AI, ML and Deep Learning @ the Movies

I firmly believe we don't need to look any further than our movies to see the future of technology. Growing up, shows like Star Trek and movies like 2001: A Space Odyssey, Terminator and Robocop captured our imaginations and fascinated us with super cool tech - stuff that today is becoming more and more common place and real! Granted we still can't teleport, but is it truly that far away from being a reality? It's just 0's and 1's, right?!... :)

But beyond the futuristic visions, I think sci-fi movies provide a simple analogy to explain some topics - like the other day, I was trying to figure out an easy way to explain the differences between the latest cognitive technology buzz words, and for the life of me, everything I could find was a little too deep or confusing for someone that is non-technical. Now I will confess - unless you are a sci-fi aficionado, this article may only confuse things more for you, so please take my definition of "simple" with a grain of salt and consider this a movie buff's approach to demystifying Artificial Intelligence, Machine Learning and Deep Learning.

First of all, think of AI to be everything and anything that we seem to have taught a machine to do, a state where the machine seems to be capable of thinking, and possessing some kind of intelligence. From enhanced computers like HAL 9000 in 2001: A Space Odyssey, ENCOM 511 in Tron, to sentient humanoid forms like Skynet's T-800 in Terminator or C-3PO from Star Wars, all of them possess Artificial Intelligence, but with varying levels of cognitive capability - simple, complex or complicated.

The simplest form of AI is the case of the robot that is programmed to do one thing and do it well. You could say the Battle Droids in Star Wars Phantom Menace were probably the closest "dumb" robots you could find. Yeah we gave "life" to a chunk of metal that could wage a war, but they were far too predictable and easily defeated in because they were programmed to simply identify an enemy and shoot. To be able to put up a decent fight or think of evasive maneuvers, these Droids needed to be better trained and capable of self-learning. In the TV show Small Wonder, Ted attempts to do this every so often by reprogramming Vicki, when she comes across new interactions that Ted simply hadn't thought of. So that's great, but with the advances in computing and data, the question that was starting to be asked was, why couldn't Vicki reprogram herself?

That's where the disciplines of Machine Learning and Deep Learning come in. They are applications or subsets of AI where the machines can now go beyond their original programming by "improving" themselves to solve more complex problems. ML is interspersed throughout our lives today, from the facial recognition feature in our phones to the spam filter in our emails, that use algorithms to train themselves to make better predictions over time. Consider the movie Minority Report where the entire crime fighting system was based on a system that identifies patterns and tries to prevent a crime before it occurs, a whole lot of data and prediction by the "precogs" were central to their learning algorithm. Another controversial example might be the whole concept of the Matrix where it is essentially a program that has bugs or "anomalies" and with every iteration using ML, the machines get better at defeating Neo and destroying Zion, till the sixth iteration where Neo reacts unexpectedly and Agent Smith is confused at the turn of events (since he hadn't "trained" for that situation).

Deep Learning is where ML now starts to mimic the human learning process, and starts becoming more capable of even surpassing what the human mind can do. A good example is one of autonomous driving - using ML we've been able to solve computer vision problems, program computers to win Jeopardy and defeat Garry Kasparov, but while being extremely complex problems, these were challenges that were definable and programmable. On the other hand true autonomous driving is simply too complicated to even begin listing out the hypotheses we would have to solve for, to perfect it - think about it, when you drive, you're monitoring traffic, weather, maps all of which are readily available data points, but what about that group of school kids who seem to be not paying attention, the distracted driver, or the flashy supercar in your rear view that you know will try something stupid? DL attempts to take all of these real-time factors into consideration to try and mimic what the human brain would do when faced with a similar situation. And if Tesla's progress is any indication, in the very near future, I expect to see KITT from Knight Rider in my garage.

One of the strongest examples for Deep Learning is probably Data's character from Star Trek Enterprise. Those that followed Data's journey through the series would remember that Data started off as a quirky and awkward machine with human-like physical features and powerful computer internals, that could truly do superhuman math and data processing but couldn't "compute" anything outside of what his programming could make sense of. The character was defined by its struggle with emotion, which was an extremely complex human attribute. However over the years, he learned how to read, recognize and express emotion, all by reprogramming his neural "positronic net". This evolution is evident when Data no longer has logical reasons for why he feels a certain way or has said something, but does respond to emotion. (Now some may say it was the "emotion chip" that actually enabled it, but the jury is out on whether he really needed the chip to begin with.) So it is probably safe to say that in movies, the highest level of AI or Deep Learning is correlated to machines and computers displaying some kind of emotion.

The paradox here is, in the case of humans, we say logic and reasoning is thrown out the window when we let emotion cloud our judgment - or we get too intelligent for our own good! And therein lies the central plot of all sci-fi movies - computers becoming sentient and attaining the highest level of consciousness through learning and control. There are innumerable movie examples of the friendly computer just trying to follow its "prime directive", but perhaps the most chilling one I can remember is when HAL in the most indifferent of voices says "I'm sorry Dave, I'm afraid I can't do that!" As this space continues to evolve exponentially, how real of a concern is it? Will the I-Robots enslave human kind? Can the Umbrella Corporation's Red Queen go rogue? And more realistically, what kind of societal change are we going to see simply due to the enhanced automation of the workplace? The answer is - we don't know, but that's why the FAAMG and IBM have come together to form the Partnership in AI to research and set guidelines in the field of AI. Till then, here's to our movies that also have the solutions to these doomsday scenarios :)