The hot-air balloon conference

Creating and maintaining space for productive uncertainty.

Hello friends,

tl;dr: Conventional conferences with fixed speakers, tracks, and panels are like trains. To stimulate new practical thinking in emergent areas, a conference needs to be like a hot-air balloon: able to drift a little and explore while still being steerable in real-time. A hot-air balloon conference creates and maintains a space for productive uncertainty. I derive some learnings about this type of unconventional conference from a recent AI conference in Singapore.

---

3.04am, 5 December 2023

Who remains awake in the maze of conference rooms under the Ritz-Carlton, in these benighted morning hours of Day 2 of the Singapore Conference on AI for the Global Good (SCAI)?

SCAI’s 5-person conference programming team has taken over one of these rooms as our Command Centre. Someone has tied an N95 facemask around the doorlatch to keep it from locking automatically. The central airconditioning is dialed way down on overnight mode, so the tropic humidity, nearly 100% RH, seeps in carrying with it cheap bathroom air freshener and paint fumes from workers in the hallway.

We are now in our fifth hour of trying to lock down what will happen on Day 2, which starts at 9am. In the 6 hours between now and then, we must finalise the space layout and time blocks, and brief the facilitation team. There are three of us remaining in the room — two of the team faded out a few hours ago to deal with other pressing matters.

We are jetlagged and wilted. I have just knocked over a bottle of water while reaching for a cup of coffee. Water has seeped under the three large glass panels covering the table. There is no way to get it out. Is this a sign?

Why have we decided to engage in this extreme form of nearly real-time conference programming?

Your Usual Conference

Your Usual Conference — with its fixed schedule of keynotes, papers, and panels published in advance — is like a train. It runs on tracks and sticks to a pre-set published schedule. Both participant and planner start at a known station and end at a known station. Late-breaking changes (delays, missing or over-loquacious speakers, equipment failures) are a bug, not a feature. Emergence cannot be expected from Your Usual Conference, but also one does not sit awake and slightly desperate at 3am figuring out what the next day’s program of Your Usual Conference will be.

Your Usual Conference structure is ubiquitous, but it isn’t always the best structure for a conference. How a conference should be structured should depend on what its objective is.

New practical thinking

Let’s double-click on one particular objective which seems to be of increasing importance: Stimulating new practical thinking and action in an emergent area.

This kind of conference is useful for groups ranging from academics at a bunch of universities working in the same field, to people developing new products inside a for-profit company in a rapidly changing industry or problem space like AI or climate change. (It works for governments, NGOs, and not-for-profits too.)

New practical thinking in an emergent space requires doing things that haven’t been done yet. The right metaphor here is a voyage to destinations unknown, but where you definitely want to get to a destination — like Mercantilist Age ocean exploration in search of non-specific new countries to colonise and exploit, but not like pleasure cruises to nowhere.

For this, what you need is Not Your Usual Conference.

Not Your Usual Conference

There are many types of conference that are Not Your Usual Conference. Hackathons and unconferences are the most well-known but they’re not well-suited for traveling to unknown destinations either.¹

I’m thinking of a different, harder to describe type of conference here. If Your Usual Conference is like a train, this type of conference is more like a hot-air balloon. A balloon drifts barely constrained by surface topography, so it has more degrees of freedom and is less predictable than a train. But it can be steered and thus approximately guided to a previously unscouted destination. Balloons can take you places trains cannot.

The hot-air balloon conference

SCAI was an example of a hot-air balloon conference, and the extreme real-time programming that kept us up was by design.

The programming team was a fluid collaboration between Singapore Government teams working on AI policy and the independent research organisation Topos Institute. My contribution was to inject process uncertainty by design and give people ulcers. Our objective for SCAI was to stimulate new practical thinking — new research programs or new governance/business models — about AI for global (not national or individual) good.

We began with the thesis that good questions stimulate emergent yet coordinated attempts to answer them. The participants were assembled from diverse disciplines, perspectives, and geographies (especially from across the increasingly tense East-West divide). We had no tracks, speakers, or panels because we didn’t want to run another conference that stopped at presenting points of view.

Instead, we ran SCAI as a 3-day process for collectively articulating good questions, with foundationality — illustrated below with the visual metaphor of a stack of stones — at the core of how we defined the “goodness” of a question. You can read more here about foundationality in the context of questions.

We began by mapping the problem space broadly so that we could elicit potential questions from participants that would cover as much of the space as possible. We wanted to then collectively cluster those questions for similarity before selecting a small number of the questions to refine. In outline, the process looks like this:

Foundationality was the goal throughout because we wanted the questions to stimulate diverse and emergent attempts at finding answers. But we also wanted the questions to actually stimulate action. So in the refinement phase we specifically asked participants to calibrate the question framings to balance between foundationality and tractability, as illustrated below.

