Before we began to research this article we had a very clear understanding of what Machine Learning is and what it meant for the search engine industry. Dirk and I have been sharing long conversations about what Machine Learning can do and what it cannot do. Our mutual fear is any piece we published that outlines Machine Learning capabilities will require constant revisions to keep the article up to date. We decided instead to provide readers a place to start learning about the who, what, where, and why of Machine Learning and in the process create a useful reference guide for those trying to make sense out of the incredible pace that technology is advancing.
When a new piece of technology makes a major breakthrough and begins spilling into other industries we always approach the issue with the mindset of an artist trying to visualize the fully realized technology existing in the world. This often leads to misconceptions about the true nature and presence of the technology existing in the world. When we humanize technology the visualizations help to understand better how people will feel about its role in their lives for better or worse.
There is always a balance between excitement and fear when there is a technological advancement made in your chosen industry. Excitement because there are so many new possibilities and fear because you may not be able to adapt fast enough to the new changes. There is a challenge being in an industry where experience offers wisdom, but often to technology that has become obsolete. How do you create a well of knowledge that allows you to stay relevant when the technology adapts so quickly?
Understanding Machine Intelligence, A Brief History
In the early 1950s an IBM employee named Arthur Samuel wrote a program that allowed a machine to learn to play checkers against a human opponent. This is likely the first instance of artificial intelligence at this scale, and for many it represents the birth of modern AI. Shortly thereafter Alan Turing developed a test that was intended to help differentiate between a real human and an intelligent machine. Although the methods of the test have changed substantially, we still call this type of exam a “Turing Test.”
Statistical sciences and research serve as a primary driving force for a significant portion of progress with machine learning. Predictive and analytical modeling of real world information using advanced pattern recognition and large data sets are used with reinforcement learning to solve incredibly complex problems. Unsupervised learning algorithms, when the input data is given no example output, often identifies meaningful insights about data sets that may have otherwise been missed.
Some computer systems are designed to mimic the structure of the human brain, allowing many “nodes” to interconnect on multiple levels. These systems, called neural networks, have been used for handwriting analysis, optical character recognition (OCR), as well as great projects like DeepMind and DeepDream. Neural networks allow many inputs and outputs to be linked together, and each link to change dynamically as the system “learns.” In the 1970s a backpropagation algorithm was introduced that allowed information to flow backward through the network and be compared and saved in real time. This innovation led to the solution of the Exclusive-OR problem and allowed multilevel neural networks to be trained much more quickly.
The First Robot Overlords, Kind Of…
While some applications of machine learning seem downright dull when compared to a walking, talking, cyborg, these systems are a bigger part of our lives than most people even realize. Financial institutions use forms of machine learning for fraud detection, employee access control, and product logistics, Cornell University is attempting to identify individual whales by their call, and many companies have some investment in predicting consumer behavior.
Most digital assistants are, at least in part, powered by artificial intelligence systems. Conversational search, natural language processing, and predictive recommendations are made possible by sophisticated neural nets and supervised machine learning algorithms. Even some mapping and navigation programs use machine learning to optimize your drive to work!
A more intriguing and mysterious application of artificial intelligence lives within Google search, specifically with RankBrain. RankBrain was quietly added to the overall search ranking system in early 2015 and was officially announced later that year. Google reported at the time that RankBrain influenced roughly one-third of all searches, and no had any idea. In mid-2016 it was confirmed that RankBrain was a factor in all searches. According to Google RankBrain “learns” from data offline and is updated once the new output is checked and verified as valid. RankBrain does not exclusively decide the rankings for searches, but represents a significant portion of the overall algorithm.
Microsoft also uses a machine learning system, called RankNet, as a factor in Bing rankings. Although different, both of these systems aim to improve search results by increasing the understanding of queries and user intent.
What is the difference between Machine Learning, Machine Intelligence and Artificial Intelligence (cognitive computing)?
According to Lynne Parker, director of the division of Information and Intelligent Systems for the National Science Foundation, “a broad set of methods, algorithms and technologies that make software ‘smart’ in a way that may seem human-like to an outside observer.” (Katherine, 2016)
Lynne Parker, explains that cognitive computing differs from machine learning in that it is more aligned with how the human mind solves problems. Every day we hear about the advancements in neurology and the progress on how many new functions we understand about the brain. However, science only touches the tiny tip of the iceberg we need to learn to replicate human thought processing. This presents a major problem when computer scientists try to model artificial intelligence to how the brain functions. Rather than being an artificial brain cognitive computing is inspired by the human mind. Cognitive systems are, “a complete architecture of multiple A.I. subsystems that work together,” Parker said.