In 1908, Henri Poincare, the famous French mathematician, and pioneer of what we now call machine learning, wrote down what would later become the four stages of the "Gestalt" model of learning and psychology: Saturation, Incubation, Inspiration, and Verification. Einstein, who found Poincare's four stages to be a tad etherial, later suggested a fifth, rather practical step: "Perspiration".
To anyone, such as myself, wishing to better understand the subtleties of artificial intelligence, these five stages of learning (and doing) offer a useful guide to practical visualisations of AI and machine learning. Let's start by substituting some potential modern equivalents:
Saturation: Datification/Digitization and Data Gathering
Incubation: Data Classification and Analysis
Inspiration: Identification of Patterns and Insights
Verification: Real-World Testing and Confirmation
Perspiration: Automated Responses and Process Improvement
Here's a simple (modern) example of how we can apply these five steps: A web designer programs for an e-commerce store programs a testing application to conduct A/B/C testing on three versions of the site's home page to determine which of the three web site designs leads to greater sales on the site (saturation) and sets the program to run for a week (incubation). At the end of the week, the program is instructed to make a decision on the most effective page (inspiration) and push all traffic to it, whereby the conversation rates can be expected to improve over the per-testing rates (verification.)
At the end of the first four steps, it can be argued that we are entering the realm of "artificial intelligence" - defined by Forbes writer Bernard Marr as "the broader concept of machines being able to carry out tasks in a way that we would consider “smart".
But if all we do is move through these four steps one time, we're not really developing intelligence, so much as building a system with a single output. As with evolution, real intelligence doesn't stop at repetition of a single successful action - intelligence evolves.
Indeed, true intelligence, whether natural or artificial, requires that the breadth and depth of the understanding of the owner of that intelligence must grow over time. I still remember the answer my tenth-grade teacher, Rod Davis, provided to us when I asked him for a definition of intelligence. "The more intelligent you are", he said, "the more situations you can deal with." I liked that definition at the time - and I still like it now. So to truly grow intelligence, either naturally or artificially, we need to do what children do, and iterate, and learn more and more from each interaction with the data.
It's at this point that we can begin to define "machine learning". Machine learning is sometimes seen as a subset or "interim step" in artificial intelligence, but an increasing number of people involved in AI view machine learning as the "enabler" - the builder/creator of intelligence. Because machine learning is about taking ownership of the steps and building more knowledge, based on changes to the data sets, the weighting of various forms of analysis, the iteration rates, the testing results... to stretch the earlier analogy, what machine learning is about, is "perspiration". Using machines, not humans.
We have more than a passing interest in this stuff - our investment strategy over the next five years depends in part on our ability to automate tasks related to the discovery and successful processing of new investments.
On that score, we're confident that the current workflow processes predominant in venture can be significantly improved and automated, enabling even the smallest funds to gain from the well-known benefits of portfolio theory. Beyond these benefits, the next phase of our evolution will involve the application of perspiration... sorry, machine learning... to the building of an increasingly robust artificial intelligence that supports venture investing. Stay tuned.