A conversation I had recently with an assistant superintendent at the Colorado Association of School Executives convention underscored this idea. We began talking about artificial intelligence and district leader said, “You know, there’s not a week that goes by that my superintendent isn’t talking about AI!” But when I asked what the superintendent wanted to use AI for, the assistant superintendent just kind of looked at me with a raised eyebrow and shrugged.
Artificial intelligence is in the water right now (some might say the Kool-Aid.) However, like many technical innovations from the past couple of decades, what it is and how it works is still a mystery to many people. Or as Arthur C. Clarke’s famous Third Law states: “Any sufficiently advanced technology is indistinguishable from magic.”
But AI is in fact just another technology and not magic at all. At the moment, it’s a technology that’s very useful for a specific class of problems with a set of questions we can ask to determine whether it’s the right technology for our particular problem.
1. Is your problem really that complicated?
See the tweet at the top of the page. In non-programmer terms, it means: “If you can write out a bunch of if-this-then-that rules explaining how a human would make these decisions, then you don’t need AI.” We’ve had software for decades that can take in data and use human-coded rules to generate an output. For example, in a hospital an alarm might go off if your blood oxygen goes below 90. That’s a simple rule that a human can code. You don’t need AI for that.
However, what if you wanted to try to detect heart attacks before they happen by looking at oxygen levels, blood pressure, body temperature, heart rate, breathing and 42 other measurable factors? You couldn’t possibly account for all the combinations of factors if you were to code it by hand. But finding signals in all that data is exactly what AI is good at. Researchers at USC are doing something similar with just a smartphone camera and some machine learning to find patterns in your pulse that would indicate heart disease.
2. Do you have known indicators for the outcomes you want?
Machine learning, which is just one type of AI that we’ll focus on here, relies on humans to provide much of its “intelligence.” For one thing, the humans need to define what the goals are in a specific way so that the computer can sift through various states of the world and determine mathematically how close they are to the desired state. If you’re not exactly sure how to value your examples with both the good and bad outcomes then AI isn’t going to magically find them for you.
3. Do you have a large dataset of those indicators?
There may be debates about the quantity and quality of data that you need, but there’s no getting around needing a large volume of data. If you can describe your desired outcomes but don’t have lots of examples of the data you’d see on your way there (or on your way off track) then the machine-learning algorithm won’t have enough to work with to be valuable.
4. Are your data points discrete and labeled with human-understandable context?
This is key and relates to my earlier post on connecting AI to the action you’d want a human to take as a result. Let’s say you have a bunch of signals from monitoring equipment. You run them through your machine-learning algorithm and you’re able to train it to raise an alarm at the right times–great! Uh, now what? How will a human take that alarm and know where to look or what to do to first, second, and third to resolve the issue? Now instead imagine that you’ve provided some context to the algorithm and then to the human (i.e. “oxygen just dropped ‘dangerously fast’ and pulse is in the ‘high’ range”). Now I might actually know what to do in that case. And over time that type of context will help me develop intuition and trust in the algorithm. Don’t discount a human’s ability to ignore your intelligent algorithm if they distrust its smarts or think it’s responding to some noise in the data.
If you answered yes to all these questions then by all means go forth and create! Artificial intelligence will continue to support human decision-making best when we can provide context with the inputs and produce outputs humans know how to act on.