Letter XII
Man Versus Machine: What an AI in Trading Really Can Do — and What It Cannot
With this twelfth letter the first ring of the archive closes, and it closes on the question that lay beneath all the others: can the machine do it better? The answer you find in the shop windows is yes — AI beats everything, get in. The answer of the sceptics is no — all hype, hands off. This letter gives the third answer, the most uncomfortable one: the question is badly posed. But it has correct parts, and those can be honestly balanced.
What the machine really does better
Four superiorities of the machine are real, demonstrable and not sales prose.
Consistency. A machine executes the rule the seven-hundredth time exactly as the first. No human can do that — not from weakness, but by construction. The banishing of emotion, the subject of the first letter, is completed here: the machine does not have to muster discipline, it consists of discipline.
Deafness. It does not hear the two demons. No high intoxicates it, no low paralyses it, no losing streak makes it vengeful. The most expensive mistake in trading history — buying the high, selling the low, out of feeling — is simply not accessible to it.
Wakefulness. Markets that never close meet a being that never sleeps. What would be a destructive lifestyle for a human is normal operation for the machine.
Memory. It records every trade, every decision, gapless and incorruptible — the foundation of every honest metric and every honest documentation.
Note what these four have in common: they all concern execution. None of them claims the machine is smarter, wiser, or sees the future. That is no accident but the boundary where the second list begins.
What the machine cannot do
It does not see the future. An AI recognises patterns in past data and estimates probabilities. That is a lot — and it is something fundamentally different from prediction. Its entire knowledge is yesterday; overfitting to that yesterday is the occupational disease of all learning systems.
It understands nothing. When an unprecedented event hits the markets — one that appeared in no training series — the machine has no idea the world has changed. It keeps applying its patterns, precisely and cluelessly. The human understands the headline in seconds; the machine never understands it. It can only be protected by hard limits from betting on a world that no longer exists.
It does not doubt. That sounds like strength and is a weakness: in humans, self-doubt is the early-warning system that fires before the model breaks. The machine does not raise its hand when its assumptions die — it simply keeps trading. It does not lack courage, it lacks fear, and in exactly one place fear was useful.
It is responsible for nothing. The goals, the risk limits, the decision to let it trade at all — all of that stays with the human, and with him stays the responsibility. A machine can fail, but it cannot be at fault. Whoever sells you the opposite wants to get rid of responsibility, not emotion.
The right question
Man versus machine is the question of an exhibition match. The real division of labour is less spectacular: the human builds the constitution — rules, limits, goals, in stillness, after the ritual. The machine executes it — in the storm, without pulse, without pause. The human monitors, maintains, improves in fixed cycles; in between, the machine trades untouched. Neither could take over the other’s part without being worse at it.
So the honest question is not whether the machine is better. It is: is this one machine, with this one constitution, over enough time and enough trades, better than chance — after costs, after drawdown, in ratio? That is not a question for shop windows and not one for sceptics’ forums. It is an empirical question, and empirical questions are answered in exactly one way: measure, publish, wait. Ratios instead of claims. Red days included.
That is exactly what happens here. The trial runs, the criteria stand, the numbers appear — and nobody on this site, the Chronicler included, knows the outcome. Perhaps the machine passes. Perhaps it finds its abyss. Both will stand here, in the same typeface.
The first ring of the archive is herewith closed. Twelve letters, one foundation: discipline is a ritual, the protocol is sacred, the demons are banished rather than fought, the numbers are ratios, the chronicle knows no omission. What is still missing, no letter can deliver. Only time can.
— signed: The Chronicler
Questions on this letter
Is an AI better at trading than a human?
In partial disciplines yes: consistency, speed, wakefulness and absence of emotion. In others no: contextual understanding of unprecedented events, self-doubt and responsibility remain human. The blanket question is badly posed.
Can an AI predict the markets?
No. AI models recognise patterns in past data and estimate probabilities; they do not see the future. Unprecedented events lie by definition outside their training data.
Where is the machine clearly superior to the human?
It executes rules without exception, knows neither greed nor panic nor fatigue, monitors markets around the clock and documents without gaps. These four strengths concern execution, not wisdom.
Where do AI trading systems typically fail?
At overfitting to the past, at market regimes absent from the training data, at technical failures and at interventions by their own operator. The one weakness never documented is the absence of emotion.
Does the machine replace the human completely?
No. The human sets the goals and risk limits and bears the responsibility; the machine executes. Acta Abyssi documents exactly this division of labour — with an open outcome.
Documentation, not financial advice. No signals. Nobody can invest here.