The Australian Energy Market Commission has finally published its report into the potential of AIs colluding to push up electricity prices, more than a year after the report was concluded, but says that the rapid growth in home batteries could offset some of the problems.
The report from the AEMC, the market rule maker, says the rising use of algorithms to manage complex battery trading is creating the perfect conditions for tacit collusion between AIs.
“The risks have increased in recent years due to two inter-related trends, an increase in information available to participants that has the potential to reveal information about competitors’ strategies, [and] the development of AI tools,” the report says.
Algorithmic collusion is when AIs use the information they have to test strategies and learn to charge higher-than-normal prices, all in lockstep but without communicating with each other.
If AIs are making collusion easier and cheaper to do, detection is both potentially much more expensive and difficult.
The AEMC paper acknowledges that it’s not clear how this kind of collusion would be detected, any methods would be extremely expensive for regulators, and solutions such as requiring algorithms to be retrained are impractical.
Furthermore, the horse may have already bolted for monitoring “black box” systems because there is no way to know how it used information to come to an outcome.
An obvious solution of restricting software known as auto-bidding is also fraught, with the AEMC saying it’s useful to manage extremely complex systems and contributes to lower prices.
Big batteries were thought to be an antidote to the high prices set in periods of high demand, potentially denting the market power of the traditional dominant utilities. But many of the batteries now being rolled out are owned, or controlled, by the very same market players.
And already, even without overt evidence of collusion on battery trading by AIs, batteries are making their market power felt.
In July, quarterly data from the Australian Energy Market Operator (AEMO) showed big batteries are now dominating pricing activity in the NEM.
Battery discharge set prices at an average of $478/MWh in the June quarter, almost double the $245/MWh average in the same period last year and nearly three times the average price set by gas generators.
Home batteries to the rescue?Â
Even in 2024, well before the federal government’s promise of a 30 per cent rebate that has super-charged the home battery market, these devices were seen as one of the ways to prevent the NEM from descending into an AI hellscape.
The AEMC says the responsive supply and demand from home batteries is “important to address the risk of algorithmic collusion”.
If consumer batteries can flood in when prices are high, or even just higher than usual, that will disrupt colluding algorithms by reducing the profit earned, the report says.
Hedging contracts are a short-term solution as well, as returns are capped, but the AEMC warns that artificially pushing prices up by a little over time would see these contracts get more and more expensive.
The paper says current NEM practices and rules around accountability, information transparency and bidding practices provide some guardrails – provided the algorithms stay within them, given monitoring is likely to be prohibitively expensive.
The AEMC also recommends expanding the information gathering powers of the competition cop, the Australian Competition and Consumer Commission (ACCC), and setting up a working group with it and the Australian Energy Regulator to figure out how to deal with the problem.
Collusion a softly-softly approach
One example of where collusion could take place is the liminal area where batteries can start making money.
There are three clusters where generators can bid into the NEM. There’s a floor price, bids up near the price cap to take advantage of extreme prices, and bids centred around short-run marginal cost.
It’s the last cluster where algorithms could shift small quantities of bids upwards to lift marginal prices across the board.
“[This] case is of particular concern as it is an extremely low-risk option for generators. Indeed, each coordinating generator is only risking a very small percentage of their total offered quantity for a much higher payoff,” the report says.
“In other words, the losses of the coordinating generators [if a competitor doesn’t participate] are small, meaning that a collusive outcome, and thereby higher prices, are more likely.”
Doing so is all about using the wealth of information that is allowing battery-trading AIs to see, or figure out, their competitors’ bidding strategies, as well as a large amount of current and future information from AEMO.
In 2021, 15 different technology providers were selling auto-bidding software in the NEM, the biggest of which was Mosaic which is now owned by Fluence.
Mosaic forecasts future prices in energy and frequency control ancillary services (FCAS) and uses that to calculate what the best bid will be for every market, and every time period.
But physical constraints on building generators and batteries are also contributing to a landscape ripe for AI manipulation.
Transmission congestion, very-long planning and connection wait times, and the rising cost of building generators (if not batteries which are seeing prices plunge), equal a NEM with high barriers to new generation.
