MIRI has a long history of thinking about how to make the arrival of smarter-than-human intelligence go well. With trends in this field increasingly pointing to disaster, in early 2024 MIRI announced a strategy shift: We would communicate our understanding of the situation to policymakers and the public, attempting to convince them to shut down the development of frontier AI systems worldwide before it is too late. This was an ambitious goal. Consequently, growing our communications team would be a major focus for 2024 and beyond.
How have we gone about this? What difficulties have we encountered? Where have we found success?
This is not a complete log of our activity. Some of our work requires discretion or timing, and we are holding off on discussion of at least one large project started in 2024 as we continue to bring it to fruition. But we had a busy year, and we would like to share some of what we’ve been up to.
Staff Changes
At the start of 2024, ours was a team of two: MIRI comms veteran Rob Bensinger (who had largely done this work himself since 2014), and the more recently added Gretta Duleba (full-time starting in September 2023). Gretta’s role as Communications Manager would charge her with growing and leading a team more appropriately sized to our new task.
She wasted no time. Our hiring rounds in the first quarter of 2024 resulted in three new hires: Harlan Stewart as a communications generalist (February), followed by Joe Rogero and myself (Mitchell Howe) as writers (late April).
Toward the end of the year, Gretta stepped back from her work managing the comms team to focus on working with Eliezer, and Duncan Sabien (already at MIRI) joined the comms team to assume some of her former responsibilities.
We’re probably not done growing yet, but we expect our next hiring needs to emerge from workflows we are still prototyping.
Direction Setting, Staff Development, Website
While MIRI staff is on the same page with regards to the nature of the AI problem and the need for a global halt, people (such as myself) have been surprised by the amount of variation within the organization when it comes to our audience models and theories of change. Learning from each other to improve our models and adjust our aim was thus a recurring objective of strategy sessions involving comms staff and senior leadership throughout 2024.
Many productive sessions of this type took place during our August comms retreat. This event, also intended as a team bonding opportunity, was rated a smashing success, and was a highlight of our year.
In the following months, we continued to adjust our models and priorities after conversations with trusted advisors and DC policy insiders. New comms team members also benefited from mentoring feedback, most notably from Rob, Eliezer, Nate, Vaniver (Matthew Gray), and valued long-time MIRI advisor Matt Fallshaw.
Despite adjustments in execution, our comms strategy still roughly matches the contours laid out by Gretta in her May post: We want to build a great foundation of “Rock” resources (explainers, briefings, websites, books, etc.) that can backstop “Wave” responses (interviews, social media posts, op-eds, etc.) to the needs of the moment, in pace with the news cycle.
As an integral part of this strategy, we’ve been overhauling the MIRI website. The existing site had various out-of-date features and did not lend itself to our new focus in terms of its design or underlying infrastructure.
We ran into several obstacles in revamping the site, such as difficulties with web developers and midstream changes in strategy.
But our biggest challenge with the site, and with comms more generally, has been the copy itself. We’re attempting to explain sometimes-technical concepts with upsetting implications to an audience that is less technical, and less proximal to the topic, than MIRI is used to. We’ve had many false starts in our attempts to combine approachability with MIRI’s precise, no-nonsense honesty. (My “Communications in Hard Mode” post goes into additional detail about our standards and their difficulty.)
While we ultimately didn’t make our planned 2024 launch dates, we were pleased to soft-launch a new MIRI website several weeks ago that includes a plain-language explainer of the AI Problem, as well as a short-form Briefing. We are continuing to make changes and add content with the help of early visitor feedback.
Media Presence
In 2023, we observed that the media had a larger appetite than we could supply for MIRI spokespeople, especially Eliezer, to do high-visibility interviews, appear on podcasts and news shows, etc. While demand slackened somewhat in 2024, we continue to see a lot of potential value in building MIRI’s spokesperson capacity. This is an especially tough role to fill, however.
Some highlights of our 2024 appearances:
- Eliezer appeared on iai.tv (June) alongside Scott Aronson, Liv Boeree, and Joscha Bach (a non-paywalled clip is available here.)