The structure of SCAI had to emerge almost in real-time, because the concrete details of each phase depended on what had happened in the preceding phase. By design, we didn’t know in advance what questions (or how many questions) participants would propose in the elicitation phase, how many coherent clusters they would form in the clustering phase, how votes for clusters would be distributed in the selection phase, or what the refined questions would turn out to be. So the output from SCAI also could not be predicted ahead of time.

The physical space layout changed over the course of each day, with tables, boards, and other material moving in and out as required by the needs of each phase. We planned for as much optionality as possible, to be more accommodating of short-notice adjustments to the amount of time and the type of facilitation each phase needed. But the operations team still needed to work from provisionally final plans. This is why we ended up working almost to the start of each day to develop plans that would work for a wide range of foreseeable scenarios based on what had happened the day before.

We had an extended team of nearly 30 facilitators and documenters drawn from the civil service and private industry supporting the participants through the process. These facilitators and documenters were briefed at most an hour before each day’s schedule began. Another team of probably 80 people handled the various liaison and operational requirements, sometimes with turnarounds of just a few hours. It wouldn’t have been possible to run a conference as emergent as SCAI without such a flexible and situationally aware team.

I can speak only for myself here, but this type of emergent, nearly real-time programming is simultaneously brutal and invigorating — I believe it is the only way to program a conference that encourages emergence.

Learning from examples

SCAI is the fourth hot-air balloon conference I’ve designed. (The only other one I can talk about publicly is MAD5, which I described in part 6 of my book.)

In structure and content SCAI looks nothing like the previous three, yet to me they are all similar. This is the unintuitive zeroth learning: Hot-air balloon conferences don’t have a fixed and easily recognisable concrete structure (like Your Usual Conferences, or hackathons or unconferences).

Hot-air balloon conferences are better understood as an approach to organising meetings which generates different concrete structures depending on who’s at the meeting and what the meeting is about. This means imitating the concrete structure of one or another hot-air balloon conference won’t work. To put on a hot-air balloon conference, it’s essential to understand the underlying logic of the approach.

Fortunately, the most important aspects of this logic can be summarised in 5 learnings:

1. The meeting objective is to generate work product — but planners and participants must explicitly understand that the exact nature of the work product is not known in advance.

2. There must be an explicit conceptual frame for recognising when desired work product is achieved — this is different from defining the desired work product concretely.

3. The schedule, physical space, and support for the meeting must be designed to be as open-ended as possible and to be fixed as late as possible — so that the program is allowed to evolve as much as possible in response to what work is produced.

4. There must be the genuine possibility of failure to generate desirable work product — this possibility of failure is what allows for the possibility of unexpected success.

5. Actively creating and maintaining space for productive uncertainty takes a huge amount of work, planning, energy, and willingness to accept failure — it is not at all the same thing as being laissez-faire.

The overarching logic of this approach is to invest a lot of conscious effort in creating and maintaining space for productive uncertainty. To mix metaphors: A hot-air balloon can take you places a train cannot, but you must pay close attention to the situation and be willing to let things go off the rails.

---

One more thing: A hot-air balloon conference creates the possibility of new practical thinking. Such thinking becomes much more powerful (the phrase force-multiplier is hard to suppress) when the conference also creates broad-reaching and boundary-spanning but also high-trust social networks through which that new thinking will flow.²

Hot-air balloon conferences don’t inherently encourage these kinds of networks but they can be designed to do so. I wrote about design principles for meetings that create high-trust, diverse community a few years ago: “At a good conference, you find people who either share or complement your own interests. There are old friends and colleagues you expected to see, friends and colleagues you haven’t seen in years, and people who are entirely new to you. This, to me, is the Real Content. The difference between a good and a great conference is: at the latter, the Real Content is not drowned out by the content.”

See you in a fortnight,

VT

1. In a hackathon, the destinations are already known (they are well-specified problems that the participants come together to solve) but the paths to get there are not well-understood. A hackathon only works when success is well-understood. The metaphor for the hackathon is a bunch of drivers in their own vehicles, set loose to get from point A to B however they choose. In a true unconference (like a foo or bar camp), there is no intended destination at all and much of the objective of the conference is for participants to enjoy whatever paths emerge and end up being taken. The metaphor for the unconference is the pleasure cruise to nowhere. This can be enjoyable and rewarding, but an unconference is not good for stimulating practical thinking. Hackathons and unconferences aren’t trains but they’re still not well-suited for structuring conferences intended to find new destinations and new ways of getting to those destinations.

2. There is a lot of sociological research on brokerage (the ability to span diverse groups) and closure (trust from close ties between members of a group). Ron Burt’s Brokerage and Closure (2005) is a good place to start, and emphasizes how combining brokerage with closure enhances innovation and performance in organisations.