Artificial intelligence is defined as the theory and development of computer systems able to perform tasks that normally require human intelligence, such as visual perception, speech recognition, decision-making, and translation between languages. In 2017 the world’s best player of what might be humankind’s most complicated board game “Go” was defeated by Google DeepMind’s “AlphaGo” computer program. The media denominated the source of the win by the following terms: AI, machine learning, and deep learning. Though it may seem extremely futuristic, artificial intelligence in reality is multifaceted and currently part of our everyday lives.
The easiest way to think of their relationship is to visualize them as concentric circles: AI the idea that came first and the largest — then machine learning which blossomed later — and finally deep learning which is driving today’s AI explosion — fitting inside both.
From Bust to Boom
AI originated in 1956 when a handful of computer scientists rallied around the term at Dartmouth Conferences. Decades since then, AI has alternately been heralded as the key to our civilization’s brightest future.
Over the past few years AI has exploded, exorbitantly since 2015. The primary reason for that can be seen by the wide availability of GPUs that make parallel processing ever faster, cheaper, and more powerful. Additionally has to do with the simultaneous two for one practically infinite storage and a flood of data of every stripe – images, text, transactions, mapping data, ect.
Artificial Intelligence — Human Intelligence Exhibited by Machines
In that conference in the summer of ’56 the hope of those AI pioneers was to construct highly complex machines enabled by emerging computers that mimicked the same characteristics of human intelligence. Conceptually we think of this as “General AI” incredible machines that have all our senses, all our reason, and think seemingly just like us humans do. General AI machines have remained in the movies and science fiction novels for good reason; we haven’t been able to pull it off, not yet that is.
“Narrow AI” falls into the concept of what we actually can do with the present technology. Technologies that are able to perform specific tasks as well as, or better than, we humans can. Examples of narrow AI are things such as image classification on a service like Pinterest and face recognition on Facebook.
Machine Learning — An Approach to Achieve Artificial Intelligence
Machine Learning at its essence is a self-adaptive algorithm that improves in analysis and patterns with experience or with new added data. Alternatively rather than hand-coding software routines with a specific set of instructions to accomplish a particular task, the machine is “trained” using large amounts of data and algorithms that give it the ability to learn how to perform the task.
Machine learning’s concept came directly from the early AI founders, while the algorithmic approaches over the years included decision tree learning, inductive logic programming. clustering, reinforcement learning, and Bayesian networks among others. Ultimately, none achieved the ultimate goal of General AI, surprisingly even Narrow AI was unattainable with early machine learning approaches.
Real-time ray-tracing the talk of the 2018 Game Developer Conference: The Difference Between Ray Tracing and Rasterization
Computer vision for many years was one of the very best application areas for machine learning, though it still required an insurmountable amount of hand-coding. People would go in and write hand-coded classifiers like edge detection filters so the program could identify where an object began and stopped; shape detection to determine if it had eight sides; a classifier to recognize the letters “S-T-O-P.” From all those hand-coded classifiers they would develop algorithms to make sense of the image and “learn” to determine whether it was a stop sign. Nonetheless, due to its brittleness and being prone to error computer vision and image detection didn’t come close to rivaling humans until very recently.
Deep Learning — A Technique for Implementing Machine Learning
Deep learning is a subset of machine learning, one that utilizes a hierarchical algorithmic level of artificial neural networks to carry out the process of machine learning. Neural Networks are inspired by our understanding of the biology of our brains with interconnected neuron nodes. However, unlike a human brain where any neuron can connect to another within a certain physical distance, these artificial neural networks have discrete layers, connections, and directions of data propagation.
Traditional programs build data analysis in a linear way, the hierarchical function of deep learning systems enables machines to process data with a non-linear approach. Each neuron assigns a weighting to its input, how correct or incorrect it is relative to the task being performed. For example, the attributes of a stop sign image are chopped up and “examined” by the neurons. More specifically, its octogonal shape, fire-engine red color, distinctive letters, traffic-sign size, and its motion or lack thereof. The neural network’s task is to conclude whether this is a stop sign or not. It comes up with a “probability vector,” essentially a highly educated guess, based on the weighting. In our example the system might be 86% confident the image is a stop sign, 7% confident it’s a speed limit sign, and 5% it’s a kite stuck in a tree, and so on. The network architecture then tells the neural network if it is correct or not.
Neural networks had been around since the earliest days of AI, and had produced dismal amounts of “intelligence.” The error was that even the most basic neural networks were very computationally intensive, an unpractical approach. Still, a small heretical research group led by Geoffrey Hinton at the University of Toronto kept at it, finally parallelizing the algorithms for supercomputers to run and proving the concept, but it wasn’t until GPUs were deployed in the effort that the promise was realized.
Referring to the stop sign example, as the network is getting modified or “trained” it’s frequently coming up with wrong answers. The network needs to see hundreds of thousands, even millions of images, until the weightings of the neuron inputs are tuned so precisely that it gets the answer right practically every time. It’s at that point that the neural network has taught itself what a stop sign looks like; which is what Dr. Andrew Ng did in 2012 at Google.
Scientists at Google created one of the largest neural networks for machine learning by connecting 16,000 computer processors, which they turned loose on the Internet to learn on its own. Ng’s breakthrough was to take these neural networks, and essentially multiply it, increase the layers and the neurons, and then run massive amounts of data through the system to train it. In Ng’s case it was images from 10 million YouTube videos. Ng truly put the “deep” in deep learning, describing the unfathomable layers in these neural networks.
An Ode to Deep Learning, AI Has a Bright Future Ahead
Deep Learning has enabled many practical applications of Machine Learning and by extension the overall field of AI. Deep Learning dismantles tasks in various ways making machine assistance seem attainable. Driverless cars, better preventive healthcare, even better movie recommendations, are all here today or on the horizon. AI is the present and the future. With Deep Learning’s help, AI may even get to that science fiction state we’ve so longed for.