- Eliezer appeared on PBS News Hour (August)
- Eliezer appeared on Machine Learning Street Talk for a conversation with Stephen Wolfram (November)
Harlan and Gretta ran our press office in 2024, coordinating media appearances and fielding requests for information, such as from journalists asking questions on background. Notably for 2024, this included coordinating a number of conversations between our spokespeople and documentary makers; some of these were for consultations early in their process, others for footage in projects that may ship in 2025.
In April, Harlan revived the now (usually) quarterly MIRI newsletter.
As X (formerly Twitter) continues to be a centrally important hub for AI discourse, several MIRI staff remained active on the site in 2024. Some of the year’s highlights included Eliezer’s posts (blog version) on why superintelligences should not be expected to leave even trivial (to them) resources for humans, and Rob’s response to Leopold Aschenbrenner’s Situational Awareness. Of the new staff, Harlan (@humanharlan) made the most investment in the platform and has seen the most traction.
Internal and External Collaborations
Recent years have seen an explosive increase in AI news and discourse. To help keep MIRI informed of the latest developments, Harlan, in addition to our externally-facing newsletter, started publishing an internal, roughly twice-weekly newsletter. More recently, he has been sharing this with other organizations working to prevent AI ruin, and also experimenting with a public version.
Harlan, myself, and others on comms have also been providing allies with feedback (mostly solicited) on a number of prominent pieces.
We also consult with, and are often consulted by, our fellow travelers with regards to strategy and audience research. Joe, in particular, has been an effective liaison with some of the smaller organizations, and with individuals looking to help.
MIRI representatives were an active presence at a number of conferences in 2024 where policy conversations were occurring, including SXSW (March), LessOnline (June), Manifest (June), DEF CON (August), the AI Security Forum (August), and The Curve (November).
In the second half of the year, our Technical Governance Team started tapping our comms writers for help getting papers over the finish line, with good results. These two teams are now working more closely together, earlier in the process.
Audience Research
By spring 2024, Gretta had concluded that the divergence in audience models at MIRI was a major obstacle to shipping comms artifacts.
Our senior personnel have hard-won insights from their time in the trenches; they are especially versed in the many ways smart people, especially within our sphere, can take a wrong turn in processing a MIRI argument. At the same time, our new comms writers were coming in with experience teaching difficult concepts to some relatively mainstream audiences who may struggle with technical arguments. Writers from each camp were frequently skeptical about the potential efficacy of competing approaches.
Some of these differences could be reconciled through discussions, as happened at the retreat. But other questions of tactics and strategy needed to be settled experimentally. Joe experimented with some systems and services for rapid message testing. Gretta collected and synthesized MIRI’s perspectives on the structure of our argument and how people respond to it; she then worked to validate these insights, with some success, by sitting down for in-depth conversations with individuals of varying levels of exposure to the AI problem. Her work inspired some of us to have model-sharing conversations with our own contacts.
This and other work have given us some new hypotheses about audience messaging that need testing. We have much more to do in this area, and are looking to collaborate with other orgs to validate our messaging approaches.
Looking Ahead
Gretta stepping back from comms at the end of the year gave us extra reason to reflect on our work from 2024 and our aspirations for 2025. While some of us were disappointed in the quantity of our public writing output, we knew from the start that 2024 would be a year for building capacity and finding our footing.
(Hidden in this description, again, are the projects we are not ready to talk about yet. If Joe has seemed underrepresented in this write-up, it’s because we can’t properly celebrate his efforts yet.)
One thing is clear: The clock is ticking in 2025. Having launched our new website, we must finish our other ongoing projects, test our ideas in public, and make waves, learning from our mistakes as we go.
We will probably err in the opposite direction of 2024; in fact, we plan to—saying what needs to be said in a timely manner even if we are less than fully satisfied with our articulation. If we are to maximize humanity’s chance of holding on to a flourishing future, the optimal amount of awkwardness MIRI should be willing to accept is not zero.
To our supporters and critics (to their credit, these are often the same people) we say: Let us know what you think of the material comms is putting out, and about the material you would most like to see us produce. MIRI is stronger because of you. And we hope, in 2025, that you will be stronger because of us.
