Can Algorithms Inherit Human Bias?

The 12 Temptations blog is an ongoing series examining how markets behave under stress. We deliberately avoid prediction and advocacy, focusing instead on structure, incentives, and behavioural dynamics.

Silver has spent the past few weeks doing something that, on the surface, looks entirely familiar. After pushing above $120/oz, it has eased back into a broad $75–85 range, moving sideways more than anything else, as markets often do after a strong run. There is nothing unusual in that. Markets pause, they consolidate, they give participants time to reassess what just happened and what might come next.

And yet, the way this particular consolidation is unfolding feels slightly different, though it’s not immediately obvious why. It’s more a matter of tone than structure. Strength doesn’t seem to hold for long, rallies fade in a way that feels almost routine, and there is a recurring sense that the most decisive selling tends to arrive at the same point in the trading day, particularly as New York comes online. Outside of those moments, the market doesn’t really assert itself in either direction. It drifts, with a mild downward bias, as though something is gently pressing on it rather than pushing it outright.

None of that, taken individually, proves very much. Markets are noisy, and it’s easy to see patterns where there are none. But when a pattern begins to repeat often enough, even loosely, it tends to shift from coincidence into something worth at least acknowledging.

A narrative that settles a little too easily

The explanation being offered is that higher oil prices are a headwind. The argument is that rising energy costs place pressure on industrial demand and tighten financial conditions. In that context, softer prices begin to make a certain kind of sense, and the narrative settles quietly into place.

The difficulty is not that the explanation is implausible. It’s that it doesn’t quite sit comfortably alongside everything else we’re seeing. Fundamentals remain strong. Physical supply is tight and getting tighter. Silver, of course, does have an industrial component, so linking it to energy costs isn’t unreasonable. But gold has been moving in a broadly similar fashion, and gold is usually framed as a monetary asset, as a hedge against inflation, as something that tends to respond positively when the cost of energy, and by extension the cost of living, begins to rise. If oil is pointing toward inflationary pressure, then one might at least expect gold, to behave accordingly.

Perhaps the market is looking through that relationship or prioritising something else entirely. That happens more often than we like to admit. Still, there is a slight sense that the narrative has arrived a little too quickly and settled a little too comfortably, even as it leaves parts of the picture unexplained.

And when that happens, it can be useful to step back from the explanation itself and look instead at the conditions in which it has taken hold.

The market feels thinner than it should

What stands out, more than the story, is the feel of the market. That’s not a precise term, and it doesn’t lend itself easily to charts or models, but it is something that becomes familiar over time. Markets usually have a certain texture to them. A push and pull between participants with different time horizons, different constraints, and different levels of conviction. There are moments of disagreement, of hesitation, of sudden urgency.

Here, that texture feels thinner. Not absent but reduced. There are fewer signs of that back-and-forth, fewer moments where price appears to be discovered through genuine interaction. Instead, the movement can feel a little more uniform, as though it is being generated within a narrower set of responses.

Part of that may simply reflect who is, or is not, participating. Retail involvement, at least anecdotally, appears to have moved to the sidelines. The volatility of the past few months has not been kind to conviction, and large volatile swings tend to have a way of exhausting interest as much as they create it. People step away, not always permanently, but long enough to leave a gap.

If that gap exists, even temporarily, it raises an interesting possibility. What if, in the absence of some of those participants, the market is increasingly being shaped by systems responding to other systems?

Do machines inherit the biases we built into them?

That idea is not new in itself. Algorithmic trading has been a dominant presence for years. But there is a difference between algorithms operating within a market that is still rich with human behaviour, and algorithms operating in an environment where that behaviour has thinned out. In one case, machines are interacting with people. In the other, they may be interacting more heavily with each other, drawing signals from the same data, reacting to similar triggers, and reinforcing similar patterns.

The models behind those systems are not neutral in the way we sometimes assume. They are built on historical data, and that data is, at its core, a record of human decision-making. Fear, momentum, overreaction, and hesitation. All of these leave traces that can be measured, modelled, and incorporated into strategies. Over time, those strategies become more refined, more responsive, and in some cases more dominant.

Which leads us to the following question.

If algorithms are trained on patterns of human behaviour, are they free from the biases that shaped those patterns, or are they, in some sense, a distilled version of them?

Recency bias does not disappear simply because it is expressed through code. If a model is calibrated on what has worked in the recent past, it will naturally give more weight to those outcomes. The same can be said, in a looser sense, for narrative. The choice of inputs, the framing of signals, even the structure of the model itself can reflect assumptions about what matters and what does not. When many systems are built along similar lines, the result can be a kind of quiet alignment, where responses cluster rather than diversify.

This is where the idea of markets as adaptive systems becomes interesting again. The notion that markets evolve, that participants adjust to each other, that strategies decay as they become crowded. All of that has been observed for a long time. But when a growing share of that adaptation is being carried out by machines, it’s worth asking what “adaptation” actually looks like.

Do these systems genuinely evolve, in the sense of revisiting their underlying assumptions, or do they primarily optimise within a given framework, becoming increasingly efficient at identifying and exploiting patterns that may themselves be temporary? And if those patterns begin to break down, does the adjustment happen smoothly, or does it come with a degree of instability?

It may be that none of this is especially unusual. Markets have always had periods where they feel less intuitive, where price action seems disconnected from the explanations offered for it. This could simply be another one of those periods, shaped by positioning, macro crosscurrents, and the normal ebb and flow of participation.

At the same time though, there is something about the current behaviour in silver, and to a degree in gold, that feels more mechanical than usual. Not in a dramatic sense, but in a way that is noticeable if you spend enough time watching it. The patterns are there, the responses are consistent, and the narratives, while plausible, don’t quite account for the whole.

None of that leads to a firm conclusion, and for now it may be enough to simply recognise the possibility that the balance between human behaviour and systematic trading is not fixed, and that at times it may shift more than we realise. If so, then the patterns we rely on may not behave in quite the same way.

And in that case, the more interesting question may not be what the market is doing, but who, or what, is doing it.