AI is pointless if it does not boost productivity
It’s hard to escape the feverish narrative around artificial intelligence. Markets are pricing in a revolution, venture money is flooding in, and the news cycle swings between euphoria and existential dread. But beneath the hype sits a simple test that should decide whether AI deserves its valuation and its cultural authority: does it measurably raise productivity where it matters?
The promise, the peril—and the purpose
AI’s trajectory is undeniably rapid. Practitioners warn that systems are improving at a pace many outsiders still underestimate, with risks that range from autonomous misalignment to abuse by criminal networks or rogue states. Safety, governance and access controls are essential.
Yet even if these concerns are handled well, a harder question remains: what is all this for? If the capital, energy and ingenuity pouring into AI don’t translate into tangible gains in output and service quality, the societal return will fall short regardless of model benchmarks or stock prices.
The staggering bill
Training and running cutting-edge models demands extraordinary investment in data centers, chips, and the energy infrastructure to power them. Governments and corporate leaders are lining up to bankroll that buildout. But the public deserves a clear productivity thesis in exchange for the subsidies, grid upgrades and regulatory flexibility now being requested.
Cutting labour costs isn’t the prize
Yes, companies see AI as a lever to tame payrolls and automate routine tasks. That’s unsurprising—and not, in itself, a bad thing. But the bigger prize is total factor productivity: doing more with the same (or fewer) inputs. If AI adoption simply shifts costs around without lifting output, it misses the point.
The UK offers a cautionary tale. Productivity growth has been weak since the 2008 financial crisis. Dig into the numbers and you find public services—particularly large, complex institutions—are a major part of the story. Everyone talks about modernising the National Health Service; few make it a genuine, funded priority with measurable targets.
Health versus education: a skewed balance
Consider spending patterns. In the late 1970s, health and education each accounted for roughly 5% of UK GDP. Today, healthcare is closer to 12%—more than double education’s share. For a country facing an ageing population, that might sound inevitable. But is it optimal?
Economically, there’s a strong case that boosting education yields larger long-run productivity gains by raising skills, enabling better health choices and enhancing adaptability. Meanwhile, some of the most immediate productivity wins may actually be more tractable in healthcare than in classrooms: streamlining administration, improving triage, reducing errors, and accelerating diagnostics.
The longer healthcare absorbs a larger share of national output without commensurate productivity gains, the more it crowds out other priorities—or forces higher taxes. The UK Office for Budget Responsibility continues to flag unsustainable debt trajectories by mid-century unless the big cost drivers—NHS, welfare, pensions and social care—are bent onto a different curve.
A productivity-first AI agenda
If AI is to earn its keep, leaders need a serious, evidence-based plan that prioritises productivity where it moves the needle most. That means clear use cases, rigorous measurement and transparent accountability. A practical agenda could include:
- Healthcare delivery: Automate scheduling, referrals and discharge; deploy AI-assisted triage in urgent care; use decision-support for imaging and pathology; target fraud and waste in procurement. Measure hours saved, wait-time reductions and clinical outcomes—not just pilots launched.
- Education enablement: Provide teacher co-pilots for lesson prep, marking and personalised feedback; free scarce classroom time for human interaction. Track teacher workload, student progress and attainment gaps.
- Public sector operations: Standardise data and processes across agencies; use AI for case management, benefits administration and analytics; publish service-level gains and cost-to-serve metrics.
- Data foundations: Invest in interoperability, privacy-preserving data sharing and quality assurance so models can work with reliable, secure inputs.
- Energy-aware AI: Tie data center expansion to efficiency targets, low-carbon power procurement and heat reuse. Report energy intensity per unit of task output.
- Governance and safety: Mandate audit trails for high-risk uses, red-team critical systems, and align procurement with open standards and model choice where appropriate.
Show the receipts
This comes down to discipline. For every major AI deployment, define a baseline, set a target, publish the delta. How many staff hours were released back to the front line? How much did patient throughput improve? Did test scores rise for disadvantaged students? Did case resolution times fall? If the answers aren’t compelling, stop, redesign and redeploy—or don’t scale.
The bottom line
AI may well be transformative. But transformation is not the goal—productivity is. If these systems cannot materially raise output and service quality where societies need it most, then the capital, carbon and attention they consume won’t be justified.
Markets can reward capacity buildouts; citizens need better outcomes. The mandate is clear: channel AI into the hard problems—healthcare backlogs, stagnant public-service productivity, skills and education—measure relentlessly, and be willing to pivot. Otherwise, the most accurate verdict on this era of AI will be the harshest one: powerful, dazzling—and pointless.
