Artificial Intelligence: Can Seemingly Collusive Outcomes Be Avoided?

Abada, Ibrahim; Lambin, Xavier

doi:10.2139/ssrn.3559308

Cited by 15 publications

(13 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition to regulating the input to the algorithm, various market design features may also prevent autonomous collusion, without impeding efficiency benefits. These could, for instance, relate to demand-steering policies (Johnson, Rhodes, and Wildenbeest, 2020), or involve forcing a disaggregation of decision-makers or introducing an additional algorithm that aims to maximize social or consumer welfare (Abada and Lambin, 2020). It would be valuable to explore such options further.…”

Section: Discussionmentioning

confidence: 99%

Autonomous algorithmic collusion: Q‐learning under sequential pricing

Klein

2021

The RAND J of Economics

109

View full text Add to dashboard Cite

Prices are increasingly set by algorithms. One concern is that intelligent algorithms may learn to collude on higher prices even in the absence of the kind of coordination necessary to establish an antitrust infringement. However, exactly how this may happen is an open question. I show how in simulated sequential competition, competing reinforcement learning algorithms can indeed learn to converge to collusive equilibria when the set of discrete prices is limited. When this set increases, the algorithm considered increasingly converges to supra-competitive asymmetric cycles. I show that results are robust to various extensions and discuss practical limitations and policy implications. "It's true that the idea of automated systems getting together and reaching a meeting of minds is still science fiction. (…) But we do need to keep a close eye on how algorithms are developing. (…) So that when science fiction becomes reality, we're ready to deal with it." -EU Competition Commissioner Margrethe Vestager (2017)

show abstract

Section: Discussionmentioning

confidence: 99%

Autonomous algorithmic collusion: Q‐learning under sequential pricing

Klein

2021

The RAND J of Economics

109

View full text Add to dashboard Cite

show abstract

“…The analysis of humancomputer interaction is important because most markets in the field are heterogeneous and firms cannot be sure of whether their opponents are using algorithms for their pricing decision, nor do they know which type of algorithm competitors might use. 39 A recent and growing literature (Abada and Lambin, 2020;Calvano et al, 2021Calvano et al, , 2020bKlein, 2020;Waltman and Kaymak, 2008) shows that markets with exclusively firms using algorithmic pricing can become collusive. This raises the question of whether algorithms also have collusive impact when they interact with humans.…”

Section: Discussionmentioning

confidence: 99%

“…In contrast to explicit agreements, they may leave no trace of collusive intent behind. Recently, it has been demonstrated that new self-learning algorithms manage to collude systematically and in a strikingly efficient manner (Abada and Lambin, 2020;Calvano et al, 2020bCalvano et al, , 2021Klein, 2020;Waltman and Kaymak, 2008). These new algorithms are not designed to collude.…”

Section: Introductionmentioning

confidence: 99%

Hybrid Collusion: Algorithmic Pricing in Human-Computer Laboratory Markets

Normann

Sternberg

2021

SSRN Journal

View full text Add to dashboard Cite

We investigate collusive pricing in laboratory markets when human players interact with an algorithm. We compare the degree of (tacit) collusion when exclusively humans interact to the case of one firm in the market delegating its decisions to an algorithm. We further vary whether participants know about the presence of the algorithm. We find that threefirm markets involving an algorithmic player are significantly more collusive than human-only markets. Firms employing an algorithm earn significantly less profit than their rivals. For four-firm markets, we find no significant differences. (Un)certainty about the actual presence of an algorithm does not significantly affect collusion.

show abstract

“…While there exists reoccurring support for the hypothesis that algorithms can learn to set non-competitive prices and develop reward-punishment strategies (Abada and Lambin, 2020;Asker et al, 2021;Calvano et al, 2020bCalvano et al, , 2021Johnson et al, 2020;Klein, 2021), it is unclear how it compares to human collusion. 6 Market environments in previous studies on algorithmic collusion deviate substantially from the setting used in experimental market games.…”

Section: Introductionmentioning

confidence: 99%

“…Note that the V (s t ) ≡ max at Q(s t , a t ). 11Abada and Lambin (2020) andKlein (2021) initialize the Q-matrix with zeros Calvano et al (2020b). andJohnson et al (2020) use an initialization that corresponds to the discounted profit if all firms randomize their prices Calvano et al (2020b).…”

mentioning

confidence: 99%