AI Is Not a Thing

Or Why Calling Everything “AI” Is Like Calling Every Fight “UFC” 

That’s UFC 

There is a certain kind of person who sees a boxing match, a judo throw, and a spinning back kick and says, with great confidence, “That’s UFC.” 

Anyone who has trained for more than five minutes knows how wrong that is. UFC is not a martial art. It is an event format. Inside it sit wrestling, boxing, Muay Thai, jiu-jitsu, karate, and whatever else happens to work under that ruleset. Each of those disciplines has different techniques, training methods, assumptions, and failure modes. Collapsing them into a single label does not merely lose nuance. It makes meaningful analysis impossible. 

That is exactly what happens when lawyers, regulators, and vendors talk about “AI”. 

AI is not a thing. It is not a technical category, a legal category, or a coherent object of ethical analysis. It is an umbrella label applied to a wide range of computational techniques that happen to share one marketing-friendly feature: they look like intelligence from the outside. 

For a profession that begins every serious analysis with definition, that should be immediately uncomfortable. 

A Bucket, Not a Category 

If you open a dictionary, you will find a definition along the lines of this: artificial intelligence is the capability of computer systems or algorithms to imitate intelligent human behaviour. That definition is broad, but it is not controversial. It is also doing far more work than most people realise. 

By that definition, AI has been embedded in ordinary computing for decades. Spell-checking is AI. Search ranking is AI. Spam filtering is AI. Fraud detection is AI. Face recognition is AI. Predictive text is AI. Cloud infrastructure optimisation is AI. Routing, clustering, recommendation systems, anomaly detection, and heuristic decision engines all sit comfortably inside the definition. 

If you are using a modern operating system, a browser, a word processor, or any serious cloud service, you are already using AI constantly and unavoidably. You cannot meaningfully opt out without opting out of modern computing itself. 

This matters because it immediately exposes the problem with sweeping statements about “using AI” or “not using AI”. Those statements are not precise enough to be true. 

Old Tricks, New Interfaces 

What most people actually mean when they say “AI” today is generative AI, particularly large language models that produce text, images, code, or audio. ChatGPT, Copilot, Gemini, Claude. That is a narrower category, and it is a powerful one, but it is not unprecedented. 

We have been here before. In the 1960s, ELIZA astonished users by simulating a psychotherapist through pattern-matching prompts. People attributed understanding and empathy to it almost immediately, despite the fact that it had neither. In the 1990s and early 2000s, computer programs that played checkers or chess at superhuman levels were widely described as intelligent, even though they relied on brute-force search, heuristics, and evaluation functions rather than anything resembling human reasoning. 

Each of these systems was, at the time, genuinely impressive. Each triggered waves of speculation about the end of human expertise. And each, in retrospect, was simply a new interface over familiar computational ideas. 

Large language models sit in that same lineage. What has changed is not the existence of probabilistic language systems, but their accessibility. Chat interfaces feel conversational, and humans are very bad at not projecting intention, judgment, and understanding onto systems that speak fluently. That psychological effect is doing much of the work in the current hype cycle. 

Which is precisely why the legal profession finds itself in an awkward position.  

Where This Actually Becomes Dangerous 

All of this matters because the risk lawyers are currently creating for themselves is not the use of technology. It is the making of false or careless declarations about that use. 

Declarations are not opinions. They are statements of fact made for a legal purpose. Professional conduct rules have always treated them seriously for exactly that reason. A declaration that is wrong, even if made without malice, can undermine credibility. Repeated or reckless false declarations can amount to professional misconduct. That has nothing to do with AI and everything to do with honesty, precision, and care. 

The problem is that many of the declarations now being made use the word “AI” without definition. That is a drafting failure. If a practitioner cannot explain what they mean by AI, they cannot truthfully declare that it was not used. Modern legal practice relies on systems that fall squarely within ordinary definitions of artificial intelligence. Pretending otherwise does not make the declaration safer. It makes it false. 

This is not a theoretical concern. We have already seen disciplinary and quasi-disciplinary attention directed at practitioners who made confident statements about technology they did not properly understand. The lesson is not “avoid technology”. The lesson is “do not declare things you have not analysed”. 

The safest approach has always been the same one lawyers use everywhere else. Define your terms. Limit your claims. If necessary, qualify statements by reference to knowledge and understanding. There is nothing unethical about using modern tools. There is something very unethical about asserting facts that are not true. 

Back to the Cage 

Which brings us back to UFC. 

If someone asks whether you “do UFC”, the technical answer is “no” (unless you are a fighter currently signed to the UFC promotion). The practical answer (to the non-technical question that they are asking) is not “yes” or “no”. It depends. Do you box. Wrestle. Train jiu-jitsu. Kick. Compete under a particular ruleset. 

Answering “no” because you do not like the label does not make you safer. It just makes you wrong. 

If someone asks whether you are “using AI”, the technical answer is “yes” (unless you are entirely off-grid). The practical answer (to the non-technical question that they are asking) is not “yes” or “no”. It depends. Once you start making declarations about it, that sloppiness stops being academic. It becomes a professional risk. 

And unlike a bad fight, you do not get to tap out halfway through a declaration.