Why Algorithms Won't Take over the World (At Least, Not Yet)

Adrian Liang on October 10, 2018
Share

HannahFry_c_PeterBartlett.JPGThese days the term "algorithm" evokes the same magical potency as Harry Potter's famous invocation "Expecto Patronum!" When an algorithm is in force, you see only what you want to see on Facebook. Advertising for items you desire follows you around the Internet. Green signal lights stay on longer during rush hour to move more traffic through congested streets.

But an algorithm is only as good as the humans who create it. In Hello World (named for the program that is the first that people often learn to write), Hannah Fry explores algorithms that work, algorithms that don't, algorithms that have improved vital areas of our lives such as health care and policing, and algorithms that have screwed things up even worse.

It might sound dry, but Hannah Fry makes algorithms sound not only quite interesting but an idea that we must understand better as they dominate more and more of our daily lives in ways we see and in many ways we don't.

We spoke with Fry by phone about Hello World: Being Human in the Age of Algorithms, which we named in September as a Best Book of the Month in nonfiction.

Adrian Liang: "Algorithm" is a word that not everyone fully understands, and I admit that I didn't fully understand it before reading Hello World. Can you define it again for me?

Hannah Fry: Absolutely. It's actually such a broad term that it doesn't really convey that much information. Officially all an algorithm is is just a series of instructions. It takes some kind of input, follows a set of logical steps, and provides some kind of output. A cake recipe could be considered an algorithm. Your input is the ingredients, and the output is a cake, and the logical steps are the recipe itself. But people tend to use the word when they are describing something that happens inside a computer. There are very obvious examples of algorithms. Google search is one of them. Amazon's recommendation engine being another. Facebook news feeds… All of those are instructions given to a computer that filters through all of the information and provides you with what you're looking for in a much more sensible way. You can also find algorithms in slightly more hidden places. So we're now using them in a courtroom, in our schools, and in our hospitals. There's pretty much no aspect of modern life which isn't now, at least to some extent, controlled or aided by some kind of algorithm.

In Hello World you explore a lot of different areas that use algorithms every day, from medicine to art to crime. You give a lot of examples of how algorithms can more effectively reach their goals than humans can, whether it's finding suspicious spots on X-rays or predicting buying behavior. So where do humans have value as the use of effective algorithms expands?

Oh, my gosh. I think humans are so central to every part of this. It is definitely true that there are some things that we're not very good at. We get tired quite easily and we get bit sloppy as well. We make a lot of mistakes and we also like taking cognitive shortcuts. If there's an easy way out, we will take it often. We're not very consistent. Algorithms, by contrast, are beautifully consistent. They never get exhausted. They're not sloppy; they're much more precise than we are. But there are things that algorithms can't do that remain uniquely human skills. At the moment algorithms don't really understand context, they don't understand nuance, they don't have empathy. Whether it's something like recommending a movie or deciding how long someone should go to jail, those are things that ultimately have room for empathy and context and understanding of nuance. So I think that this is not ever a question about humans versus machines: "Which is better?" I think it's about what we can do when we're in partnership together.

You point out in a number of chapters that the goal of the algorithm itself should be questioned. For instance if an algorithm is being used in a health-care situation, is the goal to create a cost-effective treatment plan, is the goal to reduce risk for the hospital, or is the goal to go all-out in making the patient well again? How can the nontechnical person use their judgment when assessing whether or not algorithms are actually helpful for them?

Hello World

Yes, this is a really important point. Try to think who [the algorithm] is working for and what exactly it is trying to do. A good example of this came out in a news story recently in Boston about school timetables. There's lots of research that says that teenagers [have a different] circadian rhythm—they're not very good at waking up early—and that has a negative impact on their schooling if they're asked to start school really early. So with the best of intentions, a couple of ex-MIT students were asked to design an algorithm that would reschedule all the school start times in Boston and route the buses so that as many [teenagers] as possible could start school later.

There's a few different objectives that you can ask it to do. On the one hand, you want it to minimize the amount of money that's being spent on buses so that funding can be rerouted back into school projects. On the other hand, you want to make it so that as many older kids start as late as possible. And then on the other hand, you also want to make it so the families are happy with the results. You can blend those together and come up with some combinations of those different objectives. The slight problem with this algorithm was that it didn't take the families into account enough, so it came up with something that did indeed manage to save the state loads of money and did indeed managed to make it so that the older kids were starting later. But the flip side of that was some kids in elementary school were being asked to start school two hours earlier. When you're assessing how good an algorithm is, you can't just think of what the objectives are on the surface. You have to think about what the consequences are of it not being able to consider every possible objective simultaneously.

There's a fair amount of concern about how shifting so much over to algorithms and machine learning is going to change human employment. Some think it's going to cause a labor change greater than the Industrial Revolution and it's a change we're not prepared for. Others think we'll just weather it the way we have weathered other changes and in fact everything will be better later. What are your thoughts on about how algorithms and machine learning are going to change future human employment?

I think there will be a shift. The thing that's different about this compared to something like the Industrial Revolution is the time scales in which things are changing. The Industrial Revolution took a long time from beginning to end. There were several years—well, decades really—for people to adjust to the moving work force. Whereas some of the things we've seen with machine learning, we've seen time scales of months, maybe years, in which these explosions in technology have come through. Things are going to change quickly. As to whether this is going to have this dramatic, catastrophic shift on the workforce… I'm not convinced that's going to be the case, because I just don't that think we're anywhere near the stage when you can get rid of humans. All of these algorithms that we have at the moment are still fundamentally flawed in lots of different ways. They still make really big mistakes. And when those systems aren't designed to work in partnership with people—leaving them to operate on their own accord without human assistance—I think that you can have many catastrophic results. I am concerned about factory workers. I do think that there are a lot of changes in robotics that are going to make even the jobs in factories that aren't automated at the moment be pushed towards automation. But there was a lot of talk a couple of years ago about lawyers and doctors being put out of work because of AI…I don't see that happening just yet. I think we're quite a long way away from that.

Author photo (c) Peter Bartlett


Shop this article on Amazon.com


You might also like:

ABR-RR-SUBS-yellow[1]

Sign up for the Amazon Book Review: Best books of the month * author interviews * the reading life * and more


Lists + Reviews

Best Books Literature + Fiction Nonfiction Kids + Young Adult Mystery, Thriller + Suspense Science Fiction + Fantasy Comics + Graphic Novels Romance Eating + Drinking

Authors

Interviews Guest Essays Celebrity Picks

News + Features

News Features Awards

Editors

Omnivoracious, The Amazon Book Review

Feeds Facebook Twitter YouTube