A framework for AI decision modeling

Everyone is talking about employing AI, but it is just a tool. It is a great new tool; however, tools are used to solve problems. If you apply the wrong tool to a problem, you will not get a good solution. To find the right tool, you have to define the problem. The framing of the problem is more important than the tools that are used to solve the problem.

The authors of Prediction Machines present a simple way to address problem framing called the AI Canvas. There are two parts to the canvas. One, framing the problem. Two, determining the data set needed and how to use it. The decision framing starts with a set of simple questions:

• The action - What are you trying to do?
• The prediction - What do you need to know to make a decision? 
• The judgment - How do you value different outcomes and errors?
• The outcome - what are the metrics for task success?

These four questions help the researcher determine the right tools, but to implement an AI program a second set of questions needs to be answered. 

• Input - What data do you need to run the predictive algorithm?
• Training - What do you need to train the algorithm?
• Feedback - How can you use the outcomes to improve the algorithm?  

The authors present this as a canvas because this framework can actually fit on a piece of paper.  Each box can be filled out to address the set of standard questions. If this is done carefully, it will take some time and effort before there is even a discussion of the techniques that could be employed. 

You may prefer a different framework, but the basic components should be the same: what is the objective, what data is needed, how will a model be tested, what output is needed for taking action, and how will outcomes be judged. Problem management does not start with the tools but with the framing of the problem.