Translation

What it is

Translation is the capability of moving accurately between the register, vocabulary, and model of one context and another — most urgently between the technical world where AI is built and the human world where it has to work. It is not primarily a communication skill, though communication is part of it. It is an accuracy discipline: ensuring that what is known in one context actually arrives in another, rather than a version of it that has been flattened, oversimplified, or distorted to fit the receiving context's expectations.

In the AI age, translation operates in both directions and both matter. Downward translation — from technical to non-technical — determines whether the people with authority to make decisions about AI have an accurate model of what they are deciding. Upward translation — from domain expert to AI builder — determines whether the systems being built have an accurate model of the problem they are solving. Both failures are common and both are expensive. The most common failure mode in AI adoption is not the technology failing to work; it is the translation failing to preserve what matters most about the problem.

Translation is also the capability most directly at risk from AI itself. AI can generate fluent, confident explanations of technical concepts in plain language — which creates the illusion of translation without the accuracy of it. A technically fluent summary that is missing the three assumptions that would change the conclusion is not a translation; it is a distortion that looks like one. Translation at the Guide level means being able to detect those distortions in AI-generated explanations and correct them before they reach the people who will act on them.

Why it matters now

The single largest source of misaligned AI systems is not technical failure — it is translation failure at the point where domain knowledge becomes training data, requirements, or evaluation criteria. The people who know what the problem actually requires speak one language. The people who build the system that addresses it speak another. Without genuine translation at that boundary, you get a system that performs well on the metric while missing the point.

As AI becomes more capable, the translation problem becomes more acute, not less. More capable AI can produce more convincing output in more domains — which means more opportunities for confident-sounding translation failures that are hard to detect without deep expertise in the receiving domain. Translation capability is what allows people who are not AI builders to remain genuine agents in the design of AI systems that affect their domain, rather than passive subjects of decisions made by people who speak a different language.

The five levels

Every AgenticU spine progresses through five levels: Orient → Operate → Build → Transform → Guide. Each level has a specific competency description and a defined assessment type that proves mastery.

1. L1 Orient Quiz

   Translate AI concepts and outputs for non-technical stakeholders

   Recognise the terrain and establish your first practice.

   You can explain what a model is doing, why it produces the outputs it does, and what its failure modes are to someone with no technical background — in a way that preserves the accuracy of the explanation rather than trading accuracy for accessibility. You can identify where a common AI explanation is misleading and articulate what it gets wrong without resorting to technical jargon.

2. L2 Operate Project

   Bridge the technical-nontechnical gap reliably throughout a project

   Maintain the capability through a complete project.

   Through a complete project involving AI, you have maintained accurate translation in both directions — from technical to stakeholder and from domain knowledge to system requirements. You can show a decision where your translation work changed the outcome: a requirement that was specified correctly because you caught a translation failure, or a decision that was made with accurate information because you corrected a distorted explanation before it reached the decision-maker.

3. L3 Build Project

   Build translation infrastructure — documentation, training, communication systems

   Design it for others — teams, systems, infrastructure.

   You have built standing translation infrastructure for a team or organisation: documentation that preserves technical accuracy while being usable by domain practitioners, training that closes the most consequential knowledge gaps, or communication systems that make accurate translation the default rather than the exception. The infrastructure works without you — it is a system, not a series of personal interventions.

4. L4 Transform Portfolio review

   Drive culture change through translation at organisational scale

   Lead an organisation through it at scale.

   You have changed how an organisation communicates about AI — not just improved specific documents or training, but shifted the default. Decision-makers who previously operated on distorted AI summaries now operate on accurate ones. Technical teams that previously wrote for technical audiences now write for mixed audiences without loss of precision. You can show the culture change in observable terms, not just in sentiment.

5. L5 Guide Peer review

   Guide the hardest translation challenges at societal scale

   Operate at the frontier where the playbook runs out.

   You are advising on translation problems where the stakes are highest and the audiences are most diverse — AI policy, public AI literacy, cross-cultural AI governance, or the translation of AI capability and limitation into the frameworks used by people who will live with its consequences but did not choose its design. You can identify the deepest translation failure modes in high-stakes contexts and generate approaches to closing them.

Where this spine activates

This capability cross-cuts the entire curriculum, but these domains are where it is most directly tested and most consequentially applied.

• Mind

  The cognitive science of how humans form and update models is the foundation of accurate translation — you cannot preserve accuracy in translation without understanding how the receiving mind will process what it hears.

• Society

  AI policy and governance rest entirely on the quality of translation between technical realities and policy frames — translation failure at this boundary produces rules that are either irrelevant or harmful.

• Culture

  Cultural production is where translation failure in AI is most visible — the gap between what AI generates and what human cultural context requires is a translation problem, not a capability problem.