I've spent years trying to show leaders their blind spots, turns out, we needed AI to hold up the mirror.

 A positive AI article...

The attached research is interesting for anyone interested in AI and bias, in the world and at work. So all of us, I guess. These researchers have found a way to use the Stereotype Content Model (SCM) as a simplified and highly effective way of spotting bias in LLM and mitigating for it, so efficiently reducing bias in LLM outputs. What’s exciting about their method is its simplicity. Rather than de-biasing by considering every possible bias, all its intersectional complexities, tackling stereotypes one by one, it gets to the root emotion of bias and scans for that. It’s simpler and has been found to be more effective than existing methods.

 Instead of playing bias whack-a-mole, SCM gets to the emotional root that drives all stereotyping.

That basic instinct that wants to judge how we should respond to someone new from very little information. Because it’s basic, it communicates in emotion, not logic and the language we use (and subsequent actions) reflects these emotions - so highlighting the likelihood of potentially biased actions.

Simply put, when we start feeling envy, pity, contempt or unbridled admiration for others, it's a sign our stereotyping system has kicked in, and these emotions cause us to speak and act towards others in predictable ways.

This resonates with me, since using the SCM for my MSc research years ago, I've used this tool as a helpful way to support leaders in spotting and highlighting potential stereotyping issues. For me, it’s a form of raw emotional intelligence to understand and mitigate these emotions in ourselves and others. It’s very simple and it’s always surprised me that it hasn't found more traction in the workplace to truly weed out stereotyping and bias in ourselves and our teams. Maybe AI can be the objective mirror - maybe it can show us what we couldn't see in ourselves.

When the algorithm clearly amplifies our bias it’s data. When it’s human bias, it’s personal.

And the LLM's ability to be debugged? Well machines don't have egos, we do.

Let's be honest - getting humans to recognise their own bias is like trying to show someone their blind spot while they're driving. People get defensive, uncomfortable, resistant. But AI bias? That's just data that needs fixing. No ego involved.

This research might have cracked the code on bias detection precisely because it sidesteps human psychological defences. The SCM has been shown to be simple and effective at debiasing LLMs, we can also use it to spot patterns in organisations. HR can spot departments where meeting minutes, or performance reviews show the patterns of emotional bias within. Everyone seems to be creating AI tools right now, a way to scan for this emotion in key organisational documents or minutes will no doubt be on the way. It’s only fair that we help everyone in society, but especially leaders with power, to understand the SCM and their own (and others) implicit bias.  We can all tune in our Emotional Intelligence and not get caught out by the machines.

In my view the SCM research is essentially the behavioural economics equivalent for leadership bias - showing that we're not rational decision-makers, we're emotional pattern-matchers operating on warmth/competence snap judgments.

My mission is helping organisations deepen their understanding of humanity in leadership - moving beyond the rational leader myth to embrace the emotional intelligence needed for truly effective leadership. If you're interested in building tools and approaches that help leaders see and work with their human psychology rather than against it, let's connect.

But this is bigger than any one person can tackle alone. Get in contact if you'd like to build something around this together to make a difference in the world.  The window to get ahead of AI bias while helping leaders do the same is now.

https://aclanthology.org/2022.nlpcss-1.23.pdf