Michael Schmidt is one of the “World’s Most Powerful Data Scientists” (Forbes) and founder of Nutonian, an artificial intelligence (AI) company that automatically generates forecasts and strategic business recommendations. We recently checked in with Michael to get insight on data science and the future of machine intelligence. Here’s what he shared:
Can you talk about your professional background and interest in data science? What drew you to this field?
I pretty much stumbled into data science in grad school. I was studying computational biology at Cornell, and what I kept noticing was that all of these grad students and scientists were spending a majority of their time crunching numbers and building models to understand their data. The problem was that this was taking up a significant amount of their time, when the reality is, that part of the research is neither glamorous nor particularly rewarding … and on top of that, there’s a lot of potential for human bias to sneak in.
What I thought was: What if there were a way to help scientists bootstrap the process of extracting meaning from data? What if I could leverage some of my coding experience to build a computer program that could automate the process of creating predictive and explanatory models?
What came out of that was a free prototype version of Eureqa, an AI-powered modeling software that was downloaded about 80,000 times in the first year or two. I was fortunate enough to meet some investors who saw the potential of Eureqa in the business world, and that’s when I started to build out my current company, Nutonian. Eureqa is the only software on the planet that automatically prescribes solutions on how to improve your business.
How has data science evolved since you started your career?
About 10 years ago, data science was uncool and even intimidating to most people. I think it’s starting to get a better reputation now, only because people have started to acknowledge that it could be the key to unlocking a whole lot of value from the data they’ve been collecting. But it’s still been pretty complex and time-consuming, with data scientists using things like R, SAS, Python and toiling for a few weeks or even months to solve a few problems. We’re starting to run into a scaling issue, where companies have way more questions than answers.
The biggest evolution to happen over the past couple of years has been the transition from data science being a scary, expensive and technical burden to something that can be leveraged by anyone to drive data-driven decision-making. The idea of using data science to automate data analysis, to prescribe solutions … and do it in minutes. And I can tell you, one company is light-years ahead of the others in making this happen.
What excites you about the future of data science? What trends or innovations interest you?
The term that’s most exciting to me is “machine intelligence.”
This isn’t about chat bots or sentient robots. Machine intelligence refers to the new subfield of artificial intelligence that automates the discovery and explanation of answers from data. So, whereas machine learning requires the user to exhaustively train and tune algorithms to perform a task, machine intelligence actually teaches the user.
Machine intelligence will automatically evolve the best algorithms that explain a problem, and will interpret them into plain English to answer questions like why problem X is happening and how we can solve it, such as:
- “This jet engine is in jeopardy of failing, because of these five variables. Here’s what you should do about it.”
- “This collection of stocks and bonds is likely to significantly outperform the S&P 500, and here are the 10 predictive variables that give us such confidence.”
Machine intelligence, in the hands of many users, will be an incredibly valuable tool for society, not only for making people money, but also for addressing pressing issues in areas like healthcare.
Can you tell us about the mission behind Nutonian? How are you hoping to impact the way organizations create analytical models?
Nutonian’s goal is to enable companies to make the best data-driven decisions to maximize business or a desired outcome. Nutonian has two types of users.
The first is the tech-savvy person who likes to build models. We help him/her become exceptional at that, because we automate the heavy lifting of manual model creation, but we let the user guide the process and apply domain expertise to get the best possible models.
The second type of user is the business user. The analyst, the manager, the CxO. This person doesn’t need experience with data science or modeling. This person believes in the power of information and wants to make his or her business better, more efficient, more profitable. Eureqa’s forecasting app automatically creates predictions, and literally prescribes actionable recommendations, in a matter of minutes. This is where we change the game.
What are some of the common challenges facing organizations today in modeling? What are the reasons behind these challenges?
Modeling takes time and tech-savvy. Both of these things ding a company in the long run. You might have five brilliant people working to solve data-driven problems (what’s causing customer churn? Where should we open our next store? How do we make this metal stronger and lighter?), but it takes each of them a few weeks to complete their task.
The other issue is that models today are almost always too complex to reliably interpret and understand. If I want to know what’s causing customer churn, I might be trying to interpret a black box model from R that’s three pages long. Lack of model transparency means that even if I’ve got a highly accurate model, it’s hard to actually take actionable steps to improve my desired outcome, because the model is just too hard to comprehend.
What are the benefits of using machine intelligence to evaluate models and present easy-to-understand information from them?
Machine intelligence aims to not only optimize for accuracy, but also simplicity. Machine intelligence will identify the most accurate possible explanatory models in their simplest possible form. This ensures that companies can understand the causes underlying the problem and effectively act on them.
What are some outdated and/or ineffective methods for modeling you’re observing organizations using today? What are the risks to not adapting to new methods of understanding their data?
The truth is technology is constantly evolving. I would say if you’re using R, MATLAB, SAS, Python, you’re ahead of some people, but you’re not an early adopter, and you’re certainly going to start feeling the headache of cost, speed and complexity, if you haven’t already. And truth be told, your competitors are probably already starting to dip their toes into machine intelligence.
I’m biased, but I think machine intelligence is going to be the single greatest technological advantage for corporations over the next few decades.
What are some of the most interesting and/or exciting applications for machine intelligence you’ve observed being used in business?
I’m most excited by the impact machine intelligence is having on forecasting. Because machine intelligence is about understanding and actionability as much as accuracy, it’s enabling companies to not only predict, but also to diagnose and to mend. You can predict your future and essentially change it for the better.
We’re working with a global automotive company that’s projecting sales – and receiving automated recommendations on how they can maximize that number. Everything from hiring more people at specific locations, to stocking certain car types more than others, to spending ad money through the most effective channels. We’re also working with a large retail pharmacy to forecast prescription drug demand per store and per drug, enabling them to optimally stock and staff their stores.
These are not small problems, and they’re not easily solvable. Machine intelligence is going to change that.
Learn how Syncsort can help your organization use data to evolve. Check out their Big Data solutions.