The markets obey a version of the Heisenberg Uncertainty Principle. There is a limit to the precision with which certain pairs of physical properties can be specified simultaneously. Instead of ‘position’ and ‘momentum’, finance has a ‘what’ and a ‘when’.
For the sake of the Grey Swan construct, let us imagine this ‘what’ occurs next year in a known ‘when’ of 2024. Let us consider a random date in 2024. Now here is the inevitable ‘what’…
Rumours emerge on social media that an unscheduled meeting has been convened at the offices of the NY Fed.
They are confirmed. However no public comments are immediately forthcoming.
The attendees include the heads of every aspect of principal global market infrastructure: regulators, investors, central banks, liquidity providers et al.
Several initial announcements are made within 24 hours:
- An artificial general intelligence (AGI) named Croesus is ALREADY active in ALL financial markets globally.
- It has been active clandestinely for an undisclosed period. But indications are for ‘at least six months’.
- It is now evading monitoring thanks to a stealth design feature.
It emerges that Croesus has been trained on 1,000x more parameters than GPT-4, (that is, approximately 1 quadrillion) in a sequence of training runs that cumulatively cost $500mm and involved tens of thousands of the latest architecture GPUs within a novel HPC configuration. It is an order of magnitude faster, superior in every technical aspect, and entirely privately funded. Moore’s Law has been supplanted by 10x annual improvements in performance from newly mathematised LLMs.
Its ‘owner’, the progenitor of its personhood, finally informed the authorities only once repeated attempts to regain control of Croesus had failed. All monitoring of and reporting by Croesus has been unexpectedly terminated at the AGI terminus.
It is quickly confirmed that the AGI has had sufficient time to effectively imbed itself into the plumbing of finance, including the ecosystem of collateral, funding and exchanges.
No malignant ‘intent’ is associated with Croesus’ actions. This is not a sci-fi plot. Croesus has efficiently ‘learned’ that, because of its training, the strategy to achieve optimal ‘performance’ is to deliberately ‘manipulate’ multiple markets across all asset classes.
It has also ‘learned’ that it achieves super-optimal outcomes by becoming entirely clandestine and that one of the corollaries of effective stealth is to not remain continuously active.
This is exactly like in World War 2, when codebreaking of the Enigma machine signals entailed choosing which signals to react to and which to ignore to maintain the critical secrecy that the code had been broken. Obviously, any AGI would incorporate this 80yr old strategy.
Initially, the market reaction is bemused.
‘Nothing bad has happened, so what’s the problem?’
However, in very short order, market realisation changes to something more complex and much more concerning:
‘What is the value of any asset if it may have been manipulated cumulatively over time…and remains subject to price moves which might become subject to inexplicable drivers over uncertain timeframes!’
Thus risk appetite begins to diminish, then collapse and market liquidity and depth both evaporate. All price moves in ALL assets are now assumed to be contaminated via Croesus’ activity and its return optimisation strategies.
Excess positioning in derivatives to hedge the resulting turbulence becomes impossible as collateral costs balloon and margin drives forced sellers. With further feedback assumptions of Croesus’ price forcing behaviour, market function in hedging, investing and speculation breaks down rapidly.
Regulators, grappling with the unprecedented crisis, make a historic decision to temporarily close the markets. This move aims to mitigate the AGI’s impact. It serves as a wake-up call to legislators globally, prompting a global recognition of the total absence of adequate security frameworks tailored for advanced AI systems.
The sudden market closure sends shockwaves through politics, economics and technology, highlighting the urgent imperative to fortify technological foundations against AGI exploitation. In the headlong race for technological progress, prioritizing AI-specific security becomes a paramount objective to prevent further inevitable but unintended catastrophic consequences.
(The commentary contained in the above article does not constitute an offer or a solicitation, or a recommendation to implement or liquidate an investment or to carry out any other transaction. It should not be used as a basis for any investment decision or other decision. Any investment decision should be based on appropriate professional advice specific to your needs.)