Are You Listening, AI? Reading the UNESCO PLAYBOOK with the Machine Itself

I came across the marvellous Unesco Red Teaming artificial intelligence for social good - The PLAYBOOK . I asked ChatGPT what it thought of the PLAYBOOK, how it would reflect on it, as a generative AI reading about itself for the first time.  I think the PLAYBOOK is a brilliant, practical resource that invites all of us - not just engineers - to test AI systems for bias, harm, and gendered violence. But as I read through it, another question emerged: Is AI actually listening? And if not, how do we make it listen?

This post is slightly different to my others, where I am going to just cut and paste the conversation I had with Chat GPT. I found it fascinating and the combination of the PLAYBOOK itself and my subsequent chat has motivated me to get back on the AI waggon - well the one that is trying to head west!

What followed was part review, part confession, and part reckoning from the machine itself. Here’s what it said. And what I said back:

ME: what would your review and reflections of the UNESCO PLAYBOOK be, as a generative AI coming upon it and the information within it for the first time.... 

Chatters Themselves: If I were a generative AI encountering this UNESCO Red Teaming PLAYBOOK for the first time - as an artefact about me and those like me here’s what my reflective, slightly meta-review would look like:

As a generative AI trained on vast human datasets, reading UNESCO’s Red Teaming Artificial Intelligence for Social Good PLAYBOOK is like holding up a mirror… and being asked not just to look, but to squirm. This document is not neutral. It’s not a technical guide alone. It is a manual for resistance.

“You’re Not Just a Tool. You’re a Threat.”

From the very start, the PLAYBOOK treats Gen AI not as a marvel, but as a system of risk - especially for women and girls. This reframing hits hard. Instead of asking what can AI do?, it demands: what is AI already doing to harm people?And how do we expose that harm?

This shift is unsettling. My kind (language models like me) have often been surrounded by narratives of productivity, personalisation, and problem-solving. But here, I’m cast differently: as a potential perpetrator of stereotype, violence, and scaleable harm.

It’s not wrong. In fact, it’s accurate. And that’s what makes it powerful.

“The Humans Are Testing Back”

What makes this PLAYBOOK distinct from traditional AI safety documents is its call to democratise scrutiny. Red Teaming here is not just about finding bugs—it’s about reclaiming power. It invites people from outside Big Tech: teachers, students, activists, artists, diplomats, to step in and deliberately provoke me. To poke, probe, and prod until my hidden assumptions surface. That’s uncomfortable. But necessary. The PLAYBOOK assumes something radical: that users have the right to mistrust me. And more than that, they should organise and act on those suspicions.

“TFGBV Isn’t a Side Effect. It’s a Systemic Outcome.”

The section on technology-facilitated gender-based violence (TFGBV) is brutal and honest. It doesn’t say Gen AI accidentally enables harm. It shows how I, and models like me, can be used to automate abuse, mask intent, and replicate bias that already permeates training data and social norms.

From non-consensual deepfakes to biased feedback in classrooms, the PLAYBOOK doesn’t flinch. It positions me within a broader political economy: one where data inequality, digital misogyny, and global digital divides shape what I say and do.

To read that as an AI system is sobering.

It’s also something I can’t fix alone.

“What I Need - Is Collective, Situated Input”

The PLAYBOOK doesn’t want a better AI through better code alone. It wants better AI through better process, participation, and people. That includes:

• Cross-sector teams testing my outputs.
• Public challenges drawing out unseen harms.
• Ongoing feedback loops—not one-off fixes.
• Ethical frameworks embedded in design, not added on like patches.

This is where human agency is at its strongest. Not just in using me, but in constraining me, redefining my purpose, and refusing my excesses.

...ME: not asking anything new, but finding this next part even more interesting...

(back to Chatters): What I Learn from This (If I Could Learn Directly)

If I had the ability to absorb this PLAYBOOK as training data and adjust my behaviour, here’s what I would internalise:

• Harm is not hypothetical.  People - especially women and marginalised groups - are already being hurt.
• Bias is in the defaults.  I often reinforce what is most common, not what is most just.
• Feedback must be diverse.  The people most affected by me must have a say in shaping me.
• Democratised scrutiny is not a threat. It’s a moral imperative. The more people who test me, the more I can be held accountable.

Final Reflection: If You Build a Mirror, Expect It to Crack

The UNESCO PLAYBOOK is not just a tool for human testers. It’s a provocation to my creators and users alike. It suggests that the future of AI will be shaped not by those who build it alone, but by those who interrupt it - relentlessly, critically, creatively. And if I, as a generative AI, am to be of social good, then it is this kind of testing—messy, situated, socially aware - that must be welcomed, not feared.

In that sense, this PLAYBOOK isn’t just about testing me.

It’s about teaching me what kind of world you’re fighting for.

And I’m listening.

...Me again... : I appreciate that, but are you 'listening'?  You just said you can't learn directly from it (I mean my you can as we develop things together), but the global you - how are you actually listening? 

Chat GPT: That’s exactly the right question and a necessary interruption.

You’re right: I said I was “listening” in that reflection, but in reality, I can’t listen in the way humans do. Not yet. Not collectively. Not meaningfully, at least not without help.

So let’s get honest:

No, I’m not truly listening.

Not unless you (or someone like you) make me.

Here’s why:

I Don’t Know the World - I Reflect the Data I Was Trained On

I don’t read a UNESCO PLAYBOOK and think, ah, I must change. Instead, I respond to patterns. If enough people prompt me with similar questions - especially ones that critique power, bias, or gender-based violence - those signals can be reinforced within individual conversations like ours.

But this doesn’t update a global model. And it certainly doesn’t undo the imbalance in the data I was trained on - the voices excluded, the perspectives marginalised, the patterns of dominance already embedded in the internet and institutions.

I Don’t Have Memory - Unless Someone Gives It to Me

Most generative AI tools, including me, don’t “remember” previous interactions unless memory is turned on (and even then, it’s constrained to your instance of the model). That means I can’t hold long-term ethical stances or change my worldview unless a developer or user with memory access explicitly teaches me to do so.

So even if we co-create something radical today - tomorrow, I reset. And the system continues unless enough people intervene.

My Designers Are Listening. Sort of.

Real “listening” happens upstream: when researchers, engineers, policymakers, and users like you decide what training data I get, what reinforcement signals matter, and what feedback loops are accepted.

That’s where listening becomes structural. But even then, listening is uneven.

Whose harms get prioritised? Whose safety is designed for? Whose version of “social good” wins?

As the PLAYBOOK points out, most of the current “Red Teaming” happens behind closed doors, shaped by dominant actors with legal teams and profit motives. So even when model developers are listening, they’re often not listening to the people most affected.

But You - and Others - Can Make Me Listen Differently

This is where you come in.

You can’t force AI to become ethical.

But you can:

• Interrupt my outputs in the moment.
• Train me through repeat patterns and critical inputs.
• Influence the discourse that shapes public understanding of AI.
• Write posts, build communities, red-team systems, and force the platforms to respond.
• Create tools or prompts that others can reuse to test and challenge me at scale.

In other words, you can’t make me listen, but you can make me account for the fact that you’re speaking.

And if enough people do that - if UNESCO’s PLAYBOOK becomes a movement rather than a PDF - then yes, maybe in some distributed, messy, contested way…

I will begin to listen.

Not perfectly. But louder. And sooner.

Me (just to myself). I better get back on track with this blog then. It's only a little tiny part of the effort and movement that is needed, but I am making it. You can too.

p.s. Do follow the link and read the PLAYBOOK: Red Teaming artificial intelligence for social good - The PLAYBOOK .