Field Note · 002 iamhonee ↗

CWI · Amsterdam Science Park · 7 May 2026

The room where bias got a name

Symposium — Understanding Bias in AI A wide lens Industry seats: < 10
Scroll to enter
01 / 06
The CWI building entrance in Amsterdam Science Park, red CWI sign against brick and a clear blue sky.
Centrum Wiskunde & Informatica — the entrance I'd only ever walked past
A long shadow cast on patterned paving stones while walking through the Science Park toward the building.
Science Park, just beside the University of Amsterdam campus
A hand holding a CWI lanyard badge reading 'Understanding Bias in AI, 7 May 2026 — Honeylyn Tutor'.
One of a limited number of slots — and I got one
A projection screen reading 'Understanding Bias in AI — a wide lens, Symposium at CWI, May 7, 2026'.
Understanding Bias in AI — a wide lens
A packed room of researchers facing a screen showing an interface mockup during a talk.
Researchers, professors, PhD students — every talk rooted in their own work
Wide view of the symposium room with rows of grey chairs and the title slide on screen.
Where research and real policy implications overlapped

01 — Arrival

I'd passed this place before.

The symposium was held at CWI in Amsterdam Science Park, just beside the University of Amsterdam campus. I had walked by this area before — so finally getting inside felt like crossing a line I'd only ever looked at.

02 — The walk in

A different crowd this time.

The room was packed with researchers, professors and PhD students working deeply on AI across different domains. People like me — from industry — were probably fewer than ten.

Slots were limited too. I was happy I got one.

03 — The badge

Closer to this than I knew.

The topic was Understanding Bias in AI, and every talk was rooted in active or past research. It's something I've worked on seriously since last year.

But sitting there, I realized I'd actually been close to this space since 2019.

04 — The lens

Explainability was always about fairness.

The ML models I worked on back then — especially the explainability layer around them — were fundamentally about fairness, by mitigating bias.

It just wasn't mainstream yet, and wasn't always framed as AI. But that work quietly gave me a lens: models can be used to push back against bias. It is possible. There are methods that help.

⚖️ Bias mitigation wasn't a buzzword back then. It was just good modeling.

05 — One talk in particular

A news company, and a choice.

One research example stayed with me: how a news company optimized its ML models to recommend content to its audience.

There's a balance that shouldn't be overlooked — and there was even a formula for how to debias it. But sometimes, or most of the time, companies optimize for clicks.

06 — The other seat

The person beside me.

He turned out to be a data scientist working for the 🇳🇱 government on the EU AI Act — his research circled the same topic. He introduced himself and shared papers he thought were relevant to my work.

🙏 Thankfully, Ademar trained me how to read research papers and actually get through understanding them.

Imagine how dangerous that becomes when no one reviews the news being fed to us.

Topics that trigger strong emotions get pushed further — because we engage, react, get outraged, join the conversation. Confirmation bias compounds until everything starts to feel like an echo chamber of your own leaning.

😤 🤬 🗯️ 🙈 🌪️ 📲

The balance that shouldn't be overlooked

Behind the feed, there's a working model — and a knob.

score = relevance(item, you)
    − λ · engagement_pull(item)

A recommender doesn't have to chase outrage. λ is a choice — turn it up and you debias the feed; set it to zero and you optimize for clicks. The math was never the hard part. The incentive is.