AI and the Inescapable Nature of Bias

The amplification of bad things.

AI computation at its core is just a series of ones and zeros being processed extraordinarily quickly, manipulated by very clever programming. Because of this binary heart that lies at the centre of its being, it can be regarded at one level of perception as a pure, neutral force — an impartial machine intelligence that makes decisions based purely on data.

But the technology does not exist in a vacuum — it reflects the biases, assumptions, and blind spots of the people who build and train it. The question that will be always asked from now on is: how do those biases shape the world?

The Myth of Objective AI

Every AI model is trained on datasets that have been filtered, selected, and structured by human hands. If historical hiring records show a preference for men over women in tech, an AI trained on that data will replicate and reinforce that bias. For employment assessments, the AI posited men as programmers and women as homemakers.

It is not making a decision on its own; it is perpetuating existing social structures —some of which need changing and improving, such as racism, sexism.

It is not only individuals that suffer. Whole communities do so as well. Predictive policing models trained on historical crime data do not identify criminal activity in an objective way — they direct police activity to certain communities and not others. We already know that facial recognition software performs significantly worse on darker-skinned individuals because the training data overwhelmingly consists of white faces.

While in these cases AI does not create bias, it does perpetuate and solidify it, making it harder to challenge. AI is inherently racist or sexist— it reflects the already existing systemic biases embedded in our social institutions and digitally codifies them.

The Gatekeepers

The problem may seem to only exists at the data acquisition stage. But bias doesn’t stop there. Human beings are involved in determining how that data is processed. They can govern what data can be made for public use. These are the programmers and designers of the system plus the owners of the systems.

These people need to be able to have the power to determine the appropriateness of information that is processed and passed through the system as there are laws that have to be obeyed. No one sensible wants the nasty stuff we know is out there popping up on their screen or in the outputs of their prompts.

The developers of AI platforms may claim that obeying the law is an unavoidable but mechanistic process in itself, justifying any meddling they do with the data before it reaches us.

That is all well and good, but the issue is that these programmers and developers have, as every human being does, their own biases as well. They may have views that, while harmless on their own, may not be widely shared and may negatively counter other equally valid world-views. For instance, humans are fundamentally tribalistic by nature. Our cultures have more affect on us than we might like to think. And there are plenty of cultures out there beyond those of America and Europe.

This is important because very few people, even in the engineering community, have access and control over what is allowed to be passed on to the user through the use of the systems. You can see a previous article where the AI determined that what I wrote might by upsetting for some. But what I wrote about should be upsetting to any sane reader as it was a terrible, terrible thing to happen in human history. (I haven’t mentioned the details because I don’t want another fight with AI while editing this article — it is awful to have to think like this.)

It is not for a handful of developers and owners to hide away the truths of history just because someone might be upset at being told about it. That is just morally wrong however they may justify the censorship.

Also these people are, fundamentally, the new elite. They are the brightest brains, doing the hardest work on immensely important technology, and apart from their names on academic papers, we don’t know much about them. Who knows if they may be gripped in the traditional engineering conceit that because something is clever and hard to do then that justifies its existence — maybe not, we don’t know.

They may have spent some time studying the humanities such as philosophy or classics. But I have met plenty of dumb engineers who dismiss these subjects as irrelevant while caught up in their own arrogance and superiority about the importance of the work they are doing. They are right about the value of their work but wrong about their attitude. We all live with a philosophy of life whether we know it or not. The problem is that our personal philosophies are not necessarily universal - sorry Kant. The world itself is a bigger place than a set of microvoltages flicking off and on at high speed in a tiny, tiny space on a circuit board.

What we do know about the gatekeepers is that they live and breathe in highly rarefied strata of their institutional and corporate environments that those of us with normal intelligence will never have access to or really understand. We know that groupthink, while not universal, does exist, and it is very worrying that such a minuscule subset of the population, living in such a narrow part of life, is in the position to hide important truths from us, while at the same time imposing their own.

Breaking the Cycle

Of course some of these new high priests may be self-aware enough to realise this, and there may be some real, conscious efforts on their part to ensure that the creator’s biases are treated the same as society’s biases and that the mechanisms of power and control become transparent for all to see and not lost behind a wall of assumptions built by people thinking that they know what is good for us. I hope so.