Recent advances in the impact of deep learning are finally allowing computers to see, hear and read with increased precision. These long-awaited achievements—like those in robotics and automation--are being celebrated by computer scientists, but they also have implications for the economy and society. Jeremy Howard addressed both the technical and societal issues at a recent MIT IDE seminar.
Machine learning’s history goes back to 1956 when computers amazed the public because they could play against --and in 1962, beat--a human player at checkers. With continual breakthroughs in machine learning over the decades, Howard said that it’s now widely used and “creepily good” at finding your friends and beating humans at highly sophisticated intellectual pursuits such as Jeopardy.
Howard should know. He was ranked Number 1 in data science competitions globally in 2010 and 2011, and became President and Chief Scientist at Kaggle, where data scientists compete to answer problems using predictive analytics and machine learning algorithms.
Deep Learning = Good Medicine?
As deep learning--the evolution of neural networks and artificial intelligence--greatly improved, it has also incorporated more precise speech and object-recognition capabilities. To pursue these new fields, Howard left Kaggle in August to launch Enlitic.com; he now serves as CEO.
Enlitic uses recent advances in machine learning “to make medical diagnostics faster, more accurate and more accessible to medical staff and patients.” The company's mission is “to provide the tools that allow physicians to utilize the vast stores of medical data collected today, regardless of what form they are in--such as medical images, doctors’ notes and structured lab tests.”
This computing approach involves training systems, called artificial neural networks, on huge amounts of information derived from audio, images and other inputs, and then presenting the systems with new information and receiving inferences in response. Howard (foreground, in photo) wants to mine medical data in this way and provide easy-to-use applications for health care professionals.
For instance, “there are new ways to think about tumors” that borrow from Kaggle’s analytical and machine-learning methods, Howard said at the MIT IDE seminar. Assumptions can be made by machines based on pathology analysis that are sometimes more accurate—and objective--than human results. Ultimately, however,
Howard expects that the best diagnoses will result from a combination of humans and machines; where machines crunch the data and humans add history and human interaction to the mix.
Policy Discussions Needed
While the technological strides are apparent, Howard acknowledged the potential labor impact that will result from deep learning in the next decade as computers take on more jobs and displace workers. Drawing similar conclusions to those of MIT’s Erik Brynjolfsson and Andrew McAfee, as described in their latest book, the Second Machine Age, Howard said: “As human perception and judgment are replaced, and machine learning grows exponentially, policy discussions are needed about basic living wages and technological unemployment.”
According to Enlitic, each year since 2010, the Imagenet competition has been the proving ground for state-of-the-art computer vision algorithms. The past three years have seen striking improvements in accuracy.
In addition to his current role as CEO of Enlitic and his former role with Kaggle, Jeremy Howard is on the faculty at Singularity University. He was recently a Distinguished Research Scientist at the University of San Francisco, and he advises Khosla Ventures as their data strategist. Howard founded two successful start-ups – the e-mail provider FastMail and the insurance pricing algorithm company, Optimal Decisions Group – both of which grew internationally and were sold to large international companies. He started his career in management consulting, including at McKinsey & Co. and AT Kearney, where he built a new global practice in what is now called “Big Data”.
Follow his blog here.