AI screens the candidates and translates the contract. Who answers when it is wrong?

There is a moment you may already have had. A decision that used to be yours, who to interview, how to phrase the contract in another language, which market to back, was quietly made by a model, and you could not fully explain how it reached the answer. It was probably a good answer. It was faster than you could have managed. And still a small unease stayed behind, because when a decision like that goes wrong across a border or a culture, the tool does not answer for it. You do.

Here is the reassurance underneath the unease. A 2024 University of the Sunshine Coast review of 37 studies on AI in international business is refreshingly plain about the limit: across hiring, translation, market entry and management, AI augments human work rather than replacing it, and the parts it cannot take over are judgement, accountability and cultural wisdom. Its clearest warning is a real case. Amazon built a recruiting AI that learned from a decade of largely male CVs, and it quietly down-ranked women until the company scrapped it. The tool handled the volume. The bias, and the responsibility, stayed human.

What can AI genuinely take off your plate, and what can't it?

It is genuinely good at volume and speed. The review describes AI screening applicants, translating documents, drafting business plans and profiling competitors in minutes, and used well it relieves people of mundane work so they can spend attention where attention matters. On recruitment specifically, the review is clear that AI is unlikely to replace recruiters; it is designed to take the drudgery so humans can improve the decision.

What it cannot do is carry the consequence. The review flags three limits worth memorising. Bias hides in the training data, as Amazon found. Accountability for AI translation errors in legal contracts is unresolved, so it recommends a human back-translate anything that carries legal weight. And there are still no global standards governing how AI handles employee data across borders. In every case the technology is a decision-support tool, and the decision remains a human act. This is the same truth behind why you cannot out-hustle the machine: the edge is not doing more, it is judging better.

Why does judgement matter more, not less, as AI does more?

Because the cost of a confident wrong answer scales with the tool that produces it. A biased human recruiter harms a handful of candidates; a biased hiring model can quietly do it to thousands before anyone notices. The review organises AI's challenges into three levels, and each is really a place where human judgement has to stay in the loop. At the employee level: fairness, dignity, and the reskilling that keeps people with you rather than drifting into apathy. At the country level: data privacy and regulation that differ across every border you cross. At the organisational level: accountability, and the honesty to counteract privacy breaches and what the review bluntly calls dehumanisation.

Across cultures the point sharpens further. A model can translate a language fluently and still miss what a phrase means to the person reading it. Fluency is not understanding, and the gap between them is exactly where a human who knows the context earns their place. It is the deeper reason so many AI strategies quietly fail: they optimise the output and forget the human meaning it has to land in.

How do leaders keep the human in the loop on purpose?

Deliberately, because it will not happen on its own once the tools are fast and convenient. The review points to a handful of practices worth making standard.

1. Keep accountability with a person, always. Treat AI as decision-support, and make sure a named human owns the sign-off on anything consequential.
2. Back-translate the high-stakes work. Have a human verify AI on contracts, cross-cultural messages, and anything difficult to reverse. The review recommends exactly this for legal documents.
3. Audit for inherited bias. Ask what a model learned from and who it might quietly exclude. The Amazon case was not a technical failure so much as an unexamined history fed back as fact.
4. Protect dignity as you deploy. The review warns of privacy breaches, excessive control and dehumanisation, and calls on leaders to counteract them with what it terms digital mindfulness. Deploy the tool without diminishing the people.
5. Invest in judgement, not only tools. Train people to question AI output, not merely to operate it. The scarce, valuable skill is now knowing when the confident answer is wrong.

What does this ask of me as a leader?

It asks you to be clear about the division of labour between the machine and the human, and to hold the human line where it counts. Delegate the volume without hesitation: the screening, the drafting, the first-pass translation. Keep for yourself, and for the people you trust, the parts the review says cannot be handed over: the fairness of a decision, its meaning in another culture, and the accountability for what happens next.

Technology's highest purpose is to serve human life, and in a global business that shows up in something very concrete: AI that makes your people faster and your decisions fairer, rather than quieter and less accountable. The leaders who come through this well are not the ones who automate the most. They are the ones who keep hold of what only a human can do, and let the machine carry everything else.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.