AI is a cultural question. We're treating it like a software update.

"AI is not only a technical question. It is a cultural question. It is an ecological question. It is a care question."

Most organisations haven't caught up to that yet.

I attended AI Summit 2026 in Melbourne last month. The day brought together researchers, technologists, a psychiatrist-turned-novelist, an Indigenous creative technologist and a Japanese psychologist who studies how humans perceive machines. The technical content was sharp. But the cultural thread running through every session was harder to shake.

AI is changing more than our workflows. It's changing how we think, how we relate to each other and, more quietly, whose ways of knowing get amplified and whose get erased. Here's what I took away, and what organisations should actually do about it.

Technology has always flattened culture, AI is moving faster

Matthew Ngamurarri Heffernan, a Luritja creative technologist, made a point that reframed the whole conversation. Technological flattening isn't new. When railways arrived in the 19th century, they standardised time across continents. When the telegraph ran through Central Australia, it drew language group boundaries that hadn't existed before. Each technology reshaped the culture it moved through, usually in ways that benefited whoever built it.

AI is doing the same thing, at a scale and speed we haven't seen before. The difference is that it's linguistic and cognitive, not just physical. Aboriginal English, for example, differs meaningfully from standard English, but there isn't enough training data for language models to process it reliably. Whole ways of communicating get rendered invisible, not through malice but through omission.

The implication for teams building with AI: the gaps in training data aren't neutral. Every model carries the cultural weight of what it was trained on. That's worth examining before you deploy.

We're wired to attach, and that changes things

Professor Katsumi Watanabe from Waseda University studies human perception and how we relate to machines. His research shows that people form genuine attachment to AI even when they know, rationally, that it isn't human.

In one study, brief conversation with a smart speaker created hesitation about discarding it. Sony's AIBO robot dog inspired owners to hold "funerals" and arrange parts "donation" when it broke. These aren't fringe responses, they reflect something fundamental about how humans process social cues. We're biologically primed to anthropomorphise. Add a face or a voice and our social instincts engage automatically, regardless of what we know intellectually.

This has real consequences in professional settings. Grace Chan, a psychiatrist and speculative fiction writer on the day's final panel, flagged that AI companions create an illusion of connection without the friction and complexity of real human relationships. They're always available, always accommodating, which is exactly why they're dangerous as a substitute. "There's inherent value," she said, "in being held by another fallible, flawed human mind."

For organisations: the same attachment dynamics that make AI tools feel intuitive can blur where the tool ends and the relationship begins. That matters for customer facing products, for internal tools and for how teams think about using AI for sensitive communications.

AI amplifies what it learned from, including the biases

One of the sharpest lines from the morning panel: "AI is the world's biggest amplifier." It amplifies efficiency, yes. But it amplifies existing biases at the same scale.

Early cameras couldn't reliably detect dark skin. Seatbelts were designed around male bodies. These weren't deliberate choices, they were defaults that reflected who was in the room when the decisions were made. AI inherits those defaults and runs them at speed. Professor Jie Lu gave examples of speculative future risks: autonomous vehicles making split second triage decisions based on survival statistics; medical systems making care decisions shaped by historical data that encodes historical inequality.

The practical question isn't whether your AI system is biased, it's which biases are baked in and whether you've looked. "Evaluate continuously, not just during a three-month policy exercise," was the guidance from Dr Liming Zhu, who leads CSIRO's Data61. A single sign off at deployment isn't enough when the system is actively learning and the context keeps changing.

Protecting slow intelligence in a fast-AI world

Ankit Mishra, a content specialist at Meta who led the day's workshop session, introduced a concept that's stayed with me: "slow embodied intelligence." His argument was that there are forms of knowing that only emerge through doing, through journaling, sketching, difficult conversations, walking, sitting with uncertainty. These aren't inefficiencies to be automated away. They're how humans build judgment.

The session produced a collective care protocol, seven principles for working alongside AI that the room built together. The first principle: pause before prompting. Ask what kind of intelligence this moment actually needs. Could it be answered by a walk, a colleague, a book, or silence? Not every problem is an AI problem.

The closing line landed: "Use AI only when it helps you become more attentive and more alive." That's a higher bar than most teams apply. But it's the right one.

How to adapt: Three places to start

Audit whose culture your tools reflect. Before rolling out any AI powered product or internal tool, ask: who was represented in the training data? Whose language, communication style, and context does this system handle well and who does it fail? This is especially relevant for customer facing products.

Establish what you won't hand to AI. Not as a permanent policy, but as a deliberate choice. Decide where human judgment, human presence, and human relationships are non-negotiable for your team. Write it down. Grief conversations, sensitive feedback, high-stakes decisions affecting individual people, these aren't just legal or reputational questions. They're cultural ones.

Make care a shared practice, not a personal policy. Ankit's point that AI is experienced individually but its consequences are collective stuck with me. Individual care protocols are a start. But culture is shaped by what teams do together consistently. Build the reflection in: a simple question at the end of any AI-assisted project, did this technology deepen connection or erode it?

The organisations that adapt well to AI won't just be the ones who moved fastest. They'll be the ones who stayed clear eyed about what was worth protecting along the way.