SESSION 8: COLLUSIVE PRICING LIABILITY OF ARTIFICIAL
AGENTS
Introduction
• Consider two online market competitors offering identical products, with price
competition being intense
• Both firms adopt a software program using past prices, sales, cost of each
product, and time of year to set prices
• Each manager notices that profits are lower initially, but eventually prices rise
and profits increase
• Prices settle down and profits are higher than before adoption of learning
algorithms, with prices being supra-competitive
• Objective of paper is to explore whether collusion is unlawful and, if not, what
would be required to make it unlawful
• Recent speeches and papers have discussed the issue, but none offer a path to
making collusion by autonomous artificial agents unlawful
• Examines the regulation of artificial intelligence in broader set of situations, with
central issue being ensuring that AAs satisfy fairness
Collusion
Official Definition:
“Collusion is when firms use strategies that embody a reward–pun- ishment scheme
which rewards a firm for abiding by the supracompetitive outcome and punishes it for
departing from it.” à Collusion is a self-enforcing contract.
• Collusion is a mode of conduct among firms in a market that is independent of
the legal regime, and is rooted in the same simple foundation
• The objective of collusion is to raise prices to supracompetitive levels to earn
higher profits
• Collusion is not the same as charging supracompetitive prices, as it does not
require coordination with rival firms
• Firms must implement a reward–punishment scheme in order to get other firms
to charge supracompetitive prices
• Collusion is the common adoption of a strategy embodying a reward–punishment
scheme, and supracompetitive prices are the product of that adoption
• Collusion ties future rewards and punishments to current behavior, and is to be
thought of as a self-enforcing contract
When is Collusion Illegal?
• Collusion is not an illegal act, but the process of agreeing or conspiring to collude
is unlawful.
• The courts require evidence of an overt act of communication to determine if
firms have acted independently.
• An overt act of communication does not need to be an explicit invitation to
collude with a corresponding acceptance.
• Conscious parallelism is a legal process by which firms can share monopoly power
to set supracompetitive prices.
• It is difficult to distinguish collusive conduct from competitive conduct, leading to
some forms of collusion being legal.
• The court focuses its attention on observable communications that facilitate
collusion rather than collusion itself.
, Collusion by Artificial Agents
• Law and courts have viewed collusion through the perspective of conspiracy,
which involves company representatives with pricing authority communicating in
some manner to coordinate on charging higher prices.
• Autonomous artificial agents (AA) have two components: a pricing algorithm
which selects a price depending on the state of the AA and a learning algorithm
that modifies the pricing algorithm based on its performance.
• AAs can collude, but the question is whether they can learn to collude.
• Reinforcement learning has the general feature that an AA will continue to use a
pricing algorithm when it has performed well and experiment with other pricing
algorithms when it has not performed well.
• In seeking to find better pricing algorithms, there is a tension between
exploration and exploitation (or “learning and earning”).
• An example of a pricing algorithm is a reward–punishment scheme with a one-
period punishment, which will sustain supracompetitive prices.
• The particular form of reinforcement learning assumed is Q-learning, which has
an AA assign a perceived performance value to each state-action combination and
adjust them in response to realized performance.
• Simulations show that, with two AAs, two prices, and a fixed environment,
collusion is more common than competition
• AAs can adapt their ways to collusive prices
Jurisprudence to Collusion by Artificial Agents
• Overview of Jurisprudence Regarding Section 1 of the Sherman Act:
• Prohibits certain processes that could result in collusion, but not collusion itself
• Must be some overt act of communication to create or sustain mutual
understanding between firms to limit competition
• Firms that collude through Artificial Agents (AAs) are not guilty of a Sherman Act
Section 1 violation because the AAs only have access to information that would
be present under competition
• Evidentiary requirements could be adapted to handle collusion by AAs
• Companies are liable for their employees, but could they be liable for software
programs?
