As part of UC’s Cybersecurity Summit 2026, I attended a talk from Dr Safiya Noble about how LLM has learned the sins of the past from SEO and marginalises others. I’ve previously attended a talk from her during Critical Voices 2025: Beyond the Code, and I knew I’d be in for a treat at this one.
She had a number of interesting viewpoints, but I think the one that has stuck with me the most is her suggestion on reframing the narrative around AI/LLM technology entirely. We keep calling them ‘tools’; she argues we should be calling them ‘products’. Tool is a neutralising word. It means that the described is a neutral technology that can be used for anything, whether good or bad. A hammer is not inherently bad just because you can use it to inflict damage on someone.
AI/LLM models are not simply ‘a tool’. The biases in their training data are reflected in every response they generate. There are very real impacts of everything from rare earth mineral mining for the semiconductors used, to the datacenter’s consumption of energy and water, through to the issues with slave labour used for some of the training tasks.
And the output they generate is soulless. There’s a fascinating article in Noema that argues – with citations as wide as Gartner and industry insiders – that the future of knowledge work is mixed at best, with companies increasingly returning to biological workers as a competitive advantage and a differentiator.
Now, dear reader, I don’t want you to get the wrong idea. I am not some sort of technophobe and I don’t hate the idea of AI/LLM technologies existing alongside us. I think that if used correctly, we can harness a lot out of them. But I think the way some companies have recklessly adopted them to fully replace workers is irresponsible and serves a narrative that is neither truthful nor fair: that knowledge work is something that can be productised and manufactured.
Ah, there’s that word again: product. So here we are: let’s stop calling them AI/LLM “Tools”, and start calling them AI/LLM “products”. I think that is significantly more reasonable as a framing and I agree with Dr Noble on all points.
It also doesn’t denigrate the capabilities of the technology (or its users) by calling them ‘clankers’ or ‘slop generators’. I’ve long felt uncomfortable with those terms. And to those who want to continue using such insults because “they’re unthinking, unfeeling machines”: that’s great, then why are you using an emotionally-charged insult to describe them? If you don’t want to anthropomorphise LLMs, perhaps you shouldn’t.