What Parallax Actually Means and Why It Matters for AI

In astronomy, parallax is the apparent shift in the position of an object when viewed from two different vantage points. The object has not moved. What has changed is the observer's position, and that shift reveals something true about the object's location that a single observation could never establish.

This is not a metaphor. It is the operating principle behind the assessment methodology I have spent a decade developing.

When an organisation evaluates an AI system before deployment, they typically do so from one of a small number of positions: technical review, legal compliance, or executive risk appetite. Each of these is legitimate. None of them is sufficient.

A technical review will tell you whether the system performs as specified. It will not tell you whether the specification was correct in the first place, or whether the problem the system was designed to solve is the problem the organisation actually has.

A legal compliance review will tell you whether the deployment satisfies current regulatory requirements. It will not tell you whether those requirements are adequate for the specific context, or whether the system creates harms that existing frameworks have not yet named.

An executive risk review will tell you what the organisation is prepared to accept. It will not tell you whether that acceptance is grounded in accurate understanding of what the system actually does.

Each of these reviews produces a finding that is true from where it stands. The problem is that truth from one vantage point is not the same as truth.

The Parallax methodology is a belief-reality divergence detection mechanism. Its purpose is to find the gap between what an organisation believes about its technology and what is actually present, before deployment begins.

It does this by requiring that any technology decision be examined from multiple, structurally distinct vantage points simultaneously. Not sequentially. Not as a checklist of boxes to tick in order. Simultaneously, because the gaps between vantage points are where the real findings live.

Those gaps are not errors. They are information. They tell you where the organisation's model of its own technology diverges from what the technology actually is.

A checklist is a list of things that should be present. It tells you whether you have done the things on the list. It does not tell you whether the list is complete, whether the things on the list are the right things for this specific context, or whether the absence of a finding means the risk is absent or merely undetected.

Parallax is not a checklist. It is a structured process for generating findings that a checklist cannot produce, because it is specifically designed to surface what is not visible from any single position.

The 35 international governance frameworks that Alt-TAB screens against are not a checklist either. They are a set of vantage points, each representing a distinct analytical tradition, a different set of values, a different model of what technology does to people and institutions. Running a technology decision through all of them simultaneously is not about compliance. It is about triangulation.

Every Alt-TAB assessment includes mandatory screening for three domains regardless of the technology type: technology-facilitated gender-based violence, child safety, and cybersecurity.

This is not because every technology is a risk in these domains. It is because these are the domains where the consequences of a missed finding are most severe, and where the gap between what an organisation believes about its technology and what the technology actually does is most likely to be consequential.

An organisation deploying a customer service chatbot does not typically think of itself as making a decision with implications for gender-based violence. But if that chatbot is deployed in a context where it will interact with people in crisis, or where it will be used to manage sensitive personal information, or where its outputs will be used to make decisions about access to services, the question is not academic.

The mandatory screens exist because the question should always be asked, even when the answer is expected to be straightforward.

When an organisation engages Xallarap for a pre-deployment assessment, the Parallax methodology is the engine of that work. It is not a report template. It is not a set of interview questions. It is a structured process for generating findings that are defensible, evidence-based, and traceable back to source.

The findings are not opinions. They are the product of a methodology that can be demonstrated, reproduced, and explained. That matters because the organisations that engage us are often making decisions that will be scrutinised by regulators, by the communities they serve, by their own governance structures.

When those decisions are scrutinised, "we did an assessment" is not sufficient. "We applied a methodology grounded in 35 international governance frameworks, with mandatory screening for the highest-consequence domains, and here is what it found" is a different kind of answer.

That is what Parallax is for.

Alt-TAB, the free pre-deployment AI ethics assessment tool built on the Parallax methodology, is available at alttab.afk.org.au.