A Better Approach To AI: Realize Business Value First
I’m sure you’ve seen the headlines: The artificial intelligence (AI) market is due to balloon in the coming years, and AI will change the world and solve problems we’ve never had answers to. While I believe that AI can certainly be useful and has some real-world applications already, I think a reality check of what AI does (and doesn’t do) well is in order.
The truth of AI is that production deployment lags behind claims from today’s headlines. According to research conducted by MIT, for example, AI “Pioneers,” which are defined as “organizations that both understand and have adopted AI,” make up just 20% of all organizations. What is creating the disparity between theory and production? In many cases, the way AI is frequently approached is a bit backward.
In our recent survey of line-of-business owners and technical practitioners (engineers and data scientists) across a variety of industries, respondents highlighted some disparate opinions about how to leverage AI at scale. People on the business side were more likely to imagine AI at their company making a huge impact on the world, which suggests too many companies have a pie-in-the-sky approach. In reality, they need a more practical strategy.
We need a different approach.
One of my first AI-related projects in 2016 was with a company that was trying to understand California’s drought issues. The company used visual recognition classifiers to categorize satellite imagery and determine water use on a per-housing parcel basis by classifying photos on a scale of “green” to “not green.” Working with the client, we were able to help California on its journey of reducing water use by 20%.
This AI initiative was successful because the company we partnered with began with — and kept the focus on — a business problem. Unfortunately, too many initiatives fall flat when it comes time to deploy models in production because organizations lack a clear business objective and mistakenly start with ill-defined problems and data.
The most common approach to AI that I see today is companies taking data that’s lying around and then ingesting, cleaning and transforming it. They then train, test and validate a model before wondering why the model isn’t getting deployed to production or working at scale. For successful AI, I believe organizations must change their approach.
Identify a business problem; then get real with data.
First, identify a narrow, specific business problem. Define the problem to solve, and identify the stakeholders and what level of investment (a.k.a. cold hard cash, time, priority, energy) they are willing to put toward solving that problem — and get it in writing. In our case, the problem was to identify excess water use. The problem was clear, as former California Gov. Jerry Brown had signed an executive order to reduce water usage by 25%.
Once you identify the business problem, then you can turn to the data.
Our partner felt that free data from the United States Geological Survey — satellite imagery of housing plots — would help local governments achieve the goal. This data was readily available and could be paired with existing water usage data and other government records, which lent itself well to solving our problem.
Most importantly, the images chosen were real-world examples. We knew that if we wanted a model that could classify images of yards at scale, we’d need actual images of yards. Water-use proxies were overly green lawns or pools in the backyard, indicators of high water usage. Alternatively, solar panels on the roof may have indicated that a household was environmentally conscientious.
The next task was to clean and organize the data for use in a model.
In our example, using the color-based classifiers for precision and recall tests, the client set a threshold with appropriate confidence intervals for data that human annotators tagged.
