Algorithms that tell us who to date and what to read are now fully embedded in daily life. It’s not a stretch to imagine algorithms one day deciding our career path, who we marry and who we elect to office. That’s the premise of Yuval Noah Harari’s new book, Homo Deus, and it inspired Daniel First, a data science master’s student at Columbia University, to respond. In an essay recently published in the journal AI & Society, Will big data algorithms dismantle the foundations of liberalism?, First challenges the idea that algorithms will seriously replace human free will. He recently took a break from his summer internship at McKinsey to talk about philosophy, the limits of artificial intelligence, and his own circuitous path to data science.
Why did you write this piece?
I was interested in whether Big Data algorithms could answer the first question of philosophy – how to live a good life. Every culture has its own way of answering this question – religious texts, nationalism, introspection, careerism, and now, data. Making choices has become infinitely more difficult in our complex, rapidly-changing world. Computers are in many ways smarter than us, and I wanted to push the idea to its limits. Can artificial intelligence make our tough decisions for us?
What message do you hope readers come away with?
A.I. will never replace our need to discover for ourselves how to live. A world in which users make life decisions based on advice from algorithms would be dystopic. Algorithms are ultimately designed by companies with profit in mind, not the well-being of users. Facebook “personalizes” your news feed so you’ll spend more time on the site — not to make you a better person.
But algorithms are based on math. Doesn’t that make them objective?
In one sense yes, but they are designed to solve problems, and bias creeps in depending on how you frame the problem. A hockey league asked a friend of mine to build an algorithm to optimize their game schedule. But the league wasn’t clear about its objective. Should the schedule be optimal for the players to minimize travel time, for the league to maximize revenue, or for fans to maximize their enjoyment?
What are some of the technical barriers to creating algorithms that make life decisions?
Prediction algorithms work well for short-term predictions about stable entities in a relatively unchanging environment – such as predicting whether a customer will cancel their Wells Fargo account over the next year. Predicting whether you will be happy as a doctor a decade from now is a different story. People change with time and so does society. Our life decisions also depend heavily on our family and friends, community and other social factors. These considerations are hard to build into generalizable models that can learn from Big Data.
You picked up several degrees before coming to Columbia. Why psychology and later, philosophy?
I grew up in a traditional Orthodox Jewish community. For the first twenty years of my life, I aspired to be a rabbi and Talmud scholar. Fascinated by Maimonides and his synthesis of tradition and science, I became interested in the scientific basis of desire and happiness. That made me wonder: Can science help us lead a more fulfilling life? I transferred from the Talmud study hall to Yale and worked in a neuroscience lab studying desire and self-control in cocaine addicts, examining how science can help us act in line with our ideals. I later studied for a master’s at Cambridge to examine these questions through the lens of philosophy, focusing on Nietzsche.
So how did you end up in data science?
I thought that neuroscience or psychology could show us how to live – by mapping people’s brains or running enough lab experiments. That turned out to be a flop. The traditional methodologies of psychology are unable to take into account much of the richness of human experience. I actually think data science can do better. Psychologists classically might run a thirty-minute study in a lab, while data scientists can derive insight from hundreds of hours of real-life video recordings. As we move towards a world where more and more of our lives are recorded, the potential of Big Data algorithms to help us understand ourselves soars. It’s a great time to think about the possibilities – and limitations – of A.I.
Advice for others looking to switch from humanities to data science?
It’s easier than you think. You can learn the programming you need to get started in a week – basic Python, numpy, pandas. The professors at Columbia have been excellent at ensuring everyone is up to speed when advanced mathematical concepts come up. Much of data science involves breaking down the problem logically and creatively, communicating results clearly, and persuading teams to adopt your solution – so training in the humanities actually helps a lot.
What did you learn in school that’s helped in your internship at McKinsey?
In class we were asked to predict specific variables like rent for a block of apartments or what rating a user would give a movie on Netflix. In consulting, the problem may have dozens of variables and your job is to figure out which one is the most relevant. You might be asked to analyze tens of thousands of employee records, for example, to predict what makes a team effective. How do you measure that? Andreas Mueller’s Applied Machine Learning helped us learn how to quickly assess the trade-offs of various machine learning approaches and pick the best one.
— Kim Martineau
Source viva: http://datascience.columbia.edu/daniel-first-data-science-life-decisions