AI that analyzes data to help you make decisions is seen as an increasingly important part of business tools and the systems that do it are getting smarter with a new approach to decision optimization. which Microsoft is starting to offer.
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 what causes them. A machine learning system can learn that people buy more ice cream when the weather is hot, but without a common understanding of the world, it could potentially suggest that if you want warmer weather you should buy more ice cream.
Understanding why things happen helps people make better decisions, such as a doctor choosing the best treatment or a business team looking at AB test results to decide prices and coverage. Which package will sell more products? There are machine learning systems that deal with cause-and-effect relationships, but so far this has been mostly limited to research focused on small-scale problems rather than real-world, real-world systems. because it is very difficult to do.
UNDERSTAND: How to become a machine learning engineer: A cheat sheet (TechRepublic)
Deep learning, widely used for machine learning, requires a lot of training data, but humans 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 test, a financial advisor understanding whether a low-risk or high-risk investment is best for you, or a salesperson asking you to 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 it didn’t miss anything, but if you go to the emergency room with a broken bone, the doctor It would be more helpful to ask how you broke the bone. bones and whether you can move your fingers rather than asking about your blood type.
If we can teach an AI system how to decide what the best question to ask next is, it can use that to gather enough information to make the best decision.
Cheng Zhang, principal investigator at Microsoft, explains: For AI tools to help us make better decisions, they need to handle both types of decisions.
Next best thing
“Let’s say you want to evaluate something or you want to get information on how to diagnose something or classify something properly: [the way to do that] 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 treat patients so they can stay healthy. than – I call it Best Next Works. And for all of this, scalability and personalization are important. “
Put all of that together and you get effective decisions, like dynamic quizzes Eedi online math tutoring service use to find out what students understand well and what they are struggling with, so it can give them the right mix of lessons to cover the topics they need help with, rather than making they are bored with fields that they can already handle.
Multiple choice questions have only one correct answer, but the wrong answers are carefully designed to show exactly what the misunderstanding is: Someone is confusing the mean of a group of numbers for mode or median, or do they not know all the steps to figure out the meaning?
Eedi already had the questions, but it built dynamic quizzes and personalized lesson suggestions using the decision optimization API (application programming interface) developed by Zhang and her team. creates, combines different types of machine learning to process both types of decisions in what she calls end-to-end causal inference.
“I think we are the first team in the world to connect causal discovery, causal inference, and deep learning,” said Zhang. “We allow users with data to figure out the relationship between all these different variables, like what is called. And then we also understand their relationship: For example, how much dose [of medicine] you have given will increase someone’s health, because the topic you teach will increase the general understanding of the students.
“We use deep learning to answer causal questions, recommend the next best course of action in a really scalable way, and make it usable in the real world.”
Businesses often use AB testing to guide key decisions, but that has the limitations Zhang points out.
“You can only do it at a high level, not an individual level,” says Zhang. “You may know that for this population, in general, treatment A is better than treatment B, but you cannot say for an individual what is best.
“Sometimes it’s extremely expensive and time consuming, and in some cases, you just 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, now called the Best Follow-up Question, is available in Azure Marketplace, but it’s in private preview, so organizations want to use the service in their own tools at their own pace. how Eedi needs to contact Microsoft.
For data scientists and machine learning professionals, the service will eventually be available through the Azure Marketplace, or as an option in Azure Machine Learning, or possibly as one of the Cognitive Services packaged like how Microsoft provides services like image recognition and translation. The name could also change to something more descriptive, like decision optimization.
Microsoft has been considering using it for its own sales and marketing, starting with the various partner programs it offers.
“We have a lot of commitment programs to help Microsoft partners grow,” said Zhang. “But we really wanted to find out what kind of commitment program is the treatment that helps partners grow the most. So it’s a question of cause and effect, and we also need to do it in a personalized way.”
The researchers are also speaking with the Viva Learning team.
“Training is definitely a scenario we want to personalize: We want people to be taught with material that will help them best for their work,” says Zhang.
And if you want to use this to help you make better decisions with your own data, “We want people to have an intuitive way of using it. We don’t want everyone to have to be a data scientist.”
Open source The ShowWhy tool that Microsoft built to make causal inference easier The new models have not been used yet, Zhang said, but it has a code-free interface, and the researchers are working with that team to build prototypes.
“Before the end of the year, we will release a demo for deep causal inference from start to finish,” said Zhang.
She suggests that in the long run, business users can get the benefits of these models inside the systems they already use, like Microsoft Dynamics and Power Platform.
“For general decision makers, they need something very intuitive: A codeless interface where I load data, I click a button and [I see] Zhang said.
UNDERSTAND: Ethical policy on artificial intelligence (TechRepublic Premium)
Humans are good at thinking in terms of cause-and-effect, but it’s difficult to build diagrams that show how everything is connected and what is the cause and what is the effect. These decision optimization models build that graph for you, match the way people think, and allow you to question what would happen if you varied different values. It’s very natural, Zhang said.
“I feel like people basically want something that helps them understand ‘If I do this, what will happen, if I do that, what will happen,’ because that is that support decision making.
A few years ago, she built a machine learning system for doctors to predict how patients would recover in different situations.
“When doctors start using the system, they’ll play with it to see ‘if I do this or if I do that, what will happen’,” said Zhang. “But to do that, you need a causal AI system.”
Make better decisions together
When you have causal AI, you can build a system with bidirectional adjustment, where humans teach the AI what they know about cause and effect, and the AI can test that. is that really true or not.
In England, students learn about Venn diagrams in 11. But when Zhang worked with Eedi and Oxford University Press to find cause-and-effect relationships between different topics in mathematics, the teachers suddenly realized they were using Venn diagrams to do quizzes for year 8 and 9 students, long before they told them what a Venn diagram was.
“If we use data, we discover a cause-and-effect relationship and we show it to people – that’s an opportunity for them to reflect and suddenly these kinds of real insights,” says Zhang. This interesting show.
Making causal reasoning end and scalable is only the first step: There’s still a lot of work to be done to make it as reliable and accurate as possible, but Zhang is excited about the potential. .
“40% of the work in our society involves decision making and we need to make high quality decisions,” she points out. “Our goal is to use AI to help make decisions.”
