){kind=logo}
){kind=logo}
Bias in, bias out: Lessons from the AI training ground
Across our sectors, the language of integrity, inclusion and respect is firmly embedded in policy, strategy and leadership messaging. Yet for our diverse cohort, lived experience continues to diverge from stated intent. Organisational systems continue to reward familiarity and conformity, while inclusion is promoted but not operationalised.
By narrowing the inputs that shape credibility, opportunity, and authority, systems remain designed around biased inputs, belief in confidently wrong outcomes, informal power, and outdated assumptions, which set us back. Research on the implications of human trust in artificial intelligence highlights similar dynamics and offers clear lessons. AI systems trained on incomplete or biased data can generate outputs that are confidently wrong yet still acted upon. Where systems consistently fail to align words with actions, trust erodes not only in leadership, but in the organisation itself.
We value inclusion, yet our systems reliably produce exclusion. We say we trust systems, yet we design them in ways that undermine trust.
Policies and frameworks lose credibility when they are not operationalised, and confidence in institutions weakens accordingly. This presentation draws on early research findings from trust in AI systems within DEM to examine how organisational systems function in much the same way when commitments to inclusion are not reflected in everyday decisions.
With the introduction of positive duty obligations, leaders are required to actively prevent harm and inequality, shifting accountability from messaging and compliance to system design and governance. Leading for the future can apply lessons from the AI field to organisational leadership. Just as trustworthy AI depends on representative data, clear accountability, and ongoing oversight ensure organisations rely on systems that consistently recognise diverse capability. When systems are designed to learn from differences rather than filter them out, reliability improves, trust is strengthened, and inclusion becomes an operational powerhouse rather than an aspirational promise.
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)
)