set outcomes desired by the decision-maker ie the programmer. Computers can collect and process data to scales which would simply overwhelm the average human mind.
This much is known - or rather, has been known - which is the reason for the billions made by the tech sector so far.
The novelty we are faced with is in the last stage of the chain. So far, the feedback loop from outcomes back to decisions - learning - has been left out of the mechanized chain. Analysing outcomes, auditing processes, tying it back to decisions previously made, making further recommendations to how to act to achieve desired outcomes have all been heavily manual, and consequently, heavily laborious; and some would say, highly inefficient.
The link from outcomes back to decisions, usually carried out in an iterative fashion is the intelligence bit of the chain. It is this feedback loop of information from outcomes back to decisions that is being propounded to be mechanized and automated through the deployment of computers that forms the essence of Artificial Intelligence.