Simulated annealing, or, the importance of informal EA socialising

Informal social events at EA Conferences are undervalued compared to the current paradigm of carefully chosen 1-on-1 schedules. I think 1-on-1s are great, and have had excellent facilitation by organisers in my experience. However, I think informal social events are underserved by both organisers and attendees. I appeal to an analogy of simulated annealing. A minimal solution would be to name a common venue for socialising after each day to solve the coordination problem.

An explanation of simulated annealing

Simulated annealing is a technique for exploring a landscape with many local optima. It is based on cooling metal. It starts on a fitness landscape and considers a random point. If the point is better than where it is currently, it is accepted. If it is worse, it is usually rejected, but has a probability of being accepted anyway. The worse the point is, the lower its acceptance probability is. Probability is also determined by a variable T (akin to temperature in metallurgic annealing) where higher values cause higher acceptance rates. T declines over time according to an ‘annealing schedule’, such that the probability of accepting a ‘bad’ move declines over time too. In this way, you get to explore lots of points on the landscape, gradually homing in on a pretty good mountain. If you just explore locally, you could get stuck at a subpar local hill.

GIF from Wikipedia

Similarly, some randomness is good for developing your network

If you think you know precisely who you want to speak to for 1-on-1s, and have a clear decision rule for choosing whether to accept others’ requests, then informal social events probably aren’t optimal for you, aside from the function of blowing off steam.

However, I think most people are somewhat uncertain or mistaken about the exact set of people they should meet 1-on-1, in terms of specific individuals, organisations, levels of seniority, or areas of interest. Informal social events allow a degree of randomness to affect the growth of your social network, allowing you to sample from more areas of the landscape, potentially finding higher local optima. If you, like me, lack the foresight to plan out the perfect set of people to meet, then casual events offer a chance to operationalise that uncertainty. They give you a low-commitment way to accept non-optimal moves in the landscape with the potential of finding higher peaks.

People who are new to EA have the most to gain by randomness—they could find niche cause areas they hadn’t heard of that end up being their career focus. People who are well-established in EA for many years, with a thorough understanding of the different fields and players within them, may not achieve much by adding randomness. They are less likely to be surprised, and more likely to be able to target 1-on-1 meetings effectively (although some degree of randomness may still be valuable). Most people are between these extremes.

Some randomness is good for worldview diversification and community health

Worldview diversification is an important concept in EA, and it’s important for individuals to engage in it for their own epistemology and the good of the community. For example, I’m not working in AI, and because conferences are so time-constrained I will tend not to attend AI talks or have 1-on-1s with AI folks. It would be a shame if this led me to never talk to AI people—I’d be missing out on a lot. Interdisciplinary conversations are good for the individual and the community, but are unlikely to happen organically with the current dominance of 1-on-1 meetings and formalised meet-ups. Informal events are a partial solution to this.

Casual events make hierarchies flatter

It can be intimidating to request a 1-on-1 with a senior or influential person. Casual events allow brief, low-stakes interactions that can lead to longer conversations.

Avoiding the promotion of alcohol is a good objection

A good objection to this suggestion is that informal social events in Western countries typically revolve around alcohol. Many EAs choose not to drink alcohol and/​or are uncomfortable around alcohol, and alcohol itself causes great harm in society (with ~2.8 million premature deaths/​year). A partial solution would be to focus on organising non-alcoholic social events and endeavour to make events with alcohol comfortable, safe, and happy for non-drinkers (which organisers already do to some extent). And it should continue to be the case that EA events do not fund alcohol.

However, it seems reasonable to think these compromises are not enough and that EA should avoid promoting any events, formal or informal, at alcohol serving venues. I’d be interested to see community discussion of this.


Informal social events are underrated compared to the current paradigm of carefully chosen 1-on-1 schedules. Most conference attendees could benefit from adding some randomness to their networking, with benefits being greater the earlier the person is on their EA journey. Informal socialising also has benefits for worldview diversification, community health, and flat hierarchies. Naming a common venue after conference days could be a minimal solution, but an important objection concerns the promotion of alcohol consumption.