AI: Through precision we can have it all
Artificial intelligence is often talked about as a single technology. This is unhelpful. AI is an umbrella term covering a wide range of technologies, from translation tools and recommendation engines to predictive models and the LLMs that power ChatGPT and many generative and agentic AI systems. These technologies have different lineages, are designed for very different purposes, work in different ways and raise different questions. Yet, they are routinely bundled together under the single label of AI; a simplification that obscures important distinctions and makes informed discussion and decision-making far more difficult than it should be.
Why does this matter? Because confusion has consequences. Because the way we talk about AI shapes governance, regulation, investment and its associated power, critique, and adoption. When very different technologies are treated as one, decision-making can only be ill informed. Concerns that are relevant to one type of AI may be wrongly applied to another, while genuine risks can be overlooked. Organisations may struggle to identify which tools are appropriate for particular problems, policymakers may design rules that miss their intended target, and public debate can become dominated by hyperscaler hype, fear or false choices. Therefore, understanding the differences between AI technologies is not simply a matter of technical accuracy; it is vital for making well informed decisions about their design, development, deployment, ownership, governance and future.
Greater precision in how we envisage, describe and discuss different AI technologies is absolutely essential for improving our shared understanding – whatever the context.
One of the most common consequences of imprecision is the creation of false choices that shut down legitimate scrutiny.
Take a scenario where someone raises a concern about generative AI: its environmental costs; its impact on creative labour; its concentration of power in a handful of technology companies; its tendency to produce misinformation, or the speed at which it is being deployed across society. Yet, before that concern can be examined, the conversation is redirected by someone offering a false choice.
This pattern appears frequently in media coverage and public debate.
Ashley: “Are we all being far too reckless in accepting the rapid roll-out of generative AI?
Mike: Well, the power of AI is already being used in health systems around the world to identify cancerous cells faster and, in some cases, as accurately as specialist clinicians. AI is already accelerating scientific discovery.”
At which point, often critique ceases. After all, who wants to question AI that might help tackle cancer? However, be assured, Mike’s response is a conversational sleight of hand. Critique of LLMs and generative AI become entangled with entirely different AI technologies, applications and ambitions – and once that act of misdirection is deployed, meaningful scrutiny falters.
The antidote is precision. By being more specific about which AI technologies we are discussing, we can avoid such false choices, have more honest conversations, and ask sharper questions about what kinds of AI we will accept. Precision allows us to evaluate technologies on their own merits, rather than treating AI as a single phenomenon that must either be embraced wholesale or opposed entirely.
If Ashley specified that she was talking about LLMs and the generative and agentic AI systems built on top of them, whereas Mike was referring to computer vision systems used in medical imaging, it would immediately become clear that they are dealing with two different applications of AI. Ashley could then pursue her question.
Therefore, through precision, we can support AI for cancer detection AND regulate the hyperscalers. We can have both. We can have it all.
For those wanting a little more technical detail about this example, here’s a brief summary…
Artificial intelligence is increasingly being used to help doctors detect cancer in medical images, from mammograms and CT scans to digital pathology slides. These systems are typically built using computer vision models trained on vast datasets of medical images labelled by experts. Over time, they learn to recognise subtle visual patterns associated with disease, including changes in cell shape, tissue structure and other features that may be difficult for humans to identify consistently at scale.
When presented with a new image, the system analyses it and produces an assessment of the likelihood that cancer is present. Many tools can also highlight suspicious regions, helping clinicians focus their attention on areas that may warrant closer examination.
LLM’s work differently. Rather than learning from images, they are trained primarily on enormous quantities of text and learn statistical relationships within language. Their strength lies in generating, summarising and analysing text rather than identifying visual patterns in medical imagery.
In short, AI systems used to identify cancer in medical images are not the same as the LLM’s that underpin generative AI tools such as ChatGPT, Claude and many agentic AI systems. While both belong to the broader field of artificial intelligence, they are trained on different kinds of data, designed for different purposes and raise different social, economic and regulatory questions.
Precision matters. By distinguishing between different kinds of AI, we free ourselves from the false choice that is so often presented to us: embrace everything or reject everything. We can encourage innovation AND demand accountability. In short, through precision we can have it all, which is the ability to make informed choices about which AI technologies deserve a place in our shared future.
