This is what I think has shown great leadership by New York City to create this task force in recognition that the agencies in the city are using these black boxes to do prediction, to do decision-making, to do classification, and they are worried that decisions that are being made ...

One really what I view as quite foresightful is later on this afternoon I’m actually going to be going to the New York City Accountable Decisions Systems Task Force meeting.

People are as concerned about the computer itself being fed with biased data. The model will be biased as well. This is really a hot topic of research right now in the data science, machine learning, and AI community.

Now all of a sudden, these technologies are making important decisions about individuals and their future. People are quite interested in both the use of these technologies so that a particular policy can actually be applied uniformly and not subject to human opinion or bias.

There are many, many cases right now where these machine learning algorithms are being used in our social society. One of the canonical examples right now is in the criminal justice system. Some of these decision-making algorithms are being used as black boxes to actually determine what sentence someone should ...

The typical examples are, how large should a classroom be? Should it be 20 students? Is 30 students too many? We can actually study in a counterfactual way what causality, what we have, depending on the size of the classroom. That’s just one small example.

With the advent of big data, lots of data, machine learning algorithms, and causal inference, now we have the possibility to start really using data to understand a lot of policy decisions that could be made or are being made.

There’s a whole branch of data science that’s very interested in causal inference, which has of course been a study in economics and political science and social science for decades.

It is about amassing large amounts of data in order to make these vision systems perform well. There is where the issues of accountability, bias, and so on may come to play.

That’s just one example of medical, environmental applications of this technology that’s doing image processing. What I wanted to say is all these deep learning models are very data-hungry. The more data you feed these algorithms, presumably the better the classification, the better the prediction.

Rather than do that, not only can we do better but we can do it in a more environmentally sound way by having cameras on the fishing boats that will just be always on, taking pictures of the sea, if you will, and doing a better, more reliable count.

The way in which it used to be done is there’d be a human being standing there with a pencil and paper counting the fish, looking at the fish, probably scooping them up and doing some kind of statistical analysis to do an inference of how many fish are around.

Another area of application, speaking to sustainability and the environment, is I just learned just yesterday, literally, when you think about a lot of the fishing vessels that are out there, one of the roles that these fishing vessels have is to count the number of fish around. Are we ...

This is happening already. There are startups around the world that are bringing in these techniques to help hospitals, to help radiologists, do this kind of image processing. That’s in healthcare and that’s also just about images. I could say something very similar about other kinds of media, whether it’s ...

Those are the kinds of techniques. These techniques are serving multiple purposes outside of self-driving cars. For instance, if you take the healthcare domain and imagine using these techniques to process images, medical images -- for instance, X-rays, mammograms, images that are taken to detect whether you have cancer or ...

All of this object detection, object recognition is done through these AI machine learning algorithms. It’s astounding to me the success that these techniques have had in literally just the past five years.

The whole point is that for a self-driving car, these vision systems can see the objects in front of the car. It can detect whether a pedestrian is running across the street. It can detect whether there’s a stop sign coming up.

These deep learning techniques are building models such that, as you know, face recognition is basically a solved problem. Object recognition is to the point where all the self-driving cars have cameras that are serving as the vision system of these cars.

I would say just even five years ago, we were not nearly as far along as we are today in recognizing objects in an image or recognizing objects in a video. Currently the machine learning techniques that we use are...One of the most popular techniques is called deep learning.

Let me speak a little bit about the technology itself and the capability of the technology already in place. Probably the most exciting technology that all of you heard about and read about is AI and some of the machine learning algorithms that are producing models that can be used ...

I think it’s very important that we train data scientists from day one to understand the ethics of using data, collecting data, analyzing data, and disseminate the results.

By Data for Good, I mean using data to tackle societal grand challenges such as the UN sustainable development goals, but also to use data in a responsible manner, to use data considering privacy and ethical concerns, which I think everyone understands today is really in our face when you ...

Right now, I run the Data Science Institute. One of the things that I wanted to promote for Columbia University and for data science more generally is what I call Data for Good.

In that role, I oversaw most of the academic research in computer science for the country. I set the strategic directions and gave money out. People were very happy. It was a good position to be in when you’re giving money out.

Before that I was at Carnegie Mellon University for decades as a professor and department head. I also served at the National Science Foundation for three years as the assistant director for computer and information science and engineering.

Hi, I’m Jeannette Wing. I’m the Avanessians Director of the Data Science Institute at Columbia and also a professor of computer science. Before I joined Columbia last year, I was a corporate vice president at Microsoft Research and I ran all the basic research labs worldwide for Microsoft.