Artificial intelligence (AI) has the potential to turn our lives upside down.

And it can also provide answers to questions that have so far remained inaccessible to us.

In economics, for example, researchers use them to find out how immigrants become economic climbers or how climate change and economic growth are changing remote regions of the world.

Susan Athey, a pioneer in the use of the new technology, wrote back in 2019 that machine learning - the core of AI - will have a "dramatic impact" on economics within a short period of time.

The conclusions that she and her colleagues draw based on the new methods, in turn, change our understanding of the world.

Alexander Wulfers

Editor in the business department.

  • Follow I follow

Machine learning can help both in compiling data and in analyzing it.

In the former case, it replaces what previously had to be done by hand by a large crew of research assistants.

And in a fraction of the time.

This makes much larger data sets usable, which, for example, the economic historian James Feigenbaum from Boston University knows how to use in his research.

It uses historical census data to track people's careers over time.

The data contain a wealth of information on education, family relationships and the economic situation.

By linking individual data sets from one census to the next, Feigenbaum and his colleagues were able to show, among other things,

The difficulty is that this data is usually full of errors.

Tracking the same person from decade to decade isn't that easy.

The German immigrant Schmidt became a Smith within ten years.

Or a transposed digit creeps in with the date of birth.

In order to create a consistent data set without the help of computers, tedious manual work and a large number of gut decisions would be necessary.

Feigenbaum trained an algorithm by linking people by hand for a small sample and thus teaching the software what common alternative spellings are, for example, and how much tolerance for deviations is still acceptable.

This algorithm could then be fed with the data of millions of people and found the right pairs with a relatively high degree of precision.

Insights from remote areas

Sometimes, however, economists have to start a step earlier in empirical research.

In the past, if the data wasn't neatly laid out in computer spreadsheets, but only printed out on paper, someone had to laboriously type it out by hand.

This is not only very expensive, but error-prone.

The economist Melissa Dell has developed a program that, with the help of so-called deep learning, can not only recognize numbers in scanned tables, but also digitize their layout correctly.