When we hear the term “artificial intelligence,” a variety of images are invoked—a mix of something like the robot from Lost in Space, Siri, and maybe some of the humanoids from Ex Machina. Whatever comes to mind, the notion generally stretches between what we believe is possible and what already exists in our lives. To add to the problem of defining AI, the terms “machine learning” and “deep learning” are often used interchangeably (which is incorrect). Or, in the least, all three terms exist in the same article without a thorough explanation of their differences.

So we’ve decided to start a series that deep-dives into the continual developments of AI as a whole, with articles ranging from a 30,000-foot view to the nano-specifics of this incredible innovation. To kick things off, we’ve put together a little 101 on what AI looks like today and how machine learning and deep learning influence it. What’s important to remember is that, while this particular article deals with AI in a very general way, the practical application of AI for businesses across every industry is very real and very inevitable. The key, however, to harnessing the power of machine learning and deep learning begins with Big Data, so check out the introductory blog to another one of our new series to begin to see how the combined power of Big Data and AI promise a revolution in everyday professional technologies.

AI: The Largest Circle

Long-time tech journalist Michael Copeland offers a great analogy for understanding the relationship between AI, machine learning, and deep learning: imagine them as concentric circles–AI being the largest circle, then machine learning housed inside of that, then deep learning housed inside of both:

Source: Artificial Intelligence: Deep Learning Trends and Insights for CEOs, CIOs, and IT Teams

AI itself simply refers to any man-made machine exhibiting human abilities such as problem-solving, speech recognition, planning, and learning. These abilities, however, need not be accomplished through the exact mimicking of biological functions, though the concept of artificial neural networks–constructing AI based upon the way neurons are layered in the animal brain–has always been one of the major struggles in the AI mission.

This brings us to the three types of AI:

Strong AI

Like the aforementioned neural networks, the idea here is to replicate human cognition completely and, in achieving this, also teach us more about how humans think. Strong AI is not like a human mind. Functionally, it is a human mind. No true examples of Strong AI currently exist.

Weak AI

Also referred to as “narrow AI,” this sort of intelligence is restrained to a very specific task or tasks. Here we have Siri and the chess-playing computers. Weak AI does not learn, but rather is programmed. This is why Siri can’t answer every question you ask. This is also why a chess-playing computer like Deep Blue can beat a human, but doesn’t play like a human.

In-between

This isn’t a true category of AI, but it helps to cover the flux between Weak and Strong AI. IBM’s Watson is a common example of this gray area. Its claim to fame was beating long-time champions Brad Rutter and Ken Jennings on Jeopardy!. The reason that Watson’s thinking doesn’t qualify as Strong AI is that, when asked a question, it reviewed the entire archive of previously-asked Jeopardy! questions, formulated thousands of possible candidate answers, then determined the most likely correct answer to the given question. So Watson learns, in a way, but not exactly like a human does.

Machine Learning: A Means for Creating AI

Copeland defines machine learning at its most basic as

“the practice of using algorithms to parse data, learn from it, and then make a determination or prediction about something in the world. So 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.”

Common applications of machine learning include:

    • computer vision – medical imaging analysis (diagnoses via X-rays, ultrasounds, etc.); autonomous vehicles (add links)
    • speech recognition – virtual assistants like Siri and Alexa; medical documentation via dictation systems (add links)

Some industry-focused applications of machine learning include:

  • Data Security Kaspersky Lab detects an average of over 300,00 new malware files per day, but, because these iterations are made up of 90% of the same code as their previous version, security algorithms can easily adjust to these variations.
  • Healthcare – at the individual level, computer assisted diagnosis is capable of diagnosing cancer up to a year before it’s likely to be diagnosed by a human; at the population level, machine learning can be used to predict and understand risk factors for disease.
  • Marketing Personalization – think of a Google Display ad for shoes, following you across the internet days after you last viewed those shoes in an online store; also automated delivery of targeted emails.

Deep Learning: Machine Learning’s Brightest Promise

Deep learning is a subset of machine learning where artificial neural networks (ANN) come into play. As the neurons of an actual human brain are layered, so are ANN, the major difference being that ANN must move data sequentially through neural layers, while human neurons are all interconnected.

Copeland gives an excellent example for how deep learning works:
“Attributes of a stop sign image are chopped up and ‘examined’ by the neurons —  its octogonal shape, its fire-engine red color, its distinctive letters, its 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,’ really 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 — and the network architecture then tells the neural network whether it is right or not.”

What sets deep learning apart from the generic definition of machine learning is that deep learning requires extremely large sets of data in order to work. While Watson was restricted to the rules and format of Jeopardy!, as well as an archive of thousands and thousands of past games, a true deep learning machine requires even vaster sets of data.

In Copeland’s stop sign example, the ANN would require more like millions of images in order to calibrate its recognition to the point that it’s almost never wrong. And if the ANN is steering an automatic car, that sort of precision is non-negotiable.

The Saga Continues

For next time’s installment, we’ll talk about all the big players in machine learning, as well as the degree to which deep learning fuels their technologies. As we move through our AI series, we’ll focus more and more on how these cutting-edge technologies translate to business today, not just business a century from now. As a nationally-recognized innovation and entrepreneurial center, tekMountain strives to help businesses to translate the latest tech into practical and actionable strategies within their industries.

If you’d like to learn more about how AI can put your company ahead of the game, contact tekMountain today.

 

This blog was produced by the tekMountain Team of Sean AhlumAmanda SipesBill DiNome and Beth Roddy with lead writer Zach Cioffi.

One Response to “How Does Your Company Fit into the Future?: Artificial Intelligence vs. Machine Learning vs. Deep Learning”

  1. Carl Masi

    Hi Sean and Amanda,

    Glad that you are highlighting AI and its associated applications in your blog. I would be interested in mentoring any startup focused on this technology.

    Carl Masi