We all start out with a clean slate or Tabula Rasa. At some point we recognize people and learn their names. Momma. Pappa. And so, the journey begins.
Soon we are taught the alphabet. Letters, pronunciation, learn to draw them cursively, and how they are combined to make words. Those words have meaning and can be combined to form sentences. Which are combined to make paragraphs. Which are combined to make stories. That opens up a whole new world.
Soon we are given assignments, designed to help us learn, and then we are tested, and receive a letter grade validation. If we receive high marks, we are given positive reinforcement and negative feedback for low marks. After much training, we become experts, where we transition from learning mode to production mode, taking a job to earn a living, and our training is put to good use.
So too, with computer algorithms. We create a system by which a machine is capable of receiving input, learning the patterns through positive reinforcement and tuning. After reaching desired levels, in that it can predict statistical probability with a high degree of accuracy, the system has graduated from it's learning stage. We then send real data through as input, to the trained model and the algorithms can produce some great results.
DeepMind, an Artificial Intelligence division at Google, was able to produce an environment in which the machine taught itself Atari games without any prior knowledge or expected results. Soon it mastered the game at expert level, better than almost any human. Later, it taught itself the game Go, after being fed millions of games which were scraped off an online site. The machine went on to beat one of the best players of the past decade, surprising many.
So as we can see, machines are capable of learning and performing great feats. Algorithms learn from input data over time. Then duplicates the behavior with high degree of accuracy, not just facts, but strategy. Although some may say this is the brute force way of training, it seems to work.
What's next? Well, we need more data. More input. More algorithms. More training. More developers. There are now tools available to train and monitor electronic activity via websites, for instance, gaming sights. You point your AI algorithm to the site, it learns in the background.
As more of the world goes online, with smart phones, social media, payment transactions, etc. the amount of data and events are growing exponentially. Not only that, those events are tied to people, places and things. The AI is learning the patterns from multiple sources continuously.
One way to train a model is through the technique called Supervised Learning, in which the data fed into the model is known in advance. If the data is composed of images, perhaps the label says dog or cat or bird or mountain.
Yet there's another technique used to train the model's known as unsupervised training. The input data does not have any labels or identifiable information, so the machine has to learn more without assistance. There are techniques to find the patterns, but it's more complicated than Supervised Learning.
As the data grows, and becomes more available, the AI machines have more input to crunch. Like completing a jig-saw puzzle, each single piece doesn't add much value. As you put more pieces together, a picture starts to take shape. AI will eventually put the pieces together across multiple domains, to create a giant puzzle comprised of bits and pieces to complete it's own puzzle.
Watching patterns over time is what AI systems do. Patterns can be interpreted. AI systems can detect patterns from the data, similar to fraud detection system's looking for anomalies.
They used to say any device hooked up to the internet can be compromised. How about AI systems. Anything electronic, from Internet of Things devices to cameras, social media, smart phones, payment transactions, banking accounts, location data, etc. could potentially be ingested, processed and interpreted to detect patterns over time.
Perhaps we have passed the basic learning of the letters of the alphabet, to forming words and sentences and paragraphs. And we about to graduate from our 12 years of learning, into the real world. Where our knowledge is put to the test. Artificial Intelligence is graduating from school and entering the real world. Hang on to your hats. AI is ready for takeoff.
I signed up for the Hortonworks Certified Associate exam last Thursday. Figured if I sign up, I'd have to take the test. And if I tak...
This blog post is in no way an attempt to steal other people's work. It's basically an conglomeration of notes from research I did...
Saw a post today on Twitter, " Microsoft releases CNTK, its open source deep learning toolkit, on GitHub " This is big news. Be...