The Rise of Artificial Intelligence

Topics in Economics, ESCP, 2023-2024

Author

Pablo Winant

Published

January 15, 2024

Introduction

Will only appear in HTML.

How do you see the future of AI ?

. . .

Science Fiction has explored many issues associated with AI.
- what happens to individuals, markets, firms, government?
- who wins, who looses?
Very often economic future is bleak…
Why is that so?

What is AI?

What is AI

AI can be:
- a field
- a set of technologies including
  - machine learning
  - deep learning
  - reinforcement learning
- a philosophical concept
Usual definition of AI:
- mimic human intelligence:
  - problem solving: achieve goal without being explicitly told how
  - learning: gets better over time
- without emotions

How to make the difference between a robot and a ?

Check whether it thinks like a human being
Turing test:
- a machine discussing with a human must be recognized as a 7 years old human 50% of the time
- can machines emulate human behaviour?
Chinese Room thought experiment:
- how to differentiate somebody who speaks chinese and somebody who follows rules to speak?
- turing test cannot determine whether machines think
What do you think of these tests?

Impress me?

So, what can machines do that is really intelligent?
- speech recognition, language translation
- drive a car
- play chess better than a human
- play go, better than a human, without being a specialist, by learning on its own
- create, imagine
- have empathy
- talk
- reason
- … create movies

The Classical View

“This Time it’s Different” or “Same old, same old…”?

David Ricardo Adam Smith

. . .

Do you remember the neoclassical production function?

The (neo)classical production function

What are its main properties?

production takes several factors as inputs
- capital
- labour
- … (natural resources, land, …)
each factor has a market price
marginal returns w.r.t. each factor are decreasing
factors are paid according to their marginal productivity
the technology is the particular process through which inputs are combined

AI and the (neo)classical production function

the precise description depends on the problem under consideration
what could you change to take into account the effect of AI?
- data, technological change ?

Three hypotheses about the economic nature of AI

A technological change
A new kind of factor: Data
Yet another kind of factor: Robots
Something else Completely

AI is a change in the cost structure

Ajy Agrawal, Joshua Gans and Avi Goldfarb: Prediction Machines: The Simple Economics of Artificial Intelligence 2018

Prediction Machines

many production tasks can be formulated as prediction problems
exemples:
- regression, classification: predict Y as a function of X
- student -> pass or fail ?
- should I invest in A or B ?
- even chatbot: what is the appropriate continuation for an ongoing conversation?

Will I loose my job ?

AI is a decrease in the cost of predictions
The demand for all prediction-intensive tasks will rise (law of demand)
The salary of workers with prediction-intensive tasks will rise (market price)
Value of other tasks will fall (general equilibrium effect)
More precisely:
- demand for tasks that are substitute to predictions will be low
- demand for tasks that are complement to predictions will be high

AI is Data

Chad Jones and Christopher Tonetti (Stanford) Nonrivalry and the Economics of Data (Sep 2020, American Economic Review)

Data is a factor not a technology

Data is a factor, not a technology
- Can you explain it ?
The difference between an idea and a factor? Exemples:
- idea: use machine learning to build self driving cars
- factor: each car-maker gathering his own data to train cars
Data (even anonymous) improves quality of existing products

What kind of good is data ?

Remember the classification of goods?
- nonrival: can be used with leftovers
- excludable: use can be limited to paying customers
- data is a: club good
Nonrivality implies increasing returns to scale
marginal value of new data increases more than proportionnally

AI: adds data to the production function (consequences)

increasing returns to scale implies natural monopoly
- ->GAFAMs
- increasing suboptimal monopoly rents (already a problem before existence of AI…)
- should you regulate a monopoly?
  - it depends what is the barrier to entry: data-gathering or data-processing
other relevant questions
- where are the markets? (empirically it seems “undertraded”)
- who owns the data ? Consumer, producer.

How do you regulate a Data-monopoly ?

solutions:
- split the monopolies (if deadweight loss is too big)
- outlaw data gathering (big productivity loss)
- force data-sharing: make it a public good
- let the consumer be free to decide whether to rent his data (remove externalitiess)

AI: competition between humans and robots

Economic singularity

In the very long run, could technology be bad?
Recall the neoclassical world
- market economy
- technological progress reduces production cost
- always good for consumers. Increase (real) total income.
- becomes an inequality problem
But
- whether technology reduces salaries depends on whether growth is labour augmenting or capital augmenting
- if AI is a close enough substitute, salaries of “humans” as a whole are at risk
- the is an economic singularity when salary of humans arrives below the subsistance level

Anton Korinek Joe Stiglitz Korinek Gilles Saint Paul

Two sets of authors reach very similar conclusions
- Anton Korinek and Joseph Stiglitz (left): more complete/technical
- Gilles Saint Paul (right): more political economy

Some very long run scenarios

Analysis taken from Gilles Saint Paul
Main hypothesis: all humans can be replaced by more productive robots
Comparative advantage logic:
- humans specialize in work where their comparative disadvantage is lowest (services, art, crafting…)

Scenario 1: society redistributes income from robots

Four political subscenarios:

Welfare state

robot-owners are taxed, income is redistributed
for instance as universal income
some productivity losses
what about international competitivity?

Rentiers society

robot owners invest the rent over many generations
capital concentration increases

Neo fordism

firms pay huge salaries for essentially useless jobs (powerpoint presentations, 😉 …)
useful to sustain demand

New roman empire

robot owners: patricians (top 2%)
rest of population: plebeians
- survive thanks to clientelism
robots: slaves

Scenario 2: wars, starvation, epidemic

human income (marginal productivity) falls below subsistance levels
malthusian effect: population growth decreases
not unheard of (Leontieff): consider population of draft horses

Scenario 3: the Matrix

human wage decrease
subsistance level decrease dramatically too

Something Else Completely?

Right now AI is a technology (or a factor)
What if it becomes another intelligent agent?
- has its own goals
- its own preferences
- with superhuman thinking abilities…
Response in the literature (if curious):
- Anton Korinek: if market economy survives
  - malthusian and non-malthusian scenarios
- At that stage humans might be something different completely
  - transhumanism

Conclusion

Research on AI is very speculative: especially about the long run
But concepts from classical economics still help
For next time:
- make sure you understand all concepts in bold