Luke Drago and Rudolf Laine
This stream focuses on identifying tractable policy and technical interventions to gradual disempowerment, focusing on economic disempowerment and the intelligence curse. Possible project areas include:
- Formalizing intelligence curse dynamics into a model that can be tracked and monitored.
- More durable policy solutions to mass unemployment than UBI.
- Technical interventions to extend the centaur period.
Stream overview
This stream focuses on identifying tractable policy and technical interventions to gradual disempowerment, focusing on economic disempowerment and the intelligence curse. Possible project areas include:
- Formalizing intelligence curse dynamics into an economic and/or political science model that can be tracked and monitored. We’re especially trying to find robust trends with predictive power that relate to disempowerment, like “a METR graph for job loss”. What economic indicators would demonstrate that we are heading towards more permanent labor replacement, as opposed to labor displacement, and can we construct a clear projection of AI capabilities and associated job losses? Relatedly, what political indicators would demonstrate democratic power or broader bargaining power is decreasing due to automation? For both, what evidence would show that automation is not causing these effects?
- More durable policy solutions to mass unemployment than UBI. Many have proposed wealth redistribution as a way to mitigate the negative effects of potential mass displacement from advanced AI. We are quite skeptical of pure cash transfers as a long-term intervention, given that they are conditionally granted by governments without an incentive to continue them. We’d like to mentor scholars considering more durable redistribution methods than UBI which could survive political turmoil and decrease concentration of power. We’re especially interested in regimes that maintain human bargaining power, keeping humans in the economic loop. Examples include tying redistribution to the location of physical assets (like datacenter dividends), universal basic compute, or ideally more radical and durable ideas. We want to have a high bar for which governance regimes withstand capital concentration (e.g. why has Norway been so resilient?), and whether any redistribution scheme is actually durable if human labor is no longer underpinning the social contract.
- Technical interventions to extend the centaur period. We’re interested in identifying areas where humans are likely to be most durable to pure AI competition and scoping the best interventions to augment human capability in those areas. Where are human-AI teams likely to be most competitive to pure-AI approaches, and how many people could be on such teams? Which interventions extend competitiveness in those domains? How can we use historical precedents to reason about which technologies might concentrate or deconcentrate power? Relatedly, does the duration of the centaur period and the speed of a transition to total automation affect how empowered people are at the end of such a transition?
Mentors
Luke is a member of technical staff at Thinking Machines. He is also the co-author of The Intelligence Curse, an essay series that examines the potential for mass automation to drive economic gradual disempowerment.
He previously co-founded Workshop Labs -- an AI research company building user-aligned models to combat disempowerment, which recently joined Thinking Machines. Prior to Workshop Labs, he was the AI governance and AI economics lead at BlueDot Impact. Before AI safety, he managed winning local election campaigns in North Carolina. He studied History & Politics at Oxford.
I work at Thinking Machines and previously co-founded Workshop Labs. I've written pieces including the scenario "A History of the Future" and co-authored The Intelligence Curse. Before that, I worked on AI safety research (including at MATS).
Mentorship style
We’ll meet 1:1 for 30 minute slots twice a week, once with each mentor. We’ll be active on Slack (default to over-slacking us), and can do quick ad-hoc calls as well. Once a week, we expect you to have some artifact that we will give feedback on.
Fellows we are looking for
We're excited about applications from a variety of backgrounds. Use the list below as general guidance, not as an exhaustive list.
Essential:
- Good at thinking across disciplines. Fellows will need to think across machine learning, economics, political theory, and historical precedents.
- Generativeness. We want people who can come up with new things that don’t fit within existing ideological or intellectual frameworks.
- A strong background in at least one of:
- economic modelling and empirical economic research
- political theory or political science
- history
- novel forecasting / modeling / analysis work in AI
Preferred:
- Ability to evaluate technically-grounded claims about AI trends and impacts
- Good at conveying ideas in compelling writing
- A large amount of existing reading across domains
Nice to haves:
- Good at modeling but not tied to models. The most important things here cannot be modeled well formally, but subcomponents can be (especially within economics), and fellows should have good taste in how to and when to model things.
- Policy experience inside a government body or a background in electoral politics (including internships)
- Technical AI experience
Project selection
We provide three projects as options we are excited about, but they are not inclusive of all ideas. During the application process, we will ask potential mentees to either a) sharpen these proposals into a more specific question incorporating their own interests, or b) propose their own projects.
We expect fellows to come in with inner conviction towards a starting point that fits within the above themes, and expect that the best work in this stream will come from self-directed fellows pursuing their own research taste. However, we will require sign off to pursue a project and may require fellows to shift scope if they move outside the target area.