As AI adoption accelerates across industries, why trust, not technology, is emerging as the defining issue
The debate around artificial intelligence in assessment is often framed as a binary: either AI will change marks, or it will compromise the integrity of the qualification. Recent signals from Ofqual suggest a much more nuanced position on system design rather than capacity.
Ofqual’s official guidance on AI in marking was released a while ago, but the conversation hasn’t stopped yet. In fact, it accelerated. The rapid uptake of generative AI tools by learners, the increasing complexity of assessment models, especially in apprenticeships, and increased scrutiny of fairness and integrity are bringing this issue back into the spotlight.
In that context, Ofqual’s position feels more relevant now than when it was first published. This provides regulatory principles for AI, which the AI industry is only just beginning to take seriously.
At the heart of Ofqual’s approach are clear principles. AI cannot be the sole indicator of regulated qualifications. This is not to deny technology. Rather, it reflects a deeper concern about what makes an assessment reliable.
Rating systems do not work in isolation. They underpin a broader social contract between learners, health care providers, employers and the public. A qualification is only as valuable as the confidence you have in it. From this perspective, Ofqual’s emphasis on human judgement, transparency and fairness is a safeguard rather than a constraint.
What is becoming increasingly clear is that the question is no longer whether AI can reproduce aspects of scoring. In many cases, this is already possible. The more important issue is Decisions made within AI-supported systems can be understood, challenged, and trusted.
This is where the current tension lies. AI systems can achieve consistency at scale, but often struggle with explainability. While it can produce results in a way that aligns with established scoring practices, it does not necessarily provide the reasoning behind them. In a regulated environment, the gap becomes important not only technologically but also institutionally.
This is important because valuation decisions must:
- Defensible for learners
- Transparent to providers
- reliable for employers
- and be accountable to regulators.
Without this, even highly accurate systems risk eroding trust rather than strengthening it.
Ofqual’s position implicitly reframes the role of AI. Rather than acting as a replacement for human inspectors; Support layers within the rating system. Its strengths lie in the following areas:
- Identify patterns across large numbers of responses
- Quality assurance and moderation process support
- Provide suggestive or formative feedback in low-risk situations
This reconfiguration has important implications for awarding organizations and assessment providers. This means that the challenge is not just to deploy AI tools; Design systems that intentionally integrate human expertise and AI capabilities.
Human-involved models are often cited as a solution. However, this concept requires greater precision. Simply inserting a human reviewer at the end of an automated process does not address the underlying problem. In some cases, this can create a false sense of security, where surveillance exists in theory but not in practice.
A more meaningful question is where and how human judgment is applied within the system.
- At what point does a decision escalate?
- How are edge cases identified and handled?
- What level of confidence do I need before using automation?
- How will discrepancies between AI output and human judgment be resolved?
These are design questions, not technical questions. And it requires a level of systems thinking that goes beyond individual tools and solutions.
This distinction is particularly important in the field of further education and skills. Valuation models are evolving. For example, there is a clear shift towards more continuous, evidence-based assessment in practice. This increases reliability, but also increases complexity.
As the amount and variety of evidence increases, including written responses, observations, portfolios and, in some cases, multimedia submissions, the pressure on the evaluation system increases. Consistency becomes harder to achieve. Quality assurance requires more resources. There is more room for variation.
In this context, AI is not just an innovation, but increasingly a necessity. But just because it’s necessary doesn’t mean it requires careful design. If anything, it enhances it.
When used well, AI can support systems in the following ways:
- Reduce variation in how evidence is interpreted
- Uncover insights that may not be visible at scale
- Enable more timely and consistent feedback
If used poorly, they risk creating new forms of opacity, bias, and over-reliance on automation.
This is why Ofqual’s stance is important. Conversations are guided by the first principles of legitimacy, fairness, and trust. This means that evaluation is not just a technical process; Decision-driven systems that impact the real world.
Seen from this perspective, Ofqual’s position is more inviting than restrictive. This is because this field goes beyond experiments. intentional system design.
The next stage of AI in assessment will not be defined by what the technology can do. It is defined by the choices the field makes about how its technology is integrated, where human judgment is placed, how decisions are explained, and how trust is maintained.
The future of reputation will not be determined by whether AI is used, but by whether the systems we build continue to be trustworthy.
Kavitha Ravindran, Co-Founder and Chief Growth Officer, sAInaptic
