Event recap
Impact Vibes: where social impact meets AI
An hour in a room deliberately split between people who work in social impact and people who work in AI — to see what happens when those two worlds actually talk.
Stephen Convenor — impact, AI, healthspan, Ford Castle · Adam Co-host and facilitator · Rye Smith Claude ambassador — in conversation
Roughly seventeen people gave up an hour on a Tuesday afternoon for what the hosts openly called an experiment: put people working on social impact in a room with people working on AI, and make the two worlds spend time together. The format was a fast round of introductions, half an hour of open conversation, and a promise to leave the last stretch for whatever people were still sitting on. The room ran under the Chatham House Rule, so this records what was said, not who said it.
What the room actually talked about
Every subject the hour touched, sized by how much of the conversation it took. This is the board that was on the wall while you were in the room.
Each block's area is how much of the conversation that subject took. Built live from the room's audio as the session ran — nobody tagged any of it by hand.
The brief
"The idea is to get more insights and brains and expertise from around the room shared, rather than just a few talking heads" — that was the pitch, and it set the shape of the hour. Rather than a presentation, everyone was asked for twenty or thirty seconds: who you are, and one key issue or question you've been sitting with at the overlap of social impact and AI.
That round took nearly half the session and was, by common agreement afterwards, the best part of it. A tool was also being trialled on the wall — a QR code that let the room propose and vote on topics and submit questions without having to speak. Barely anyone used it. As the facilitator put it: "there are not too many that have popped up in the app, and I think partly because we've decided that being human's nicer."
What the room worked through
Social licence, and the automatic no
The first topic raised, and it kept resurfacing. Not whether AI works, but whether anyone has asked permission — and the mirror-image problem of people rejecting the whole category on sight.
Where is the social licence to — excuse the French — fuck up your workforce, leave a dent, and move on, because there will be something else? I don't think the conversations are connected, either vertically or horizontally. And there's a whole group of people who, if we don't have the right conversations, can't participate.
From the opening round
It doesn't live in a cloud. It lives in a place.
The most distinctive thread of the hour, and the one the room came back to hardest at the end. Data centres are not abstract — they sit in suburbs, draw power and water, and vent heat.
We sometimes think this stuff lives in a cloud. It doesn't live in a cloud. It lives in a place, and those places have water, they have heat output, they have energy use. Communities give something up for these places to exist. What are they getting back?
From the closing discussion
And they're being built in the wrong places
A specific, checkable pair of numbers that landed hard: a proposed data centre west of Melbourne would draw more energy than Victoria's largest power station generates — and that is roughly the same quantity of renewable generation currently going unsold in western Victoria because the transmission links into Melbourne don't exist.
They're building that data centre in the wrong place. You're putting the data centres where the people are, not where the energy is — and with inference latency, it doesn't matter.
From the closing discussion
The magic box problem
Several people located the core risk not in the technology but in how it is sold — and in what happens when a business believes the pitch.
The fundamental problem is a misperception perpetuated by people selling the tools: that it's a magic box that can do anything, rather than a tool that is good at some things and not good at others. People believe the magic box dream, and then they're going to blow up their business, or blow up their life, because they haven't thought about it in enough depth.
From the table discussion
What happens when the lifetime of learning goes
The same question the Melbourne fireside had circled the night before, arrived at independently here — the difference being that this room framed it as a knowledge problem rather than a hiring one.
I can see the value, with my lifetime of learning, of knowing when to challenge and when to ask. But what happens when we lose that lifetime of value? How are we going to transition that process? What does good knowledge look like? What do knowledgeable people look like in the future? And how do we keep creating them?
From the opening round
The creative industries are already through it
Not a forecast in this room but a report from the other side. One account of watching a partner's design industry be stripped by image generation, and the identity crisis that followed, became a book on the psychological and emotional impact of AI on artists and creatives. A designer in the room described a studio that hasn't adopted image generation and still carries "a deep sense of fear... we just don't know what's going to happen."
When everything's going in the same direction and everyone's doing the same thing, being original is going to be something you can't put a price on. You've seen that with AI slop. Brand was valuable beforehand — it's even more so now, and nobody's talking about it.
