Empathic Computing Laboratory

07/10/2026

A quietly important post from Google Research last week: they have moved the federated learning behind Gboard's next-word prediction into trusted ex*****on environments, so the privacy logic can be checked by someone outside Google rather than taken on trust. Access policies go into a public transparency log, operators only ever see anonymised model weights and metrics, and there is a whitepaper alongside it. This lands closer to our work than it might look. Our lab works with faces, voices, heart rate and gaze, signals people share with us because they trust the setup. Verifiable rather than merely claimed feels like the right bar for that kind of data. Would you hand over your physiological data to a study if the privacy guarantee were independently auditable?

We announce a new Federated Learning system that provides externally verifiable privacy guarantees while shifting computation to the server to improve training speed, accuracy, and device coverage.

06/10/2026

Writing in Waatea News last week, Matthew Tukaki made the case that Māori should not simply be consumers of AI designed somewhere else, but should be shaping how it gets built. The occasion is Aotearoa's first Indigenous Tech Summit, at Tākina in Wellington on 26-27 November, convened by Dan Te Whenua Walker of Te Whare o Toitū AI and bringing Māori, Pacific and international Indigenous technology leaders together. The question underneath it - who governs the data a system learns from, and on whose terms - sits close to our own work. We build systems that read faces, voices and physiological signals, which is about as personal as data gets, so consent and governance cannot be an afterthought for us either. What should a lab like ours be doing differently?

02102026 Dan Te Whenua Walker Māori technology leader Dan Te Whenua Walker says Māori and Indigenous communities need to help shape artificial intelligence rather than simply become consumers of technology […]

05/10/2026

IEEE ISMAR turns 25 this week. The 25th International Symposium on Mixed and Augmented Reality opens today in Bari, Italy, and runs through Friday, with ACM SUI following in the same city on 10-11 October. ISMAR has been the home venue for augmented and mixed reality research for a quarter of a century, and a good deal of what we work on in our lab - shared AR workspaces, gaze and gesture cues between collaborators, measuring how present people actually feel - grew out of arguments that started in ISMAR sessions. If you are in Bari this week, we would love to hear what has caught your attention. And if you are following along from Aotearoa, which paper or demo should the rest of us be reading?
https://www.ieeeismar.net/2026/

24/09/2026

A group of New Zealand academics is calling for a dedicated national AI office, of the kind Australia, the UK, Canada, Singapore and Japan already run. As IT Brief reports it, the case is partly practical — somewhere to coordinate work across universities, industry and community — and partly about Aotearoa having a voice in international AI governance rather than inheriting other people's rules. One figure in the piece stuck with us: New Zealanders are among the heaviest users of AI assistants in the world, and among the most distrusting of AI. That gap is a research question as much as a policy one, since trust is won or lost in the interface as much as in legislation. If such an office existed, what would you want it working on first?
https://itbrief.co.nz/story/new-zealand-academics-call-for-ai-office-as-fears-grow

With deepfakes and other harms already reaching users, academics say a dedicated office would help New Zealand regulate AI and join global talks.

23/09/2026

Meta Connect runs this week, with the keynote at 4pm PT on Wednesday 23 September — 11am Thursday here in Tāmaki Makaurau. The previews point to camera-free AI glasses aimed at the privacy complaints, a refreshed Ray-Ban Display, and a lightweight tethered headset, reportedly codenamed Phoenix, that leans on hand and eye tracking instead of controllers. All of it is rumour until Zuckerberg is on stage, but the eye tracking is the part we will be watching. In our lab, gaze and physiological signals are how we make a collaborator's attention and state legible to the people working alongside them, so when that sensing lands in consumer hardware the questions about consent and interpretation land with it. What are you hoping to see announced?
https://www.engadget.com/2262932/what-to-expect-at-meta-connect-2026-new-ai-glasses-a-mixed-reality-headset-and-more/

There's a lot for Mark Zuckerberg to cover at this year's keynote.

22/09/2026

Alibaba's Qwen team released Qwen3.8-Omni-Flash on Friday: a single model that takes text, images, audio and video in, with a million-token context and audio input across 113 languages and dialects. The detail we keep coming back to is what they call agentic perception. For long video the model decides what to watch and listen to rather than grinding through every frame start to finish, which cuts tokens by roughly 46% while improving accuracy on OmniVideoBench. That is a selective-attention problem, and it rhymes with one of ours: working out which signals from a collaborator actually matter in a shared space, and which are just noise. Has anyone tried it on multi-party video yet?
https://qwen.ai/blog?id=qwen3.8-omni-flash

Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.

16/09/2026

Bloomberg reported last week that ByteDance is building a model for generating real-time spatial video — interactive 3D scenes you can move through, in the same family as Google's Genie. It reportedly builds on their Seedance video model, could launch as early as October, and has an obvious destination in their Pico headsets.

We find the research question underneath more interesting than the product race. If a scene is generated on the fly, what happens to the shared ground two people rely on when they collaborate in the same virtual space? Presence and mutual understanding depend on both people seeing the same world, and a generated one may not hold still.

Would you trust a generated environment for real collaborative work, or does it need to be stable first?



Read it:

ByteDance Ltd. is readying an AI model geared for real-time spatial video generation, taking on Meta Platforms Inc. and Alphabet Inc. in an arena with applications in robotics and autonomous systems.

15/09/2026

Spare a thought for everyone who was awake at odd hours late last week: the CHI 2027 papers deadline closed on 10 September (AoE). The round again uses the revise-and-resubmit model — reviews come back on 5 November, authors have until 3 December to respond, and final notifications go out on 17 December. It is a kinder rhythm than a single-shot decision, and it gives authors a real chance to answer their reviewers rather than start over. CHI 2027 itself is in Pittsburgh, 10–14 May. A few of us in the lab have submissions in the mix, and we are looking forward to seeing what the HCI and XR community sent this year. Did you submit? We would love to hear what you worked on.

14/09/2026

How do you know where your remote collaborator is looking — or what they’re pointing at — when you’re sharing a Mixed Reality workspace?

Huidong Bai’s research explores how gaze, gesture, and visual cues can make remote collaboration in MR feel more natural, coordinated, and human.

Highlighted research:
→ https://dl.acm.org/doi/abs/10.1145/3313831.3376550
→ https://link.springer.com/article/10.1007/s12193-020-00331-1

Why it matters: in face-to-face teamwork, we constantly use attention cues to coordinate. In remote MR, those cues can get lost. This work investigates ways to capture and share them so distributed collaborators can better understand each other’s intent.

14/09/2026

Te Wiki o te Reo Māori runs from today until 20 September, and this year's theme is Te Reo Towns — a nod to the communities that have kept the language alive. Waipapa Taumata Rau is marking it with events at ReoSpace in the General Library all week.

It is also a good moment to think about language technology. Speech and language tools are only as good as the data behind them, and Māori-led projects — Te Hiku Media's kaitiakitanga licence among them — have shown what it looks like when a community keeps guardianship of its own data rather than handing it to whoever trains the biggest model. That question matters for anything we build that listens to people.

What is one kupu you are using more this week?