Is what is happening in analytics, in industry, an evolution or a revolution?
Analytics is the science of data analysis. Those who practice analytics self-identify as analyst, digital analyst, marketing scientist, data engineer, researcher, among many others. Tukey (1962, The Future of Data Analysis, The Annals of Mathematical Statistics, (33), 1) called them all practitioners.
The goal of the practitioner depends on their context. That context largely, but not always, depends on the state of knowledge, state of the culture, or sometimes, normatively, the state of maturity, of the group they belong to.
Large organizations can have a large amount of difference within them. It’s not uncommon for an operations department to be extremely mature and for its marketing department to be comparatively immature, and vice versa. Small organizations are more likely be homogenous on the inside, because the opportunity to develop differences is much lower. If its senior leadership has a high degree of maturity, it’s often the case that the stems and the leaves of the organization will have a high degree of maturity. And vice versa. Some leaders don’t know what they don’t know.
For some practitioners, their entire goal is to flood the organization with enough data to induce inquiry and curiosity. For others, their goal is create justifications for predetermined courses of action or to engage in post-hoc exculpatory explanation – both are forms of convenient reasoning. For others, it’s to establish a surveillance apparatus directed either at their employees or at intra-organization competitors. For some it’s about using information to help their people learn how to make the decisions in their control just a little bit better. For some, it’s about the optimization of existing processes. For others, it’s about the discovery of new processes. For others, the goal is to predict better futures. For few, it’s to automate decisions. For some it’s entirely about creating sustainable competitive advantage through learning. Many share a common goal creating better outcomes. Some exist to persist.
The diffusion of deep learning technology is at the core of what’s changing.
Inside the coma of all the hype of Artificial Intelligence, there is a core disruptive technology called deep learning. Deep learning is broadly a set of mathematical techniques rendered commercially viable in just the past few years. The techniques were discovered in the eighties. The processing power required to make it commercially viable came in the early 2010’s.
Deep learning is not Linear Regression / Ordinary Least Squares (OLS) alone. OLS was discovered in the late 1700’s, and diffused into policy and commercial life slowly in the early 1900’s. It became increasingly useful as a result of machine readable data formats in the mid-20th century, and by the early 1980’s, some executives were doing their own linear regression to make business decisions. OLS took a long time to diffuse compared to deep learning.
Many deep learning algorithms reduce a problem space into two dimensions where ultimately a straight line can be used to make a decision, but I wouldn’t reduce all of deep learning to being linear regression with networks.
Deep learning is hyped in part because it is an effective technique for making accurate predictions when given large volumes of data.
If it is the case that the diffusion of deep learning is accelerating, I believe we are heading towards revolution. If the diffusion is linear, I believe we are evolving.
There are at least two trends: the rate at which organizations are maturing in their use of analytics, and the rate at which deep learning technology is improving.
It’s very difficult to quantify the rate at which organizations learn because it’s often not linear. When a system goes from one state of equilibrium to another, it is often marked by a line that is static, an exponential change, and then more static. With so much noise, it’s hard to tell for sure.
It’s easier, but still difficult, to quantify the rate at which deep learning is improving. To be generous, deep learning researchers are engaging in a very spirited debate about which measures to use to quantify their progress. I’ll leave it at that.
There are a few ideas in deep learning that are over and above regular optimization/search research. I’m optimistic that the seeds for the next revolution are there.
I can’t tell if, in aggregate, there’s an evolution or a revolution in analytics. The truth likely depends on your perspective and where you are. From where I sit, some days it feels like a total revolution. And then I have some days that it feels as slow as biological evolution. I can’t tell. Maybe you can.
What it means
It means interesting times for the analytics practitioner, the data engineer, and the machine intelligence developer.
If, in aggregate, there’s a revolution afoot, it means exponential change at the human-decision boundary.
If, in aggregate, there’s it’s an evolution, it means a longer arbitrage opportunity.
It took a long time for OLS to be understood to the point that it was deemed safe enough to be used to inform decisions. There was a lot to explain and it took a long time for leaders to trust the technology. What is a square? What does mean even mean? What’s special about least? What is regression? You’re telling me you can boil all of that data into an equation? It took awhile for each of these concepts to become understood well enough to be comfortable. To feel safe around. It takes time for people to learn.
It may take a long time for deep learning to be deemed safe enough to be used to inform decisions by most organizations. What is a neural net? What is learning? How can it learn? Is it fair? Is it accountable? Is it transparent?
Deep learning has been embraced by organizations that value learning. It’s like it’s an advantage on top of an advantage!
It may mean that for those that this technology is a revolution, they may enjoy a significant advantage over those organizations that are evolving. And those organizations may enjoy exponentially greater margins over those that are evolving.