AI As An Excuse

(This is the fourth article in the series looking at the human cost of Artificial Intelligence)

A phrase is starting to creep into corporate strategy decks: “AI-enabled workforce optimisation.” It’s vague, sounds clever, and usually comes right before a wave of layoffs. Some teams get cut. Some roles get merged. There’s a nod to transformation, then silence.

What’s really happening is this: AI is becoming a convenient reason to shrink the workforce.

And it’s not because the technology can already do everything. Most of what’s being introduced today into offices, contact centres, HR functions, and marketing teams isn’t true intelligence at all. It’s automation with polish.

What most companies are adopting now falls under Artificial Narrow Intelligence (ANI). These systems are designed to perform specific tasks. They don’t learn outside their boundaries, and they don’t understand what they’re doing. A search ranking algorithm. A fraud detection tool. An automated scheduler. All useful. All narrow. All ANI.

The vast majority of ANI being adopted in business settings today falls into two camps. The first is generative AI: tools that produce text, images, summaries, code, and so on. These are usually large language models (LLMs) or systems built on top of them; the kind powering generative AI tools like ChatGPT, Copilot, Gemini, and the like. These are only ANI but they often look or feel broader because of their flexibility. They generate emails, summaries, code snippets, outlines, images, answers to support queries. They do it fast, and with a tone that feels human.

But they don’t know what they’re saying. They generate by guessing the next likely word, not by reasoning. And while they seem to get better with each update, there’s a growing issue: the more fluent they become, the more plausible their mistakes. They’re hallucinating more often, but in a way that’s smoother. Subtler. Easier to miss.

The second type of ANI is a hybrid of sorts: models that combine LLM outputs with traditional rule-based logic (symbolic AI). These try to combine the flexibility of LLMs with the control of rule-based systems. You might see them in customer service platforms that let a chatbot handle FAQs but pass you to a real person when the request gets specific. Or in a legal drafting tool that uses an LLM to propose language but relies on predefined templates to check that the result stays within the law. These hybrid models offer a bit more control. But they’re still narrow. They’re not general-purpose minds.

Are they useful? Absolutely. Game-changing? Not really. Not yet. They are prone to error. They hallucinate. They need context, structure, and curation. What’s more, as these models grow in size, their tendency to produce fiction with confidence actually increases. In short: the smarter they sound, the more carefully they need watching.

And this matters. Because when someone says, “We’re replacing ten jobs with AI,” they rarely say, “and we’ve budgeted for the people we’ll need to check its output.”

You see, these tools aren’t really autonomous. They’re not self-aware. They don’t understand. They can’t validate. They’re guessing based on probabilities. Yes they’re incredibly fast and often seem to have an impressive grasp of tone, but they’re guessing all the same. And that’s not intelligence. It’s pattern mimicry at scale.

Which brings us to Artificial General Intelligence (AGI). This is what people imagine when they hear “AI is going to take your job.” AGI is a different beast entirely. It wouldn’t just complete tasks; it would reason, reflect, adapt across contexts and domains, and make choices based on goals it could formulate. AGI would be less like a search engine and more like a person you could delegate to. It could handle unfamiliar tasks, draw connections, shift between goals. It wouldn’t just imitate; it would understand. That’s what would make it genuinely disruptive.

But here’s the reality: we’re nowhere near it. Not via LLMs, anyway. AGI is about cognition, something that LLMs can never have due to how they are designed. And the further we stretch LLMS, the more brittle they show themselves to be.

And we’re even further from Artificial Superintelligence (ASI); the theoretical point where AI outpaces human intelligence in every respect. That’s the sort of AI governments and ethicists worry about in the long term. Right now, though, the challenge isn’t superintelligence. It’s misused narrow intelligence wrapped in overconfidence.

So why the rush? Why the sweeping claims for what AI can do, and why the mass exits?

Because for some organisations, AI isn’t really about intelligence. It’s a justification. A narrative that makes cost-cutting sound like innovation. If you say you’re reducing headcount because of AI you might even get a bump in the stock price. But say you’re doing it because you want cheaper labour and you might get awkward questions.

It’s not always calculated malice. Sometimes it’s just wishful thinking at scale. Organisations who have believed the hype. A belief that with the right model, the right prompt, the right plug-in, you can automate not just tasks but judgment, context, and care. Sorry, but you can’t.

Businesses are making serious decisions (redundancies, restructures, role redesigns etc) based on tools that don’t yet offer true reasoning, context awareness, or reliability. These are tools that generate, not tools that judge. They can assist, yes. Replace? Not safely. Not yet.

And here’s the most overlooked risk for businesses: if you scale back human involvement, you remove the very safety net that makes these tools viable.

That’s why the EU AI Act has made human oversight mandatory for higher-risk systems. Hiring algorithms. Medical decision tools. Education platforms. Financial risk assessments. Anything that affects someone’s future in a significant way must involve people; not just in principle, but in practice. There need to be eyes on the output. Human brains checking the logic.

