AI books are usually extremely technical—filled with Python and mathematical equations—or too general, without enough actionable insights.
But a new book called Real World AI: A Practical Guide for Responsible Machine Learning does not fall within either category. Then again, the authors include Alyssa Simpson Rochwerger, who is a director of product at Blue Shield of California, as well as Wilson Pang, who is the Chief Technology Officer at The app. In other words, both have real-world experience on leveraging AI. They also avoid writing with the complex AI jargon.
So what are the main takeaways with this book? Well, there are quite a few. But let’s take a look at some that stand out:
AI Strategy: Yes, it may seem like a stretch for many companies—especially for smaller ones—to put one together. But it is really a good idea to get started. Here are just some of the questions to consider: What kind of data do you have? What processes can be automated? Where are there areas to save costs? How can predictions help?
No doubt, these require a general strategy. The good news is that you do not have to create a detailed document. Even a PowerPoint with fewer than ten slides should be good enough.
Accuracy: This is certainly important as AI is about probabilities. So you want to make sure that a model is grounded in sound outcomes. And this is definitely the case for areas like fraud detection or cancer diagnosis.
Yet accuracy is not everything either. Just because a model is accurate does not make it good. Let’s face it, if the result is that it leads to discrimination or unfairness, then a model should be re-evaluated.
Accuracy can also be misleading. For example, suppose you have built a model that predicts if a person has brain cancer—and it is 99% correct. Sounds great, right? Not necessarily. The reason is that brain cancer afflicts less than 1% of the population. So if a model says “No” most of the time, it will be accurate.
This is why accuracy has various approaches and levels, such as with the use of false positives and true positives.
Understanding AI: You certainly do not have to be a data scientist to be effective with AI. But you still need to understand the limits and use cases of this technology.
The irony is that AI is often not the right solution to a problem. Usually a simple approach is the best. Thus, before you embark on an AI project, make sure that you are focused on the right kinds of problems. These are often where there is a large amount of data that can provide effective correlations.
Focus: In an AI project, a common mistake is to do too much. It’s about boiling the ocean.
Yet the better approach is to start small and focus on clear-cut and measurable goals. For example, you might use AI to help reduce the case resolution time, not try to automate all inquiries that go to the call center.
According to the book: “The best Goldilocks problem is small enough that you can solve it quickly. Problems that involve classifying something into one or two buckets—password reset request, yes or no?—are great candidates. It’s usually fairly easy for reasonable people to agree on those types of classifications.”
Tom (@ttaulli) is an advisor/board member to startups and the author of Artificial Intelligence Basics: A Non-Technical Introduction, The Robotic Process Automation Handbook: A Guide to Implementing RPA Systems and Implementing AI Systems: Transform Your Business in 6 Steps. He also has developed various online courses, such as for the COBOL and Python programming languages.