Anthony Goldbloom is a world-renowned expert in using big data to tackle big tasks. His company, Kaggle, uses predictive modeling to solve complex problems for NASA, automakers, insurance companies, and drug and medical device manufacturers, among others. No problem is too big for Kaggle's team of scientists to resist tackling. Goldbloom, named by MIT as one of the top 35 innovators in the world, believes that "machine learning" is the most powerful branch of AI and will be responsible for much of the disruption we can expect to see in the workplace. Machine learning is the technology that allows machines to learn from data and, in some cases, mimic things humans can do. Kaggle operates on the cutting edge of machine learning, which gives Goldbloom a unique perspective on what machines can do and what they can't.
"By construction, machines are very good at learning things that have been done before and repeating them again and again and again," Goldbloom told me. "But in order to touch somebody emotionally, you have to surprise people. Machines have made very little progress in tackling novel situations. They can't handle things they haven't seen many times before."
One project undertaken by the Kaggle community of data scientists demonstrated the promise—and limitations—of machine learning. In a competition co-sponsored by the Hewlett Foundation, scientists were invited to create software that could grade a student-written essay as well as or better than a human evaluator. It's an important area of software development. The Hewlett Foundation is a philanthropy dedicated to providing America's public school students with the skills they need to excel in the twenty-first century. Chief among those skills are critical thinking and effective communication. One way to improve the quality of instruction in those areas is to move from multiple-choice tests to essays, which require higher-level thinking and writing skills. However, grading essays by hand is expensive and time consuming. That's why Hewlett challenged the Kaggle community to take their best shot at automating the task of scoring essays.
The results of the Kaggle/Hewlett challenge were promising. The winning software evaluated 22,000 hand-scored essays. By analyzing sentence structure, spelling, and punctuation, the software did a reasonably good job of replicating scores given by human graders, especially for the average essay. But according to Goldbloom, the algorithm came up short in one crucial area. It failed to recognize essays that were above average—unusual, novel, and groundbreaking. In fact, creative essays received lower grades than they actually deserved! Machines learn from crunching large amounts of existing data while we humans use our imaginations to propose and communicate novel ideas that, by definition, have never been seen before. "We can connect seemingly disparate threads to solve problems," says Goldbloom. "This puts a fundamental limit on the human tasks that machines will automate." If a computer can recognize average, it can replicate average. Average simply isn't good enough to stand out in the digital age.
In another essay experiment, this time in Japan, mathematician Noriko Arai sparked a wave of anxiety when she built an AI system that outperformed 80 percent of high school students in a competitive college entrance exam. The "Todai Robot" scored in the top 1 percent in the math and science section of the exam and could write a 600-word essay better than most students. Despite these results, Arai is convinced that humans can thrive in an AI-saturated world if they reconsider the types of skills they must master. In Arai's experiment, she discovered that AI did better than 80 percent of students because it could retrieve facts more quickly and accurately, which is what most students are taught to do—memorize facts and repeat them. But Todai failed to beat the 20 percent of students who stood out because they could think creatively and extrapolate meaning "beyond the bounds of a question." In other words, AI doesn't read or think as humans do. The Todai Robot recognizes keywords and combines text and facts retrieved from existing information to provide an answer to a question. According to Arai, if "knowledge" means memorizing and retrieving facts, then AI can do what humans do, only better. She says the skills that give humans an edge are those that no robot or machine can currently replace: critical thinking, creativity, and communication.
Machines are fast; humans are creative. Machines glean insights from data; humans shed light on what the data means. Machines teach us about the past; humans build the future. Machines make us more productive; humans improve the world in imaginative, unexpected ways. Lin-Manuel Miranda didn't win a MacArthur Foundation Genius Award for writing a song faster than a computer. He won for igniting and inspiring the human spirit.
A TECTONIC SHIFT IS HERE
Emotional connection is, indeed, the winning ticket in a world where technologies such as automation, big data, artificial intelligence, and machine learning are eliminating millions of jobs and disrupting entire industries, businesses, and careers. People around the world are understandably anxious about the pace of change and what it could mean for the future of work. The good news is that fears of a "robot apocalypse" might be overblown, at least according to the evidence we have from the last 500 years.