The consequences of AI-based technologies for jobs

• Author: Peter Cappelli, University of Pennsylvania
• Pages: 652-675

652

Recent discussion f‌irst in the business press and

then in related public policy communities has

considered the notion that industrial countries

are on the verge of important changes that

stem from information technology (IT) broadly,

including notions of artif‌icial intelligence (AI),

and its implications for how work is performed.

The size and pervasiveness of these discussions

merits a serious look at the ideas behind them

and the fundamental question they ask: is

there something happening already or about

to happen in information technology that will

change in a fundamental way businesses and

organisations, jobs, and outcomes like pay and

unemployment? I consider these issues below.

1. The nature of the discussion

Before considering the arguments and

assertions about the implications of evolving

IT, it is worth thinking through the context in

which those stories take place. Followers of the

media are well aware that there is a bias toward

reporting stories that represent something new,

especially something new and dramatic. That

includes claims about developments that will

Summary

This contribution follows the recent public

debate on the changes across industrial

countries that stem from information

technology, including notions of artif‌icial

intelligence and its implications for how work

is performed. While acknowledging the size

and pervasiveness of these discussions, the

article discusses the core arguments related

to the impact of information technology on the

way businesses and organisations operate,

how these changes could translate to the

labour market, and other potential outcomes

such as lower wages or unemployment.

The argument begins with an introduction

to the two ways in which people tend

to anticipate future developments. This

either happens through estimates based

on prior experience (commonly known as

forecasting) or through a belief in a real

uncertainty of future developments and

reliance on other kinds of evidence besides

traditional forecasts. The article maps the

projected impact of technological uptake

on the labour markets and reviews the

empirical evidence. It touches upon many

of the above-discussed trends, such as

skill-biased technological change or routine-

biased technological change, and their

implications for skills demand. Applying an

historic perspective, the article argues that

predictions based on the past may be less

relevant in the current context. Although

new equipment and practices could

eliminate certain jobs, on balance they do

not necessarily destroy jobs because their

overall ef‌fects on improving productivity and

overall wealth create jobs elsewhere.

To understand why assumptions claiming

that the future is like the past are not correct

and extrapolations from prior experiences

are unlikely to be accurate predictors of the

future, read this chapter.

653

CHAPTER 11

happen, even if there is little or no evidence of

them yet. We may notice these stories especially

when they relate to health, e.g. epidemiological

studies showing that some particular food group

is associated with either remarkably better or

worse life outcomes. It is extremely dif‌f‌icult to

run a story that says, for example, ‘still nothing

new in ef‌fective weight loss’. A f‌irst question

to ask is whether the apparent magnitude of

the stories of technological change ref‌lects

a change in the nature of the media and public

discourse rather than ref‌lecting something

about the merits of the arguments themselves.

There have been changes in the media that might

help create the impression that particular stories

are more important than would have been the

case in the past, such as the fact that there are

now many more outlets for stories, including

social media, where surprising or frightening

accounts are repeated and reinforced over and

over. There is also considerable expansion of

organisations focused on public policy, especially

those businesses which advocate ideas that

are important and support those that attract

attention. Hosting discussions, producing reports,

commenting on media stories are standard

practices for such organisations. Every major

consulting company now produces reports and

markets their views on policy-related stories,

including technology and workplace topics.

The fact that there is a great deal of discussion

about IT certainly suggests that it is a topic

worth investigating, although it is not prima

facie evidence that the arguments which

provoke that discussion are correct. The truth

is typically more boring than the speculations.

2. Anticipating the future

Assessing the merits of arguments about

the potential ef‌fects of IT in the workplace or

elsewhere should begin with thoughts about

epistemology: what is it that we know, and

how can we know it? Specif‌ically, how can

we distinguish reasonable belief from mere

opinion? What constitutes knowledge is always

a pertinent question, but it is especially important

in this context because of the unique nature of

the claims being made. They are claims about

the future rather than the present, although

they may well be informed by the present.

There are at least two quite dif‌ferent types of

claims about the future that are made in the

social sciences. The f‌irst concerns probabilities

and risk: we have very little idea about, for

example, whether my house will burn down but,

based on prior experience of houses like mine,

we can estimate with considerable accuracy

what the odds of that are.

Forecasts move us from predictions about

common events and about individual units in

a population to anticipating events that have

not happened before. They go a step further

than identifying average experiences in the

past to extrapolate from the past. To predict,

for instance, the unemployment rate in a year’s

time, they look back to previous unemployment

rates and to variables that determined them

or at least were associated with them. If

the model using those variables explained

a reasonable amount of the variation in

previous unemployment rates then we will

try to use it to extrapolate into the future.

We do so by assuming that the structure of

the model remains the same going forward

or, in practical terms, that the coef‌f‌icients of

regression-related models in the future will be

the same as they are in the model. Assuming

we have more recent values for the variables

in the model, we apply them to that model and