Fascinating event in Brussels today on the competition law challenges of pricing algorithms and artificial intelligence. Drawing on the brilliant writing of Ariel Ezrachi from Oxford. There are many interesting issues, but most of all is the prospect that rival machine-learning algorithms could teach themselves collusive behaviour. The speed of interaction on line could make the process of reaching agreement lightning quick – and the black box nature of AI systems mean responsible firms might not even know what their computers were up to.

Tacit collusion is not a crime. So, goes the argument, the algorithms can happily rip off consumers and exploit an enforcement gap. I can see at least three responses to this:

– It is simply not a problem. There is a reason why tacit collusion is not enforceable in the ordinary course. If markets don’t produce competitive outcomes, this will stimulate new entry and introduce disruptive new forces into the market.

– Technology will solve its own problems. Maurits Dolmans from Cleary Gottlieb argued the point well today that one cannot look at the risks only from one side. We can expect high tech tools to continue to evolve give consumers the power to band together and exploit disruptive buyer power, or to circumvent higher prices by obtaining personalised and secretive discounts.

– The enforcement tools are already there. Abuse of a dominant position is already prohibited under EU law, including excessive pricing. Where cases in the offline world would fall down on (i) proving tacit collusion to find joint dominance; or (ii) establishing that prices are genuinely excessive. Could concrete algorithmic evidence help overcome both these evidential hurdles? Or could the rapid machine-learning interactions between two algorithms cross the threshold for a 101 infringement based on information exchange?

But it’s not clear that these responses will apply in every case. There could be a genuine enforcement gap here, where EU competition authorities could struggle to find a nexus for intervention without ex ante regulatory tools. That option looks very unattractive in dynamic technology driven markets. So option 3 looks like the most likely to me. For now these arguments are based on theory rather than evidence from case work – but how long before we see a dawn raid that aims to seize an algorithm?