The Algorithm Doesn’t Lie. That’s Exactly the Problem
By Janet Thompson, SVP, Material
I’ve given and attended enough briefings to know how this goes. Someone puts up a slide. There’s a neural network graphic, maybe a gradient. The word ‘scalable’ appears at least 4 times. And somewhere in the deck, there’s a sentence that reads: ‘AI removes human bias from the equation.’
Everyone in the room nods silently. Either unaware or unwilling to admit that algorithmic bias powerfully reinforces and legitimizes existing inequities. It is as biased as you get.
Consider:
+60% of websites are in English, only 20% globally, speak English fluently
<1% of content is in African languages, yet Africans are ~17% of the global population
One of the most-used AI training sources, Wikipedia, draws 80–90% of its content from Europe and North America
Only ~15–20% of Wikipedia biographies are about women
If we accept AI outputs as fact, we risk allowing historical bias to define our future. This is critical moment where words and actions matter. Giving AI editorial control isn’t just lazy, it’s dangerous.
I get why it’s appealing. Clients want certainty. Everyone wants to limit liability. ‘The data decided’ is a clean exit from messy conversations. But data reflects the biases of those who create, fund, and distribute it, not objective reality. Algorithms amplify these historical inequities, turning them into system-wide decisions.
Accepting this false reality is not just pessimistic, it’s a brand risk, a cultural risk, and increasingly, a regulatory risk.
The plot thickens.
AI no longer trains on just human-generated data. It’s training on AI generated data. The outputs of today become the inputs of tomorrow. Which means bias doesn’t plateau. It compounds. What begins as skew becomes pattern. Pattern becomes norm. Norm gets laundered into objective insight. And once something achieves the status of objective, it becomes nearly impossible to challenge in a client meeting or a courtroom.
We stop asking if it’s right. We start defending why it is.
So, how is the industry responded to this threat?
‘We need more diverse data.’ Sure. But data collection is itself a power dynamic. You don’t solve representation by extracting more from communities that have historically been underserved. That’s not inclusion. That’s better-informed extraction with a DEI footnote.
The real question isn’t who’s in the dataset. It’s who decides what gets collected, how it’s interpreted, and what it ultimately gets used to justify. Until that changes, more data doesn’t fix the system. It just makes the bias more statistically confident. Which, if you think about it, is worse.
So, what does ‘doing it right’ look like?
A few thoughts worth pushing for inside your agency, and with clients:
Audit the Inputs, Not Just the Outputs.
Most agencies review outputs (creative, targeting, messaging). That’s too late. Responsible stewardship means interrogating:
What datasets are being used
What’s excluded (geographies, languages, income brackets, behaviors)
What proxies are quietly standing in for sensitive attributes
Bias isn’t a glitch in the output. It’s a design choice upstream.
Build ‘Bias Stress Testing’ Into Go-To-Market.
Before anything launches, pressure-test it like you would a financial model.
Ask:
Who does this system systematically underserve?
Who does it over-prioritize?
What happens at the edges (non-majority users, atypical behaviors)?
Make this as standard as QA or brand review. Not a workshop. A gate.
Redefine Performance Metrics.
AI systems optimize for:
CTR
Conversion
Efficiency
All of which can reward bias. You, need to expand the definition of performance to include:
Distribution fairness (who is reached vs excluded)
Outcome parity (who benefits vs who is penalized)
Long-term brand trust impact
If bias improves short-term performance, the system will choose it every time unless you redefine what ‘winning’ means.
Diversify Data Partnerships, Not Just Teams.
Agencies talk a lot about diverse talent. Less about diverse data sources.
Push clients and partners to:
Incorporate non-Western datasets
Include multilingual and non-dominant market inputs
Balance commercial data with cultural/contextual data
If the data is narrow, the thinking will be too no matter who’s in the room.
Shift the Narrative: From Ethics to Advantage
Responsible AI sounds like a constraint. Reframe it as:
Better data = better targeting
Broader inclusion = expanded market opportunity
Reduced bias = stronger long-term performance
This is not about being good. It’s about being right and winning because of it.
Build accountability into the system.
Transparency reports don’t shift incentives. Real consequences do. The EU AI Act is a start. Watch it carefully and get ahead of it, because your global clients already are.
Create a ‘Bias Escalation Path’ (Like Brand Safety)
Brand safety has escalation protocols. Bias should too.
Who flags it?
Who investigates it?
Who has authority to pause spend, retrain models, or kill work?
If there’s no escalation path, bias becomes everyone’s problem, and therefore no one’s responsibility.
The real threat isn’t a rogue algorithm. It’s the perfectly functioning one.
The popular fear is AI gone rogue; autonomous, uncontrollable, dramatic. That makes a great doomsday movie but it’s not the actual risk profile we’re operating in. The real danger comes from the algorithm that operates precisely as intended, automating and entrenching bias while presenting it as fact.
No alerts. No visible failure mode. Just millions of compounding small decisions, all pointing in the same direction, quietly determining who gets access, who gets seen, who gets the offer, and who gets filtered out.
We still have a window. It’s narrowing.
The systems being built right now will define the decision-making infrastructure of the next decade. Once embedded at scale, it won’t just influence outcomes; it will determine which outcomes are even imaginable.
Our industry has always had a complicated relationship with the question of who we’re really building for. AI doesn’t resolve that question. It amplifies it, accelerates it, and eventually makes it very hard to walk back.
Start here: design systems that reflect the full complexity of human reality, not a flattened, convenient version of it. Because if we don’t do this deliberately, we are not building intelligence, we are industrializing bias. And worse, we are disguising it as objectivity.