Machine Learning is slated to be as disruptive and revolutionary as the internet when it was first introduced so it’s time perhaps to visit the subject in as non-technical a manner, as possible, and unpack some of the terms that are conflated when it comes to it. For a quick introduction to machine learning and deep learning (and what the latter means) you should perhaps visit my article on it at Plus Your Business.
For the purposes of the piece here we need to go back to the 50s where it all started in earnest. The phrase “artificial intelligence”, as a matter of fact, was coined in 1956 by John McCarthy, who organized an academic conference at Dartmouth dedicated to the topic. At the end of the conference, the attendees recommended further study of "the conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it. An attempt will be made to find how to make machines use language, form abstractions and concepts, solve kinds of problems now reserved for humans, and improve themselves."
Their premise has also become the basis of what we call artificial intelligence today, to which I shall be returning presently. Also in the 50s (1959 to be exact), Arthur Samuel defined machine learning as "the ability to learn without being explicitly programmed." And he went on to create a computer checkers application that was one of the first computer programs that could learn from its own mistakes and improve its performance over time.
The difference between artificial intelligence and machine learning, from their point of inception, is now apparent. Artificial intelligence is the precise (or explicit) description of functions (and presumably hardware) so that it can simulate human intelligence while machine learning is aimed at helping machines learn on their own and perfect their operation, so it could more correctly be called “machine intelligence”. Machine intelligence is nothing more than an optimization algorithm (a little akin to the Cylon creed of being "the best machines they can be").
What we know today is that the human brain is capable of so many permutations that its explicit expression is virtually impossible to describe and even if that were possible the computing power required to emulate its function makes it impractical to actually try and set up a computer that would try and match it.
This isn’t to say that artificial intelligence is not possible in a more rigorously defined sense nor that machine learning is not part of it. As a matter of fact the AI revolution that we hear so much about in the popular media is the direct result of the convergence of the two.
Where AI meets Machine Learning
Machine learning is the direct result of data mining. Because we truly live in the age of data, in order to navigate it better we find ourselves in need of better tools. Data mining is a field where we use specific data-capture and analysis tools to make sense of data in response to specific needs we have. Machine learning provides just the right combination of energy-efficiency and massive data-crunching power to deliver the results we need.
In the case of semantic search, for instance, data mining uses algorithms designed to index, understand and classify the information on the web, map it onto real-world entities and deliver more accurate results that reflect the contextual ambiguity of real-world situations and the fluidity of natural language usage. The average user sees none of that complexity. All they know is that when they go to Google search they input a query and they get back a reply.
The Los Angeles Police Department uses data mining algorithms in a predictive analytics capacity, allowing it to better manage its resources and allocate manpower in areas more likely to experience crime. The communities they police and the city they serve only see the end results, a reduction in crime that in some areas is as high as 25%.
Not every AI effort in the past included machine learning. The hugely ambitious CYC Project for instance attempted to create an ontology-based knowledge base of everyday objects (entities) through hand-coded facts that feed an inference engine (basically a virtual computer). The limitations of that system can be seen when compared to IBM’s Watson that uses machine learning to data crunch and classify its data and a rule-based AI sitting atop information retrieval (i.e. search) algorithms. Watson is what IBM calls “cognitive computing” which is to say that it is a system that displays applied intelligence.
Applied intelligence (or applied AI) is the result of complex mathematics and computer architecture that provides an extremely good simulation of human intelligence in a narrow field. Google Assistant’s conversational capabilities, AlphaGo’s spectacular defeat of Lee Sedol, Google’s optical character recognition within images and Facebook’s face recognition capability are all examples of applied AI.
Intelligent Machines Are Here
Artificial intelligence is now all around us. It is in the smartphones we have that remind us of appointments without being told and let us know of traffic congestions based on their awareness of our location. It is in the fitness bands on our wrists that monitor our steps, tabulate calories burnt and tell us what more we need to do. It is in the email applications that draw appointments and flights booked from email messages and filter possible spam mail by checking our list of contacts.
None of these are intelligent in the general sense of sentient beings (what we call General Artificial Intelligence, the kind of which we see in Terminator or Westworld). But as computer science and mathematics are making fresh breakthroughs we will realize that what allows us to do specific tasks will increasingly become subject to analysis and replication by machine learning algorithms within AI parameters.
What makes us who we are however, the totality of our consciousness, will continue to evade our reductionist analytical tools for some time to come.