Blue Dots Partners

In this first of three articles on Machine Learning, we delve into the profound impact it is having on many businesses, touching on a few basics along the way. In Part 2 we will make the case that CEOs and leaders should start machine learning initiatives immediately for important aspects of their growth strategy. And finally, in Part 3 we’ll construct a framework for integrating “intelligent machine” thinking in your business and how to align these disciplines with your growth objectives. By understanding what it is, why it is important and how to think about it, our intent is to help you see the strategic importance of machine learning to your business.

In November 2015 a machine emerged that will save thousands of lives. No, it wasn’t a machine you’d typically see next to a hospital bed monitoring every vital sign. This machine sat in an office, with no knowledge of the individuals whose lives it would save.

This was a machine that had been trained to look at lung CT scans of patients being screened for the presence of cancer. It examined hundreds of scans looking for early signs of lung cancer. Expert radiologists with years of experience have been examining such scans for years. But this time, it was the machine that outperformed the experts. The machine had a 0% rate of false negatives – i.e. it never missed a single diagnosis when there was cancer present – no matter how small – in the scan. The radiologists, however, had an error rate of 7% – i.e 7 of every 100 patients with cancer did not get diagnosed as having cancer when in fact cancer was present. Machines just earned some life-saving cred!

Machine learning is driving enormous change in our world today. Daunting problems like accurate image recognition, early detection of disease and racial injustice in the court system – once unrecognizable or thought to be persistent and unsolvable problems – are yielding to machine learning and its related disciplines. Kindred but not synonymous topics like artificial intelligence, big data and robots further contribute to the sense that a major disruption is underway.

For decades, “machine learning” and “artificial intelligence” have been the stuff of doomsday science fiction. Super intelligent PhDs were seen toiling away in the far reaches of research university campuses in search of formulas and data sets that could make a machine possess human-like intelligence. But those arcane perceptions are being steadily eroded and it is time to consider what machine learning means for your business today.

Profoundly, in the business arena, advances in machine learning are acting as new competitive weapons for those prepared to use them. For those that ignore them, they will become significant barriers to entry in some markets. Some businesses will be forced to retool and significantly upgrade their workforce skills and knowledge just to keep pace. Some businesses will simply become obsolete. If you are the CEO of a company or leader of an organization, there’s little doubt “the Machines” (of sci-fi and comic book lore) are already among us and now is the time to take note!

How is machine learning being used today? Amazon and Netflix use big data and machine learning to increase the accuracy of their recommendation engines. Google has perfected its search algorithms through machine learning. Everything from logistics to disease management are benefiting from improved machine intelligence. If we are to harvest the huge potential and benefits of machine learning, we need to bring the principles of machine learning into every day decision making. Let’s begin with a brief definition and some history.

Definition and history

The notion of machine learning was first developed by scientists who were intrigued by the challenges of constructing artificial intelligence. They believed it might be possible to create a machine (possibly a robot) that could simply learn by feeding it more and more data. Ultimately, they hoped, the machine could reach a conclusion that it was not taught directly. So it makes sense that artificial intelligence, robots, and data (i.e. the data which the machine is supposed to learn from) are all closely related to machine learning. Machine learning, however, more narrowly applies to the conditions when the machine has the capability of generating new rules and conclusions beyond that which it was initially prepared to make.

To be certain, machine learning is not new. The quest to build artificial intelligence, for example, goes back at least to 1959. In that year, Arthur Samuel, an early pioneer in computer gaming and later Stanford professor, posed this definition of machine learning: “[the] field of study that gives computers the ability to learn without being explicitly programmed”.   For decades, this “field of study” barely made its way out of research labs.

In the 1980’s, the focus turned to developing so-called “expert systems”. Essentially these systems attempted to learn (or were pre-programmed with) some data or a series of events so that the machines could subsequently attempt to apply that domain of knowledge to a new situation in an optimized fashion. IBM’s Deep Blue – the chess-playing computer – not only put such expert systems on the map, but also demonstrated their limitations, having been defeated by an expert human chess player in its first incarnation.

Machine learning ingredients

What makes a machine “learn”? There are essentially three aspects to every machine learning endeavor – the data, algorithms and the tools to rapidly prototype the application.

Data

Every machine learning exercise begins with a set of data. The technology has not yet advanced to the state where you plug in the machine for the first time and it instantly starts to learn about the world in which it has been placed. Today, we must be intentional about providing information to the machine that it needs to know.

This initial data is often referred to as “training” data because it is used to “train” the machine about specific known relationships or outcomes. For example, if you wanted to use machine learning to help predict commute time congestion on local freeways or plan alternate route improvements, you might start by providing the machine with historical data on traffic patterns, commute times, etc.   To be sure, machine learning can also be applied to data sets about which the machine knows nothing. This is particularly powerful where you might be searching for a previously unknown or unrecognized pattern.

Algorithms and tools

Algorithms are what the machine does with the data you feed it.   One way to think about algorithms is that they are like pre-set photo filters that can enhance the image that you see. Algorithms don’t change the data, they simply organize the data for readability and understanding. Standard algorithms (that go by names like Bayes and Nearest Neighbors) can be applied to any data set, but some algorithms will make better sense of the data than others. By applying a particular algorithm to a data set, particular patterns emerge.

Graphic 1 demonstrates the concept. The upper left hand square is a data set plotted in space. The other nine blocks demonstrate the results of applying an algorithm to the data set. The number in the lower right expresses the degree to which the algorithm explains the data where a “1” means that it perfectly describes the relationships inherent in the data.

Graphic 1: Same data, different algorithms, different relationships

Fast prototyping and rapid iteration

Like photo filters, there are a myriad of algorithms. Applying two different algorithms to the same data set may yield very different outcomes. While choosing which algorithm to apply could be important to learning from the data you have, this turns out to be less critical as it is relatively trivial to change algorithms. By rapidly trying different algorithms it is more likely that an optimal fit will be found between an available algorithm and the particular data set to which it is applied. This is one advantage of machines – they never complain when asked to repeat a task!

Increasingly, real-time customer decisions and manufacturing processes are creating new data sets that continuously feed machine learning algorithms. As conditions change, the machine may or may not be equipped to handle the new conditions. Today’s tools and technologies enable the ability to rapidly update or prototype changes to the machine’s knowledge and rule sets.

Conclusion

Machine learning is out of the lab and is rapidly influencing the trajectory of businesses. It is ready for prime time and is already solving some problems better than humans. Furthermore, the elements of machine learning – data, algorithms and fast prototyping tools – are widely available today.   Competitive and organizational landscapes will be reshaped by this technology. Machine learning will change the trajectory of your business. To learn more about why machine learning matters, look for the second part of this series in our First Monday post next month.