Epistea Summer Experiment (ESE)

By nora, Jan_Kulveit, Anna_Gajdova, Daniel Hnyk @ 2020-01-24T10:51 (+23)

This post was written collectively by the organizing team of the Epistea Summer Experiment.

Cross-posted to LessWrong here.

Epistea Summer Experiment (ESE, /ˈiːzi/) was an experimental summer workshop in Prague combining elements of applied rationality and experiential education. The main goals were to:

• Try new ideas about rationality education, such as multi-agent models of minds, and ideas about group epistemics and coordination
• Try to import insights and formats from experiential education
• Connect people interested in rationality education

We consider the event successful and plan to use the insights gained for creating more content along similar lines.

The remainder of the post will outline our motivation for focusing on these goals, our takeaways, and future plans.

Motivations

Group Rationality

Most of today’s rationality curriculum and research is focused on individual rationality, or ‘how to think clearly (alone)’. The field of group rationality - or ‘how to think well and act as groups of humans’ - is less developed, and open problems are the norm.

[...] I feel like group rationality remains a largely open and empty field. There’s a lot of literature on what we do wrong, but not a lot of ready-made “techniques” just sitting there to help us get it right — only a scattering of disconnected traditions in things like management or family therapy or politics. My sense is that group rationality is in a place similar to where individual rationality was ~15 years ago [...]. (cited from ‘Open Problems In Group Rationality’, by Duncan Sabien)

The central problem is that people use beliefs for many purposes - including tracking what is true. But another, practically important purpose is coordination. We think it’s likely that if an aspiring rationalist decides to “stop bullshitting”, they lose some of the social technology often used for successfully coordinating with other people. How exactly does this dynamic affect coordination? Can we do anything about it?

There are more reasons why we believe progress in the domain of group rationality is needed. There are for example the classical problems in coordination (e.g. Moloch), communication (e.g. common knowledge and miasma), social dynamics (e.g. groupthink, pluralistic ignorance, etc.) and others. In particular, we have little know-how about how to teach the skills which are important for effectively solving any one group rationality problems (this is also called pedagogical-content knowledge or PCK).

Group rationality and group epistemics are important for all kinds of EA, long-termist and x-risk efforts, possibly more so than in most other domains. It has been observed in the past that topics related to EA are prone to coordination failures. Given the sheer size of these problems, it is unlikely any one of us will solve them alone.

Overall, we are confident that the field of group rationality is both important and neglected.

Experiential Education

Experiential education is an educational tradition with a roughly 40-year history in Czechia. The central idea is that the process of learning often happens through directly experiencing the content - in contrast to other, more passive ‘classroom-style’ approaches to learning. Clearly, different types of content lend themselves more to one or the other pedagogical approach. For example, advanced mathematics is hard to experience directly. Understanding team dynamics or the role of trust in cooperation, on the other hand, can effectively be taught via relevant experiences.

Several members of our team have organized events of this type, some of them targeted to talented high schoolers. Judging by our personal experience, the methodology succeeds in giving people more agency and helps them coordinate in a group context. This is why we were excited to transfer the pedagogical know-how from the experiential education school of thought to teaching group rationality in particular.

Multiagent Models of Mind

We developed a new technique for increasing internal alignment called ‘Internal Communication Framework’ (ICF) [1]. On the theoretical side, it is based on multi-agent mind models, predictive processing and game theory (for more theory behind our effort, see Jan's post and Kaj's sequence). On the practical side, it draws from Internal Family Systems (IFS) and Internal Double Crux (IDC).

Our goal for the event was to gather more data and we have anecdotal evidence that the technique is useful for increasing internal alignment in individuals.

Workshop

Participants

Since the event was highly experimental, we did not extend our outreach efforts to a broad audience and instead targeted the event to people who already had prior experience with applied rationality. Our participants were CFAR alumni, members of the Moscow Kocherga hub and other people with a similar background. The motivation behind this was to minimize the risks in cases where our content wouldn’t work. Specifically, we wanted to avoid several failure modes, including idea inoculation, PR risks and destabilizing participants prone to mental health issues. (We also warned everyone that the content was experimental.)

Curriculum design

In creating the activities, we drew a lot of inspiration from the Czech Experiential Learning Centre. Over the years, they have accumulated an impressive amount of knowledge and a library of activities ranging from outdoor games to deep introspection methods. We adjusted some of these for our goals and created a couple of new ones. We also incorporated a few classes which were taught traditionally (i.e. classroom-style) to introduce relevant concepts. Part of the content was created originally by Epistea, based mainly on theoretical reasoning.

After generating a set of individual activities, we arranged them into a 6-day long program while taking into account program coherence, the need to balance mental and physical intensity for participants, the overarching plotline, and other factors.

Examples of activities or techniques used are:

• Role-playing game focused on the mechanisms of depleting shared resources (tragedy of the commons)
• Game in which participants work in teams on a range of problem-solving tasks while one or two of them are secret “saboteurs”
• Trust-o-meter: similarly to CFAR's toolification of accessing implicit representations of surprise (i.e. surprise-o-meter), we can access and calibrate over time implicit representations of various social variables
• "Tea tasting": accessing implicit models (similar to some forms of Focusing)
• Class on Internal Communication Framework (ICF)
• Campfire ritual prepared by participants
• Outdoor games

Outcomes

Overall we consider the workshop successful:

• Results of the survey filled in by the participants at the end of the event indicate that experiential education may be a good medium to convey group rationality concepts. [2]
• For example, two experiential games focused on coordination were by half of the participants evaluated as "extremely useful to improve external coordination".
• We received positive feedback on the Internal Communication Framework.
• “ICF is a more general and powerful tool than the stuff I was using before (IDC, NVC, CBT) and I think it might help me to make more robust inner peace.”
• “ICF has succeeded where several similar systems were less successful.”
• “[ICF] appears to have unblocked me with respect to introspection and system 1.”
• We deepened connections among the currently somewhat scattered European rationality space.
• All of the participants reported they would be comfortable to reach out for help with a rationality related task to more people than before the event. (The imprecise estimate of the average delta is ~8 new people.)
• The team gained valuable insights into group rationality and how to teach it.
• Some of the activities from ESE were successfully re-used at ESPR (European Summer Program on Rationality).

We also made mistakes:

• We created a logistical nightmare for ourselves by running the event back-to-back with the Human-aligned AI Summer School. (This was motivated by the wish to make it easier for people to participate in both events, but was likely not worth it.)
• We were miscalibrated about people's physical abilities and overestimated e.g. their stamina.
• Some of the activities were prepared last minute at the expense of the energy of organizers who could otherwise use it to make the event run more smoothly or get more rest.

Conclusion

Though the participants’ feedback was positive, the sample size collected last summer is small. We will continue to experiment with rationality education along similar lines and gather more data. Our future plans include events focused on group rationality, epistemics, ICF and also exchange of ideas among rationality teachers. Further information on these will follow soon. If you want to make sure not to miss information on future events we organize, you can fill in this form.

[1] As the technique and the right method of teaching it are still in development, there isn’t any public write up available.

[2] There is going to be a follow-up survey to check if the participants retained the knowledge and how they evaluate the experience after a longer period of time.