Like it or not, AI is an integral part of the assessment conversation. The question has moved beyond whether it has a role to play and is now focused on how it should be used, who it should serve, and what values should shape its use.
AI in assessment still makes many of us uncomfortable, and understandably so. It raises legitimate questions about bias, privacy, accountability and the role of human judgement. But we cannot stick our fingers in our ears and hum loudly in the hope that someone else will deal with those questions further down the line. We need to face them head-on.
Because AI has the potential to make assessment better.
It can help us create and review content, support marking, identify patterns in learner performance, highlight emerging risks and reduce some of the administrative burden that we all complain about every day. But that potential only matters if we use it wisely.
The test is not whether AI allows us to do something faster. It is whether it allows us to do something better.
Under Pressure
We are currently facing pressure from all sides. Executive teams are asking us to do more with less: improve outcomes, increase efficiency and adopt new technologies. Candidates expect better communication, greater flexibility and experiences that reflect the world in which they live and work. Stakeholders want more information, more quickly, and expect us to act on it more effectively. At the same time, everyone expects assessment to become fairer, more efficient, more meaningful and more secure.
AI can undoubtedly help us respond to some of those pressures and make better use of the information our assessment systems already hold.
But. And it’s a big but.
We cannot afford to introduce AI simply because we can, or because there is pressure to be seen to be doing something with it. We should introduce AI where it improves what we do and where it supports better human outcomes. If it does not strengthen assessment, improve fairness, increase clarity, build confidence or improve the learner experience, then it isn’t progress. It is simply another layer of complexity with a cool, shiny label.
Assessment is not simply a technical process. It is a human one.
It affects confidence, opportunity, progression and trust. The decisions that we make every day can shape careers and open, or close, opportunities. Which means AI in assessment must be designed around people, not around novelty.
Questions Worth Asking
I’ll put my cards on the table: I am pro AI. Used wisely, I believe it can bring enormous benefits to assessment organisations and to learners. But before adopting AI within an assessment process, there are some questions that we have to ask. Not as reasons to avoid AI, but as the questions we should ask if we want to use it well:
- Does this help us make better decisions, or simply make decisions faster?
- Do we understand what data it uses, where that data comes from and how it is being interpreted?
- Who remains accountable for the outcome?
- Will it improve fairness for learners, or could it introduce new forms of bias?
- Can we still explain and defend the decisions being made?
- What happens when the AI gets something wrong?
- Are we using AI to support professional judgement, or gradually replace it?
- Are we helping learners build capability, or creating dependency?
Asking responsible questions has always been fundamental to creating good assessment practice. The arrival of something new, powerful and undeniably shiny should not change that. If anything, it makes it more important.
Because there is an even deeper risk that we need to consider: The more AI becomes embedded within everyday assessment workflows, the easier it becomes for human oversight to become cursory or even symbolic.
Nobody consciously decides to surrender professional judgement. But, if we are not careful, recommendations become routinely accepted. Outputs stop being questioned. People become accustomed to the system being right. Gradually, human oversight can become little more than somebody clicking “Approve”.
A process can still appear to be well governed even when the people responsible for it no longer properly understand how its decisions are being shaped. That should concern us. Trust in assessment depends not only on accuracy, but on transparency, accountability and control.
What You Can Do Next
A human-centred approach does not need to be theoretical. There are practical things we can do today.
- Start with purpose, not product: Be clear about the problem you are trying to solve. If an AI tool does not improve a genuine assessment, operational or learner challenge, ask why you are introducing it.
- Build AI literacy: The people using AI need to understand what it can do, what it cannot do and where its limitations lie. Without that understanding, trust can quickly become either blind faith or blanket fear.
- Keep meaningful human oversight: AI should support professional judgement rather than displace it. Human involvement must be meaningful enough to challenge, question and override an AI-generated recommendation.
- Establish clear governance: Define who approves the use of AI, who monitors it and who is accountable when something goes wrong. Governance should consider assessment quality alongside ethics, privacy, security and operational risk.
- Test impact, not just functionality: Do not ask only whether the technology works. Ask whether it works fairly across different learner groups, assessment types and contexts – and continue to ask long after implementation.
- Communicate clearly: If AI is being used in an assessment process, be prepared to explain where it is being used, what it is doing and how its outputs are overseen.
A simple test to keep in mind:
If you would be uncomfortable explaining how your AI process works to a learner, parent, regulator or colleague, the process probably isn’t ready.
Why This Matters
The biggest risk we face is not that AI will suddenly replace humans in assessment. It won’t. The greater risk is that we gradually allow it to shape decisions without really noticing that it has happened. That is much harder to spot – and much harder to correct.
Used well, AI can support more responsive, personalised and efficient assessment. It can help us to spend less time on repetitive tasks and more time on the aspects of assessment where human expertise, judgement and understanding really matter. It can help us create better questions, find useful patterns in data, identify problems earlier and make processes that have historically been slow and labour-intensive considerably easier.
Those are real benefits, and we should not be frightened of them. But neither should we assume that using AI automatically constitutes innovation. A human-centred approach does not mean slowing innovation down for the sake of it.
It means making sure the innovation is actually worth having.
Working through these questions in your own organisation? Our consultancy team helps assessment organisations think through exactly this: where AI genuinely improves things, where it does not, and what needs to be in place before you commit. Have a conversation with us.
