The Internet and artificial intelligence are changing the way we live. That’s not a surprise. We fret about how distracted we’ve become, or about whether robots will take our jobs. Those are real concerns. But the impact of computing technology could be even bigger and deeper – forcing us to reckon with our place in the world and our ability to understand it.
“I certainly do not want to be in a position of denying science – that’s an easy starting point - or the sorts of laws that science has discovered over the centuries,” Weinberger said. “I’m not about to pick an argument with Newton.”
But Weinberger says that artificial intelligence should make us stop and think about – even question – the fundamental idea that humans are capable of understanding how the universe works, and of using that data in a predictive way.
“The new thing about machine learning is that you feed it in the data. You don’t tell it how you, the human, think this data should go together – what the general principles and rules and laws are,” Weinberger explained. “Instead, it just iterates on this massive amount of data and it finds correlations, and those correlations can be extremely detailed.”
In many cases, Weinberger says, the relationships that machine learning infers are so complex that we cannot understand how they work. That is often the source of skepticism or fear, but Weinberger says we should be taking advantage of it.
For example, human doctors can understand the relationship between cholesterol and heart disease, and use that to predict when a patient is at risk. But machine learning systems can be fed hundreds of thousands of patient records with hundreds of data points for each patient and – without any knowledge or understanding of how diseases work – diagnose patients at risk for a range of conditions.
“Some of these systems can predict the onset of diseases more accurately than humans can and, in some instances, can predict the onset of diseases that humans simply cannot predict,” Weinberger said.
Sometimes, the associations that machine learning algorithms turn up can lead to a deeper understanding of a disease or other complex system. But, in other cases, the best efforts of scientists have failed to turn up an explanation, even when the diagnosis or prediction is correct.
Of course, machine learning is not completely independent of human influence. Humans are the ones generating the data from which machines learn.
“We are placing the sensors around the world that are gathering the data that matters to us for one reason or another,” Weinberger said. “Information is not a natural artifact. It’s something that we create.”
The choice of what data gets collected, and how, can have a huge effect on the ultimate outcome of machine learning. For example, health data collected from one population can lead to misdiagnoses of people from different genetic backgrounds – a real problem that researchers are working to fix right now.
Weinberger fully acknowledges the potential pitfalls of letting algorithms make decisions for us, but he urges us not to throw the baby out with the bathwater.
“If it helps us to see more of the world that we’ve written off as accidental, or because we didn’t see exactly how it fits, or the data is beyond our capacity,” Weinberg said. “If we start to see the world more for what it is, then I think we are at really important and exciting inflection point in our history.”
And it’s not just the world around us that artificial intelligence could help us understand. Weinberger says it might just change how we understand ourselves.
“Our tools have tended to shape our understanding of ourselves and especially our minds,” Weinberger said. “Assume that machine learning is going to be a dominant technology. How might that change how we think about who we are and what our position in the world is?”
Weinberger says he doesn’t have an answer, but that the question needs attention.
David Weinberger’s new book is “Everyday Chaos: Technology, Complexity, and How We’re Thriving in a New World of Possibility.”