Buzzwords in Big Data: What do they mean?
Today, we’re going to go more into another couple of terms, that we find frequently — almost in every case, in the same breath as ‘big data’. Those two are “artificial intelligence” and now more commonly, “machine learning.”
While some people use them interchangably, that is not the case. They are in the same realm — ad closely related, but they are not the same.
Once we define each of them, can we properly understand what each means in a specific context — and in the context of big data.
Even a little over a decade ago, the common man’s idea of Artificial Intelligence — or AI — was a cyborg, either adapted or made to look only vaguely human-like, in some subtle — and very mild way. More recently, AI has taken on some dark connotations — in doomsday-sayers who think “robots will take over the world,” or something along those lines.
The first misconception there, for which we could blame pop culture, science fiction and films of the past, is that AI is a literal robot. It’s not.
It is not so much a tangible thing as it is the IDEA that a machine can carry out complex tasks one would normally expect a human to do, or function in a way human thought and logical processes work.
That starts off really simply. Artificial Intelligence is when a computer is able to ‘see’ or perceive things as humans do, to imitate what people expect as “human” thought processes, recognising speech and speech patterns, decision making and more.
A couple of basic examples in real life — that many computer users have seen in real-time, so to speak, come to mind.
Say you have a Facebook account. Years ago, it couldn’t tag people unless you clicked on people in photographs manually, or parts of photographs, and labeled them. Then, a few years ago, it was able to recognise faces, in the sense of being able to detect that there was a face at all in a particular part of the photograph, and how many.
Now, we’ve seen that same social media not only single out faces but indeed, which faces are there — and so the next time your social media asks you if you “want to tag X in this photograph”. That’s an example of AI — and just coincidentally also an example of machine learning, but I’ll explain more about that later.
In 1996, chess Grandmaster Garry Kasparov took on IBM’s AI — Deep Blue in a game of chess — although Kasparov did win a number of subsequent games. Chess cannot entirely be predicted — and the skill of a GM lies in thinking at least two steps ahead of his or her opponent — and being able to predict not only the moves the opponent will make, but what steps the opponent thinks they may take — to feint then.
With every one of those aspects in place, Deep Blue still managed to pull off a defeat of Kasparov.
So what then is machine learning? It is more an application of the use of Artificial Intelligence (further referred to as AI), where systems learn something — and then hone and independently improve that skill rather than have a human need to constantly improve, change and manually update those systems.
Back to that example of the evolution of tagging and recognising photographs on social media that I was telling you about. Remember I said that back in the day, you could click on any part of the photograph to tag anything — whether on Facebook or not, then it went to specific faces, and in the past few years, that ‘tagging’ feature could now zoom in on — and recognise — your friends and acquaintances?
That’s machine learning. When machines have access to that large amount of data — as Facebook does. If you’re on the network you will have pictures of yourself there, leading that AI — with more and more accuracy and detail — to be able to recognise a specific person — or you, yourself — off a photograph you took whilst out to dinner with your closest friends. But what looks like a photo to you — made up of visible, and measurable, pixels, for the machine is data that it parses in numbers — just like every other bit of data out there.
Interestingly, the connections it (the machine) forms with other bits of data to be able to parse it the way it does, is quite similar to the human brain and its network of nerves that connect and work together — and that’s what gives us what’s known as a “neural network.”
Just as the human brain and nervous system understands, makes connections and decides on the best course of action for the body at any given point in time, so does the neural network. When a specific algorithm uses a neural network — the algorithm — here, in essence, the brain, and the neural network — the nervous system — will figure out from the data (like stimuli, so to speak) the next course of action, how it can be improved, and how a specific goal can be achieved.
You might remember that photo showing how a ‘zoomed-out’ view of the universe looks like the nervous system. Well, even the tiniest bits of data and how humans have designed advanced machines to function resembles it, albeit not in the same way optically.
Neural networks are a big part of machine learning. And from those neural networks comes deep learning.
Similar to how the human brain makes certain networks or pathways, when neural networks work in complex layers, deep learning systems are able to use — and learn, over time, how to hone their results.
Remember how last week, I told you about how a linear system might use a single way to figure out a solution to a set of data? Well, a deep-learning neural network will use multiple factors to detect and hone in on a ‘solution’ for a set of data — therefore making it far more accurate in terms of results.
Here, I’ll remind you of the advertising example I’d used earlier. Say you have a product, and that product is aimed at men from 18–25 years old. That’s only one factor, and in a world population of 2.07 billion, that’s a lot.
Then you would want to hone it down to 18–25 year olds in a specific geo-location — for example, let’s say in the northern part of the United States — or to be even more specific, maybe all northern, coastal states in the US.
Then, and this is all from data users themselves provide on the internet, the neural network can learn that a male user between the ages of 18–25 who say, has provided his interests as swimming, surfing and sports, and has displayed a certain type of purchasing behaviour, would be likely to purchase a specific product.
Ever seen a sponsored post in your feed either vaguely — or even closely — related to the products you use, your internet search or your ‘likes’ and interests on social media? That is not manually entered, but machine learning in terms of targeted advertising.
Or a simpler example. Let’s say you want to catch a criminal — a male between the ages of 45–50. You (in this case, let’s say you’re a police officer) — are trying to catch him on the basis of knowing his general geographic location, but a social media post or some internet activity of his tipped you off.
The first layer of your neural network will just be the search term, or the post that was made. That is the basic foundation of the information you’re looking to go on. Then, that information percolates down to the second layer of the network, which can give you a internet service provider and an IP address. That IP gives you a tangible location on the map that you can work from. With several layers of information in that neural network, you then have a concrete way to catch — and apprehend — the criminal.
Lately, neural networks are even being used in medicine, in a fascinating way. It is quite possible the human mind may miss some data — but machines do not, so neural networks and deep learning have been analysed to use pharmaceutical data and information about diseases to see if existing medications have uses or implications for newer illnesses and diseases. ONE PART of the long and short of it?
Deep learning in a neural network (and in an ideal world), could help scientists cure cancer some day. Let’s hope that day is in the near future.
Did I miss any buzzwords you’d like to know? Leave a comment or email us, and we’ll be sure to include it in next week!