The counterpoint, from the table
Equity, bias, and who is in the training data
Raised early, and the single most common answer when the room was asked what it wanted to go deeper on. Two distinct concerns: who can afford access, and whose experience shapes the model in the first place.
If it's predominantly male, how does that sway the model, and what's the output of that? You're missing half the story. And you're missing the bundle of people who don't have access to the tool at all — so it's a very privileged learning, and that scares me.
From the closing round
The wrong people are in the room
A recurring diagnosis about enterprise adoption — and a deliberate echo of an earlier failure.
In these big implementations we're looking at the IT teams, and that's the wrong people. They made the same mistake with digital transformation the first time around, and they're making it again.
From the table discussion
Sovereign capability as the Australian play
The most optimistic thread. Local models running on-device, an education system as the practical vehicle, and an argument that Australia has the inputs if it chooses to use them. It was noted that South Australia is the only state permitting AI in its school system, in a controlled framework from reception to year seven.
We've got the people, we've got the space, we've got the tech, and I think we've got the capital. We have the cheapest energy on the planet if we want it — we could do a lot with this and do it the right way here. But we have to choose, and we don't always make the best choices.
From the closing round
Regulation is not the enemy anyone assumes
Recent federal moves on data centres were read as a genuine step, and the room noted the irony underneath the usual framing.
There's a perception that industry doesn't want regulation, but if you see what they've been doing, a lot of them do — as long as we're all in the same boat. I don't think regulation or governance means a slowdown of innovation.
From the table discussion
The economics won't stay this good
A clear-eyed read on subsidised pricing, and the term the room reached for repeatedly.
We're in the happy place right now — everyone's building market share on investor capital. At some point those investors want a return, and suddenly it gets a lot more expensive. We're living in glory days; enjoy it while it lasts, because you'll be paying the market rate at some point.
From the table discussion
Numbers cited in the room
Lines worth keeping
Unattributed by design — the room ran under the Chatham House Rule.
When I think about AI I think of the Simpsons scene where Homer accidentally boils his pet lobster Pinchy, then eats him bite by bite — savouring each mouthful, and crying at the loss of his friend. That's how I feel every time I use AI. This is delicious, but oh my God, I'm killing the world.
Adam · co-hosting
Remember that cartoon during COVID — people in the ring being smashed by COVID, and behind it this giant gorilla that was climate change? I see AI as a gorilla twenty times the size of that one. I'm fascinated by its potential to do great, and terrified by the negative side of it.
From the opening round
We've trained this on humanity — and humanity is terrible.
From the closing round
What I'm constantly surprised by is the lack of strategy, overlaid with a rush to the door and to quick solutions. I don't see anything different here.
From the opening round
Productivity gains for us equal more funds that go to girls and women. So we feel we have an obligation to use it and mobilise it. But then there's the third overlay — bias within AI, and the negative impact that has on girls in the world.
From the not-for-profit sector
What does it mean now that we're augmented with this capability, when people's values and passion are so much of what they bring day to day? How do we not lose that — how do we harness and amplify it? I don't think we get that for free or easily. It could easily tip the other way if we're not careful.
From the opening round
We've already had however many years of capitalism and we haven't figured out what the role of companies in our society is. Now we're creating this superpower, and we're still not healthy. It's just going to go where capitalism has gone, to an extreme — and I think that's deeply dangerous.
From the opening round
Get rid of the FOMO. Two companies I'm working with are asking what the competitor down the road is doing. Nobody knows. If someone tells you they know everything that's going on in this space, they're full of it.
From the table discussion
There's going to be a contraction. Some really rash, horrible decisions will get made, and then people will realise — hang on, we actually need these people to think, to be the visionaries, to get that creativity going. That process of contraction and expansion is not going to be a fun time.
From the table discussion
You get to a point where you're no longer dealing with the pure, you're dealing with the applied — and that's where the model stops. There is an emergence of a new round of jobs coming from the collision of a whole mess of different things, of which AI is only the underpinning.
On ageing and healthcare technology
The room has no boundaries.
Answering "so who actually needs to be in the room?"
What the room wants to go deeper on
Closing round: one topic each, for whatever comes next. Equity and energy came up most.
