Artificial Intelligence: Machine Learning
I have written, some time ago, several columns covering artificial intelligence in my attempt to dispel the rumors, gossip and hype surrounding the notion.
Plagiarizing nature …
In this month’s column, I wanted to revisit the subject. In particular, I thought I’d take on the challenge to better understand “machine learning” and to break it down in terms of where we are with it; what is currently understood and, ultimately, where we are heading. Firstly, I’d like to start by echoing my persistent viewpoint regarding what is understood today as artificial intelligence (AI) and reiterate that it is currently nothing more than clever programming and smart technology (you can read more about that in “Artificial intelligence: I think, therefore I am?”).
Machine learning (ML) along with deep learning for me, follows a similar theme in that it is again nothing more than clever programming. However, I have become somewhat lost in the history and the parallels used to shape and architect how ML mimics the behavior and function of the human brain with software.
You see, many researchers and analysts regard machine learning as a subset of AI, with other facets such as deep learning, predictive analytics, robotics, automation and self-diagnosis. I largely agree, and I also think it’s a great way to organize and shape the hypotheses surrounding AI. More so, we can perhaps liken it to the structural components within the human brain, which I touched upon in, “Artificial intelligence: embryonic.”
So, machine learning is a large field of study which broadly surrounds the development of software (a computer program) that automatically improves with experience, according to Tom M. Mitchell in his book, Machine Learning. The subject is not entirely new either, since it’s been circulating for several decades and in many guises. However, today we now associate the term with “big data” and data modelling or science, where retailers, for example, like to harvest information about your shopping habits so, in turn, they can more accurately target their advertising.
It's just clever programming!
Although, many academics, researchers, analysts and, indeed, the wider industry have yet to agree upon a definitive statement as to what ML actually is; for me, it’s essentially an ability for a software program to have an adaptive set of processes, based upon experience through the data it has received, both previously and presently. I can arguably compare it to an “empathic” agent, of sorts, since the software program has only the ability to “see” our world, if you like, through data. As such, from its perspective, it then interprets a representation and an understanding of “its” world, where it can, in turn, make empathic and educated decisions based on what it understands presently and what it has “learned” in the past.
I know what I have done: I have seemingly painted a picture where this piece of software is “all-knowing” and that’s certainly not the case. Nevertheless, it is certainly one healthy base of source code and it again resonates with my common theme, that is this AI-thing is still nothing more than clever programming!
Machine learning is so much more
I confess that I may have embellished my portrayal of a mere software program or algorithm to be more than it is, perhaps, in fact, more capable than it really is because it does, after all, lack consciousness – something which can often be more of a curse than a blessing to us mere humans. As conscious beings, we are all innately “programmed” with a set of rules that govern our thinking and decision making. Our belief systems are nurtured by our families and friends and our social values are punctuated by the many rights and wrongs we learn throughout our lives, shaping our personalities and impacting upon the choices we make every day.
In contrast, our “intelligent agent,” is an adaptive program that can make decisions without the conflicts of an individualized belief systems. As a result, its ability to diagnose and resolve is often far more accurate, making it an invaluable resource for much of today’s industry.
It's tarot card reading
There are however times where predictions simply can’t be taken as accurate. For example, the algorithms used to predict weather can often be completely wrong as they are prone to continual environmental change, making them a moving “goal post,” if you like. Yet if we were to take the same algorithms and apply them to a “known” scenario such as manufacturing and production, for example, then this is a largely consistent and non-volatile environment. Moreover, we can liken this to a “black box” where at one end we place “A” and expect “B” at the other – what happens in the black box is managed by our intelligent agent.
I’m sure we will continue to develop ever-enhanced software and algorithms that will, over time, be able to more accurately provide our weather forecasts but, right now, it’s akin to a random tarot card reader attempting to predict our future.
Until next time …
But I’ll end with focusing on the incredible possibilities that this technology presents. What, for example, if we tasked our intelligent agent to tackle decisions that ultimately solve real-world problems? This conjecture is not entirely new, and some may say this is already happening in abundance with artificial neural networks (ANNs) being used to detect cancer cells far more accurately than actual physicians; in fact, this is something that my wife Sarah talks about in “Fighting breast cancer with AI early detection.”
And, in this example, our algorithms have been put to good use, with excellent pattern recognition techniques and are certainly well-placed for future medical development, where we might develop them further to detect patients with the coronavirus and help predict how the virus might mutate.
So, this is where your “AI psychologist who’s always learning” Dr. G signs off.
Originally published in Technically Speaking.