Now, the UK isn’t bound by the EU AI Act, having chosen a different regulatory path post-Brexit. But for any British firm doing business in the EU, or handling data that touches it, compliance still matters. If you sell to Europe, you follow the rules. If you use high-risk systems, you need explainability. You need oversight.

But whether or not the regulation applies, the logic behind it still holds. These tools make convincing mistakes. And unless someone spots those mistakes, they get amplified. They go out into the world as decisions. And they’re hard to undo.

This is exactly why AI literacy needs to be a central part of the workforce. Not just for the developers or the data scientists. But for everyone. Managers. Analysts. Frontline staff. The people using these tools need to understand what they’re doing, what they’re good at, and where they fail. Because they will definitely fail. They will continue to hallucinate, and someone has to spot it. That includes understanding how plausible hallucinations are becoming more common, not less. It means training people not just to click the button, but to ask: “Does that really make sense?”

This kind of scepticism isn’t a blocker. It’s a requirement. If we want AI to be used effectively and safely in real organisations, we need people who know their subject. People with enough confidence to question what looks like a finished product. People with an eye for detail. Who aren’t swept away by slick formatting and confident tone. That’s not the “yes men”, but people who aren’t afraid to challenge. That kind of judgment and confidence doesn’t come from a weekend course. It comes from time. From experience. From context.

And that’s exactly what many businesses are cutting.

Because running underneath this technological shift is a very human assumption. A belief that people will simply reskill on their own. That once a job goes, the worker will go home, open a laptop, download a course on Python or prompt engineering, and return a few weeks later with a new profession and a sunny outlook.

This is fantasy. And it’s one that conveniently lets employers off the hook.

In the real world, most people can’t afford to stop and learn full-time. They’re managing rent. Kids. Parents. Burnout. Uncertainty. Stress. Health. Debt. They’re juggling multiple fronts just to keep life steady. Telling them the solution to redundancy is “just upskill” is like pushing someone off a boat and throwing them a manual on how to swim.

When companies benefit from automation, when productivity rises, headcount shrinks, and the bottom line looks healthier, they also carry a responsibility. Not just a moral one, though that matters. A practical one. Because there is no AI without people who understand it. And no future for a business that burns through its workforce without building new capabilities in return.

Support doesn’t mean a single lunchtime webinar. It doesn’t mean a vague note about “career pathways” on the intranet. It means funded training. Real courses, with protected time to attend. It means career coaching tailored to the person, not a PDF packed with generic tips. It means transitional roles, linking old skills to new needs, so that people aren’t asked to leap across a chasm on day one. For more on this have a read of my recent article In the In-Between.

If a company values its staff, that value shouldn’t end the minute someone becomes economically replaceable. Because the message you send, when you cut someone loose without a plan, is that they were only ever useful. Not trusted. Not valued. Just convenient.

Some firms justify this by citing competition. If they don’t adopt AI, they say, someone else will. If they don’t cut costs, they’ll fall behind. There’s some truth in that. But it also shows a lack of imagination.

Because there’s another way to compete. Some companies will earn loyalty by guiding people through transition. By offering direction when others offer a shrug. By showing belief, not just in the tech, but in the people expected to use it.

This isn’t just about ethics. It’s about resilience. You can have all the tools in the world, but if nobody around you knows how to challenge, interpret, or adapt them, you’re not ahead. You’re exposed.

If companies really want to adopt AI responsibly, they need to build environments where questions are encouraged, not suppressed. Where pointing out an error isn’t seen as a failure, but as essential maintenance. That requires confidence. And confidence comes from knowledge.

Replacing staff with a tool you don’t fully understand is not transformation. It’s negligence dressed up as strategy.

But there’s another path. One that doesn’t treat AI as a substitute for people but as a support. One that recognises what these systems can do well, and where they are fundamentally flawed.

There’s still time to make this shift. To invest not just in the tech, but in the people expected to work alongside it. To stop using AI as a pretext for downsizing, and start using it as a reason to upskill and reimagine.

If that sounds idealistic, perhaps it is. But if you’ve read this far, you probably know the cost of not trying.

And in a world full of noise, hype, and half-truths, clarity is a rare thing. Businesses can choose to provide it, or keep chasing illusions.

A company is more than software and systems. It’s people. Memory. Intuition. Skill. Judgment. If you cut too deep, you lose more than staff. You lose identity. And that’s not something you get back with a platform subscription.

Efficiency gained by erasing the human element is not clever. It’s fragile. It looks strong on paper until something goes wrong, and then there’s no one left who remembers how to put it right.

If businesses want to share in the gains of AI, they must also shoulder some of the disruption. Not by writing it off, but by preparing for it. Supporting it. Planning for it with as much care as they planned their last software rollout.

Anything less is cowardice wrapped in a balance sheet.

And if that’s the legacy we leave, tools that displace more than they support, businesses that shrink before they build, workforces trained to click but not to think, then AI won’t be the problem.

We will.