- Equity and access who is being pushed further back, who can't afford the tools, and the gendered shape of both the workforce building AI and the data it learns from.
- Energy, water and data centres where they're built, what powers them, what they cost the communities hosting them.
- Sovereign capability local models, Australian-trained models, and whether there's a CSIRO or university-system play.
- The applied, not the speculative "there are significant problems in this world; I want to find the real-world problem, stick to it, and work through it."
- Ethics with teeth "in drug development you have to go through ethics. Why don't we take the same approach?" Plus: what does it actually mean when a company claims to be the ethical one?
- Mental and psychological health applications, and where regulation is stepping in.
- Reimagining learning and skills "a three-dimensional approach, not our typical two-dimensional approach to skills."
- Re-democratising democracy , and the private ownership of critical infrastructure.
- Real human stories of adoption and resistance — enough of them to get pattern recognition, rather than everyone's single anecdote.
- Where it's genuinely working "we've justified some of the potential cons; where can we actually progress, and be open about it?"
- More of these, in person. The most repeated answer of the round.
Who was in the room
Names left out under the Chatham House Rule — but the mix was the whole point of the exercise.
- From social impact: a global girls' and women's charity, a platform community for impact-aligned organisations, a climate-communications specialist, a brand-purpose consultant, and an operator working across impact businesses.
- From AI and tech: a digital agency principal, a software engineer building tools for the disability sector workforce, an AI-adoption consultant, a context-engineering specialist and author, and a veteran of Apple, Wang and IBM who worked on Watson's arrival in Australia and now works in healthcare technology in Japan.
- From energy and infrastructure: an energy consultant working on batteries, virtual power plants and a fully renewable-powered data centre precinct.
- From regulated industry: financial complaints and scams prevention, in the middle of a major system change.
- From design and the creative industries: a wayfinding agency, a video marketing firm, and second-hand accounts from graphic design.
- Also present: a psychologist working with entrepreneurs on founder mental health, and several independent consultants — a running joke of the introductions.
How the hour ran
- 2:10pmWelcome and acknowledgement of countryFramed openly as an experiment: connect the social impact world and the AI world, and see what happens. One hour, described by the host as "very ambitious — I realise it was a bit silly."
- 2:15pmRound the roomEveryone: who you are, and one question you've been sitting with. Ran long, as these always do, and was widely named afterwards as the most valuable part.
- 2:50pmOpen conversationSocial licence, the magic box, regulation, sovereign capability, and where the ROI actually goes when the investor subsidy ends.
- 3:25pmQuestions from the floorAviation software safety, the economics of subsidised subscriptions, and the rebirth of jobs in applied domains like ageing and care.
- 3:40pmOne topic eachA closing lap for what people would want a whole session on. Equity and energy dominated; "more of these, in person" was the most common answer of all.
What happens next
Tell the hosts what you want more of
There was an explicit request to fill in the Luma feedback, and to say what you'd want from this forum — more topics, a different format, or something else entirely. That input decides whether and how the next one runs.
The Impact Lab, three weeks on
A hackathon-style Impact Lab was being organised for roughly three weeks after this session, gathering students, builders and people from rooms like this one to work on the problems these conversations surface.
A possible book club
Half-joking, half-serious, from the closing remarks — prompted by Eric Ries's new book on building organisations fit to solve the world's big problems.
Mentioned in the room
- "Cognitive surrender" — picked up by two people that same morning, and one of the phrases the room kept returning to
- Enshittification — Cory Doctorow — invoked for where subsidised AI products are heading, and in a piece that morning on aviation software
- Eric Ries — his new book on incorruptible organisations — from the author of The Lean Startup; on building organisations fit for solving big problems
- The Governance Institute of Australia AI study — 200 Australian businesses; the ROI, fear and training figures above
- RMIT's new centre for AI excellence — announced the day before this session
- South Australia's schools AI framework — the only state permitting it, reception to year seven
- Waymo's expansion into Tokyo — raised as applied AI for an ageing population that lives around train stations
- Open-weight and local models — raised repeatedly as the sovereign-capability route, alongside on-device inference
Captured and summarised live with Nexus. This recap is a record of a conversation, not a transcript — quotes are lightly tidied from live audio.