AI that analyzes data to help you make decisions will become an increasingly important part of business tools, and the systems that do that are getting smarter with a new approach to decision optimization that Microsoft is starting to make available.
Cause and effect
Machine learning is great at extracting patterns from large amounts of data, but not necessarily good at understanding those patterns, especially in terms of their cause. A machine learning system could teach people to buy more ice cream in warm weather, but without some common sense from the world, it’s just as likely that if you want the weather to get warmer, you’ll need to buy more ice.
By understanding why things happen, people can make better decisions, such as a doctor choosing the best treatment or a corporate team looking at the results of AB testing to decide what price and packaging will sell more products. There are machine learning systems that deal with causality, but until now this has been mostly limited to research that focuses on small-scale problems rather than practical, real-world systems, because it was difficult to do.
TO SEE: How to become a machine learning engineer: a cheat sheet (TechRepublic)
Deep learning, which is widely used for machine learning, needs a lot of training data, but people can gather information and draw conclusions much more efficiently by asking questions, such as a doctor asking about your symptoms, a teacher giving students a quiz, a financial advisor who understands whether a low or high risk investment is best for you, or a salesperson who will let you talk about what you need from a new car.
A generic medical AI system would probably walk you through an exhaustive list of questions to make sure you didn’t miss anything, but going to the emergency room with a broken bone makes it more helpful to the doctor to ask how to move the bone and if you can move your fingers instead of asking your blood type.
If we can teach an AI system to decide what is the best question to ask next, it can use that to gather just enough information to suggest the best decision.
In order for AI tools to help us make better decisions, they need to handle both kinds of decisions, explains Cheng Zhang, a principal researcher at Microsoft.
The best next thing
“Suppose you want to assess something, or you want to get the information on how to diagnose or properly classify something: [the way to do that] is what I call the Best Next Question,” Zhang said. “But if you want to do something, you want to make things better – you want to give students new teaching materials so they can learn better, you want to give a patient treatment so they can get better – I call that Best Next Action. And for all these things scalability and personalization is important.”
Put all that together and you get efficient decision making, like the dynamic quizzes that online math tutoring using Eedi to find out what students understand well and what they struggle with, so that it can give them the right mix of classes to cover the topics they need help with, rather than boring them with areas they can already handle.
The multiple choice questions have only one correct answer, but the wrong answers are carefully designed to show exactly what the misunderstanding is: does someone confuse the mean of a group of numbers with the mode or the median, or just don’t know all the steps to get to the mean? to decide?
Eedi already had the questions, but it built the dynamic quizzes and personalized lesson recommendations using a decision optimization API (application programming interface) created by Zhang and her team that combines different types of machine learning to handle both types of decisions in what to terminate their final causal inference.
“I think we are the first team in the world to bridge causal discovery, causal inference, and deep learning together,” Zhang says. “We enable a user who has data to find out the relationship between all these different variables, such as what calls what. And then we also understand their relationship: for example, how much the dose? [of medicine] you have taught will improve one’s health, to what extent what subject you teach will increase the student’s overall understanding.
“We’re using deep learning to answer causal questions, suggest in a really scalable way what the next best action is, and make it usable in the real world.”
Companies routinely use AB testing to guide important decisions, but that has limitations, Zhang emphasizes.
“You can only do it at a high level, not at an individual level,” Zhang said. “You can find out that for this population in general treatment A is better than treatment B, but you can’t say which is best for every individual.
“Sometimes it’s extremely costly and time-consuming, and for some scenarios you can’t do it at all. What we are trying to do is replace AB testing.”
From research to no code
The API to do that, currently called Best Next Question, is available in the Azure Marketplace, but is in a private example, so organizations wanting to use the service in their own tools the way Eedi should contact Microsoft.
For data scientists and machine learning experts, the service will eventually be available through the Azure Marketplace or as an option in Azure Machine Learning or possibly as one of the packaged Cognitive Services in the same way that Microsoft offers services such as image recognition and translation. The name can also change to something more descriptive, such as decision optimization.
Microsoft is already considering using it for its own sales and marketing, starting with the many different partner programs it offers.
“We have so many engagement programs to help Microsoft partners grow,” says Zhang. “But we really want to know what type of engagement program is the treatment that will help a partner grow the most. So that’s a causal question, and we need to do it in a personalized way as well.”
The researchers are also in talks with the Viva Learning team.
“Training is definitely a scenario that we want to make personal: we want people to be taught with the material that will best help them in their work,” Zhang said.
And if you want to use this to make better decisions with your own data, “We want people to be able to use it intuitively. We don’t want people to have to be data scientists.”
The open source ShowWhy tool Microsoft built to make causal reasoning easier to use doesn’t use these new models yet, but it has a no-code interface, and the researchers are working with that team to build prototypes, Zhang said.
“Before the end of this year, we will release a demo for the deep end-to-end causal inference,” Zhang said.
She suggests that in the longer term, business users can take advantage of these models in systems they already use, such as Microsoft Dynamics and the Power Platform.
“For general decision-makers, they need something very visual: a no-code interface where I load data, I click a button and [I see] what are the insights,” said Zhang.
TO SEE: Ethical Policy for Artificial Intelligence (Tech Republic Premium)
People are good at causal thinking, but making a graph that shows how things are related and what is cause and effect is difficult. These decision optimization models build that graph for you that fits the way people think and allows you to ask what-if questions and experiment with what happens when you change different values. That’s very natural, Zhang said.
“I feel like people fundamentally want something to help them understand ‘If I do this, what happens, if I do that, what happens,’ because that’s what helps with decision-making,” Zhang said.
Several years ago, she built a machine learning system for doctors to predict how patients would recover in different scenarios.
“When the doctors started using the system, they played with it to see ‘whether I do this or that, what happens,'” Zhang said. “But for that you need a causal AI system.”
Making better decisions together
Once you have causal AI, you can build a system with two-way correction where people teach the AI what they know about cause and effect, and the AI can check if that’s really true.
In the UK, school children learn about Venn diagrams in year 11. But when Zhang worked with Eedi and the Oxford University Press to find causal links between different subjects in mathematics, the teachers suddenly realized that they were using Venn diagrams to create quizzes for students in Years 8 and 9, long before their graduation. told them what a Venn diagram is.
“When we use data, we find out the causal relationship and we show it to people — it’s an opportunity for them to think and suddenly these kinds of really interesting insights appear,” Zhang said.
Making causal reasoning end-to-end and scalable is just a first step: there’s still a lot of work to do to make it as reliable and accurate as possible, but Zhang is excited about the potential.
“40% of jobs in our society revolve around decision making, and we need to make high quality decisions,” she noted. “Our goal is to use AI to aid decision-making.”