What do you think is the future of AI?
The future of AI — what do I think, or what do I want?
A Google search will, of course, answer the prompt in full. By 2040, I expect AI to run invisibly beneath the surface of vehicles, appliances, medical diagnostics, and classrooms.
As an observer, I would advocate for win-win business models. And for AI, its purpose should be to empower people, not to replace them. Unfortunately, where there is substantial gain on the industry side, there is often a big hit on the receiving end — both economically and socially. But who is the “gladiator” to mediate the intense geopolitical superpower rivalries and the massive financial incentives for companies possessing frontier models, and to fend off rogue users and criminals alike?
On misuse, I make no claim to a defensible perimeter: jailbreaks, open-weight fine-tuning, and ordinary fraud long predate AI, and regulation can reduce these harms but not eliminate them. I say the government is duty-bound to tackle these dilemmas. What we need is transparent, tightened rules, enforceable standards, and institutions accountable to the public rather than to shareholders. I do not claim governments can do this alone, or that they are immune to the incentives they are meant to check; regulatory capture is a real risk, and cross-border coordination is an unsolved problem. But naming the target is the precondition for designing the mechanism.
As news reports surface, governments are accelerating regulatory frameworks like the EU AI Act to tackle deepfakes, synthetic data limits, energy demands, and model alignment. This is not to say regulation is costless. Compliance burdens can slow the very reliability improvements we want, and poorly designed rules can entrench incumbents. The task is to regulate for safety and transparency without regulating pace — a distinction regulators have not always managed.
Yet beyond technological improvements, the utmost concern is to protect the overall integrity, economic stability, and governance of society. Labour market disruption has already begun; jobs are lost every day, and it will further creep up into white-collar and specialised fields such as accounting, law, and teaching. Inevitably, automation will reshape the entire workforce, but grave concerns over large-scale unemployment, economic inequality, and social instability must be addressed sooner rather than later.
Personally, I find AI an efficient research assistant and a good translator. Its speed for search results or translation work is just a split second, compared to a day of manual work. It helps me hasten my work and save time for more tasks. But AI does not do my work — it empowers me, not replaces me. My work is still my thinking, my architecture, and my own signature that AI cannot replace or replicate. Yet AI’s interactive intelligence and responsiveness make it an indispensable tool for many fieldworks.
Because AI has served me well in this limited, supervised way, I encourage my students to use the tools in research and translation — but not to copy without understanding and diluting the information. The four-pronged approach (URRU), with or without the aid of AI, works for me, and now works for them in all studies: Understanding — Remembering — Recognising — Use. Take a student using AI to translate a primary source: Understanding the passage’s argument, Remembering the key terms, Recognising where the machine translation distorts idiom, and Using the tool only after that sequence — that is URRU in practice, and it is the difference between a tool that teaches and a tool that substitutes.
As for the future of AI, I want two things that are often conflated. I want accuracy — outputs that are reliable in analysis and information, so that users can trust what they receive and URRU is genuinely facilitated. But I also want accountability — systems that detect, disclose, and correct their errors, and institutions that answer for them when they don’t. Zero error is a wish; zero unaccountable error is a goal worth regulating for.