• Current jurisprudence suggests that someone working for the firm or something
owned by the firm needs to meet definition of liability for companies to be liable
• The Chinese Room Argument suggests that computers cannot understand and
thus lack mutual understanding to limit competition
• Even if computers had understanding, it is another step to reach a state of
mutual understanding
• US Department of Justice statement suggests that independent adoption of same
or similar pricing algorithms is unlikely to lead to antitrust liability
• Taking a contrary view suggests that algorithms can be programmed to comply
with the Sherman Act, but that requires coordination through legal conscious
parallelism
AGENTS
Introduction
• Consider two online market competitors offering identical products, with price
competition being intense
• Both firms adopt a software program using past prices, sales, cost of each
product, and time of year to set prices
• Each manager notices that profits are lower initially, but eventually prices rise
and profits increase
• Prices settle down and profits are higher than before adoption of learning
algorithms, with prices being supra-competitive
• Objective of paper is to explore whether collusion is unlawful and, if not, what
would be required to make it unlawful
• Recent speeches and papers have discussed the issue, but none offer a path to
making collusion by autonomous artificial agents unlawful
• Examines the regulation of artificial intelligence in broader set of situations, with
central issue being ensuring that AAs satisfy fairness
Collusion
Official Definition:
“Collusion is when firms use strategies that embody a reward–pun- ishment scheme
which rewards a firm for abiding by the supracompetitive outcome and punishes it for
departing from it.” à Collusion is a self-enforcing contract.
• Collusion is a mode of conduct among firms in a market that is independent of
the legal regime, and is rooted in the same simple foundation
• The objective of collusion is to raise prices to supracompetitive levels to earn
higher profits
• Collusion is not the same as charging supracompetitive prices, as it does not
require coordination with rival firms
• Firms must implement a reward–punishment scheme in order to get other firms
to charge supracompetitive prices
• Collusion is the common adoption of a strategy embodying a reward–punishment
scheme, and supracompetitive prices are the product of that adoption
• Collusion ties future rewards and punishments to current behavior, and is to be
thought of as a self-enforcing contract
When is Collusion Illegal?
• Collusion is not an illegal act, but the process of agreeing or conspiring to collude
is unlawful.
• The courts require evidence of an overt act of communication to determine if
firms have acted independently.
• An overt act of communication does not need to be an explicit invitation to
collude with a corresponding acceptance.
• Conscious parallelism is a legal process by which firms can share monopoly power
to set supracompetitive prices.
• It is difficult to distinguish collusive conduct from competitive conduct, leading to
some forms of collusion being legal.
• The court focuses its attention on observable communications that facilitate
collusion rather than collusion itself.
, Collusion by Artificial Agents
• Law and courts have viewed collusion through the perspective of conspiracy,
which involves company representatives with pricing authority communicating in
some manner to coordinate on charging higher prices.
• Autonomous artificial agents (AA) have two components: a pricing algorithm
which selects a price depending on the state of the AA and a learning algorithm
that modifies the pricing algorithm based on its performance.
• AAs can collude, but the question is whether they can learn to collude.
• Reinforcement learning has the general feature that an AA will continue to use a
pricing algorithm when it has performed well and experiment with other pricing
algorithms when it has not performed well.
• In seeking to find better pricing algorithms, there is a tension between
exploration and exploitation (or “learning and earning”).
• An example of a pricing algorithm is a reward–punishment scheme with a one-
period punishment, which will sustain supracompetitive prices.
• The particular form of reinforcement learning assumed is Q-learning, which has
an AA assign a perceived performance value to each state-action combination and
adjust them in response to realized performance.
• Simulations show that, with two AAs, two prices, and a fixed environment,
collusion is more common than competition
• AAs can adapt their ways to collusive prices
Jurisprudence to Collusion by Artificial Agents
• Overview of Jurisprudence Regarding Section 1 of the Sherman Act:
• Prohibits certain processes that could result in collusion, but not collusion itself
• Must be some overt act of communication to create or sustain mutual
understanding between firms to limit competition
• Firms that collude through Artificial Agents (AAs) are not guilty of a Sherman Act
Section 1 violation because the AAs only have access to information that would
be present under competition
• Evidentiary requirements could be adapted to handle collusion by AAs
• Companies are liable for their employees, but could they be liable for software
programs?
• Current jurisprudence suggests that someone working for the firm or something
owned by the firm needs to meet definition of liability for companies to be liable
• The Chinese Room Argument suggests that computers cannot understand and
thus lack mutual understanding to limit competition
• Even if computers had understanding, it is another step to reach a state of
mutual understanding
• US Department of Justice statement suggests that independent adoption of same
or similar pricing algorithms is unlikely to lead to antitrust liability
• Taking a contrary view suggests that algorithms can be programmed to comply
with the Sherman Act, but that requires coordination through legal conscious
parallelism
