Did You Just Opt Out of Innovation?

Did You Just Opt Out of Innovation?

Think about the last time you clicked "reject all cookies." Or unsubscribed from a mailing list. Or declined to share your data with an app. It felt like the right call, a small act of control in a world that constantly asks for access.

Most of us have been trained to think of opting out as the safe, sensible choice. The privacy-conscious move. But when it comes to AI, opting out carries a cost that rarely gets mentioned, and it is not the one you might expect.

AI systems are only as good as the data they learn from. When you opt out, your face, your voice, your language patterns, your health history, your context are absent from that learning. The system does not pause. It does not wait. It trains on whoever stayed in, and it gets better for them. Opting out does not keep you safe from AI. It just means AI was not built with you in mind.

And if you think that does not matter, that you can simply choose not to use AI, it is worth knowing that AI is already making decisions about you. In hiring. In healthcare. In credit assessments and government services. The question is not whether AI affects your life. It is whether the people who built it had anyone like you in the room.

What We Mean When We Talk About AI Sovereignty

In Australia, "AI sovereignty" has become a serious policy conversation. It covers roughly three things: data sovereignty (who owns and controls data about Australians), compute sovereignty (whether the infrastructure running AI sits on Australian soil), and model sovereignty (whether Australian institutions can build and run their own models rather than depending entirely on foreign ones).

These all matter. But they are institutional framings. They are questions about who controls the system, not about whether the system actually works for you.

The missing piece is personal sovereignty. It means your agency over how AI systems represent, affect, and include you. Right now, the answer most people are given for that is consent. Opt in, or opt out.

It sounds like empowerment. I want to argue it isn't. I am not about to tell you the answer is to opt in to everything. The problem runs deeper than which box you tick.

The Two Exits, and Why Both Are Bad

When you are given an opt-in or opt-out choice around your data and AI systems, you are being offered a decision with two genuinely bad outcomes.

You have done this before, even if you did not think of it in these terms. Every time you click "reject all cookies," decline a marketing list under CAN-SPAM, or navigate a GDPR consent banner, you are exercising a version of this right. These frameworks were a genuine step forward. They gave individuals visibility and a mechanism to say no. But notice what happens after you click reject: the website still loads, the company still operates, the system still functions. Your absence changes nothing about how it was built.

That is the limit of consent as a tool. It was designed to protect you from a system. It was never designed to make the system work for you.

With AI, the stakes of that distinction are higher. If you opt in, your data enters the system, but you had little say in how it was collected, how it was labelled, or what assumptions were baked in before your data ever arrived. Representation is not the same as accurate representation.

If you opt out, you protected your privacy, but the system now works less well for people like you. Your face, your accent, your medical history, your context are all less legible to AI than the people who stayed in. The system was built, the defaults were set, and the model learned from whoever was in the data. You opted out of that process, and the process continued without you.

There is also no clean undo. If your data was used to train a model, requesting its removal from a database does not remove what the model already learned from it. The model has been shaped. This is an open legal and technical question that regulators are still working through, and no existing framework has resolved it.

This is not a neutral outcome. Research consistently finds disparities in AI outcomes, accessibility, and representation among diverse groups, stemming from biased data sources and a lack of representation in training datasets. Facial recognition that performs worse on darker skin tones. Voice recognition that struggles with non-standard accents. Medical AI trained predominantly on data from Western, male, and younger populations. Hiring algorithms that replay the underrepresentation already embedded in historical records. These biases affect how people are seen, categorised, and treated by AI systems - in job applications, in healthcare, in interactions with government services.

And here is the harder part: the people most likely to opt out are often the people who can least afford to be invisible. Research in biomedical AI has found that consent rules are neutral on their face but have discriminatory impacts, because willingness to consent varies by race, ethnicity, gender, and socioeconomic status. The most vulnerable groups have the most rational reasons to withhold consent. Migrants navigating complex legal situations, communities with legitimate reasons to distrust institutions, people whose data has historically been used against them rather than for them. Their reasons are valid. But the consequence is that distrust compounds underrepresentation, quietly, over time.

Existing privacy law does not bridge this gap. GDPR and Australia's Privacy Act protect individuals from misuse of their data. They do not give communities the right to be accurately represented. You can request erasure of your own data, but you cannot erase the pattern a model learned from your entire community. There is no collective mechanism to say "the way our community appears in this system is wrong." These frameworks give you the right to leave the room. They do not give you the right to have been invited in properly to begin with.

This is not an individual failure. It is a design flaw.

We Have Solved This Kind of Problem Before

This is not the first time we have built a system around the wrong default and called it neutral. For decades, women were routinely excluded from randomised clinical trials. Not because anyone made a deliberate decision to harm them, but because the male body was treated as the standard, and no one was required to think otherwise.

The consequences were measurable. A 2022 analysis of over 300,000 clinical trial participants found that on average only 41.2% were female, despite women making up roughly half the population. Women experience adverse drug effects at nearly twice the rate of men, a disparity researchers now attribute in significant part to their historical exclusion from the trials that set the safety standards. The harm was not visible until someone thought to look for it. And we only looked because enough people named the cost out loud.

It was not until 1993 that the US Congress passed a law requiring the inclusion of women in NIH-sponsored clinical trials. The fix was not better consent options. It was a requirement built into the design of the system itself — an obligation to include, not just a right to opt in.

Some communities are already building equivalents for AI. The AIATSIS CARE Principles — developed for Indigenous data governance — offer a framework built around Collective Benefit, Authority to Control, Responsibility, and Ethics. Rather than asking individuals to make isolated yes-or-no decisions about their data, it treats data as a community asset and builds accountability into how systems are designed from the start. Te Mana Raraunga, the Māori Data Sovereignty Network, operates on similar principles.

These are not abstract ideals. They are working frameworks that locate the obligation where it belongs: with the people building systems, not the people deciding whether to trust them.

What You Can Actually Do

The question I would invite you to take into any room — any meeting, any briefing, any procurement conversation — is this:

Who is in your training data, and what happens to the people who aren't?

If you work in AI or technology, that question points toward auditing your training data for representation gaps. Not just checking whether a consent box was ticked, but asking whose experience shaped what the model learned and who is missing from that picture.

If you work in policy or government, it points toward pushing for representation requirements alongside consent requirements. Australia's Privacy Act reforms are in progress. They are worth watching, and worth engaging with — and worth asking whether they go far enough on the question of inclusion, not just protection.

If you are a citizen, a patient, a worker whose life is increasingly shaped by automated decisions, the question is still yours to ask. Of your hospital. Your insurer. Your employer. Your local council. Most of them will not have a ready answer. That is useful information.

We named the cost in medicine, eventually. It took too long, and the harm in the interim was real and measurable. AI is moving faster than medicine ever did. The window to get this right is not indefinite.

Opting out is not sovereignty. It is just a quieter way of being left behind.