I finally watched the movie Ex Machina this weekend (disclosure: I’m not a movie buff, and am always decades behind on movies!). It was a pretty impressive plot, would have liked the screenplay to be a little better. But what stood out for me was the application of the Turing Test to break the frontier – not just testing for human actions, but for human consciousness. I’ll be suspicious of pretty cyborgs from now on.
That said, I was trying to deconstruct for myself why AI is more hot today than it was say, a decade ago. I strongly believe this is the golden time for commercialization of AI, making it more humanly accessible (no pun intended!) and making it more of a democratized technology than a research lab construct. The three driving forces for AI being ready for real-time enterprise applications today are accessibility of computing horsepower, availability of abundant data to make decisions, and the emergence of an ecosystem of algorithms, tools and frameworks. Let’s explore this a little bit.
Accessibility of computing horsepower
If the objective of AI is to augment human cognitive abilities through computers, then we need a lot of (parallel) computing horsepower that is also cheap. Of course, Moore’s Law drove better utilization of processor real estate and made it possible for large things to come in small packages. GPUs are the go-to hardware architecture to run more complex AI jobs today. All that said and done, the bigger impact was not just the availability of faster technology but the accessibility of computing through public cloud vendors.
15 years ago, to build a piece of AI software, you needed to own a lot of computing hardware, you needed to get it configured, tune it, maintain it and basically have just those basic skills to get started. Having lived through the virtualization and cloud revolutions in the Enterprise, I know this is no small task! Today, thanks to cloud computing, you can rent these capabilities for little and not have to worry about the rest of the headache. Cloud computing was a huge driver in making computing power available for cheap and more importantly, accessible. Today, more than just cheap computing power, both AWS and Azure have tailored offerings for your machine learning projects:
AWS machine learning: https://aws.amazon.com/machine-learning/pricing/
Thanks to this, a crucial class of learning technology, known as neural networks, has gone from being prohibitively expensive to relatively cheap (more on that coming!).
Availability of (abundant) data
In their 2012 article, Big Data: The Management Revolution, MIT Professor Erik Brynjolfsson and principal research scientist Andrew McAfee noted that 2.5 exabytes of data are created every day, and that number is doubling every 40 months or so. A petabyte is one quadrillion bytes, or the equivalent of about 20 million filing cabinets’ worth of text. An exabyte is 1,000 times that amount, or 1 billion gigabytes. Since then, every day, we create 2.5 quintillion bytes of data — so much that 90% of the data in the world today has been created in the last two years alone. But it’s not just about the volume of data available today, the sheer variety of this data, generated by humans, wearable devices, IT systems, machine to machine, social and eCommerce interactions, makes it impossible for humans to comprehend, analyze and act on them without machine assistance.
Not only is there a massive amount and variety of data now available, if all of the past decade’s data infrastructure investments in enterprises is to pay off, organizations will have to figure out a way to leverage technologies like machine learning and natural language processing to make sense of all that data.
A maturing and fast growing ecosystem - algorithms, frameworks & tools
Though the term 'Artificial Intelligence' did not exist until 1956, the advances and ideas from the preceding decades evoked many of the future themes. So, we are talking about a technology that has been baking for a long time with some rapid growth spurts for over six decades (Here is a timeline: https://en.wikipedia.org/wiki/Timeline_of_artificial_intelligence). A more entertaining read is Superintelligence: Paths, Dangers, Strategies by Nick Bostrom.
These six decades were key to the development of algorithms and took frameworks to democratize AI development. Take deep learning for example. Digital neural nets were invented in the 1950s, but it took decades for computer scientists to learn how to compute the astronomically huge combinatorial relationships between a million—or 100 million—neurons. The key was to organize neural nets into stacked layers, where the progressive layers performed partial functions that aggregated to the big picture. This took a while and in 2006, Geoff Hinton, then at the University of Toronto, made a key tweak to this method, which he dubbed “deep learning.” He was able to mathematically optimize results from each layer so that the learning accumulated faster as it proceeded up the stack of layers. The code of deep learning alone is insufficient to generate complex logical thinking, but it is an essential component of all current AIs, including IBM’s Watson, Google’s search engine, and Facebook’s algorithms.
Another parallel development was the awesome frameworks that make it easy for developers to develop their our ML algorithms using popular, often open source elements. A combination of high level programming languages in easily usable frameworks like Spark/MLLib, Apache Singa, Caffe, Neon, Tensorflow or even ready to deploy dev infrastructure with Azure and AWS has made the development of commercial AI applications possible.
This is the golden age for AI. The big question however remains in that we still have to wait and see who the winners and losers are in the long run. The focus should be on solving real problems that deliver tangible business outcomes for enterprises rather than making religious or speculative bets on certain players or vendors — that part will shake itself out eventually!