Responses & Interventions
Summary
Section titled “Summary”This section covers what can be done about AI risk—from technical research to policy interventions to building societal resilience. Understanding the landscape of possible responses helps individuals, organizations, and policymakers decide where to focus.
Response Categories
Section titled “Response Categories”Technical Responses
Section titled “Technical Responses”Direct work on making AI systems safer:
| Category | Description |
|---|---|
| Technical Approaches | Alignment research, interpretability, evaluations, AI control |
| Research Agendas | Specific research programs and their theories of change |
Governance Responses
Section titled “Governance Responses”Policy, regulation, and coordination:
| Category | Description |
|---|---|
| Legislation | Enacted and proposed laws (EU AI Act, US EO, etc.) |
| Compute Governance | Hardware-based mechanisms like export controls, compute monitoring |
| International | Summits, treaties, and international coordination |
| Industry Self-Regulation | RSPs, voluntary commitments, industry standards |
Institutions
Section titled “Institutions”Organizations and bodies working on AI governance:
| Category | Description |
|---|---|
| AI Safety Institutes | Government bodies focused on AI safety |
| Standards Bodies | Organizations developing AI standards |
Epistemic & Coordination Tools
Section titled “Epistemic & Coordination Tools”Technologies and mechanisms for collective intelligence:
| Category | Description |
|---|---|
| Epistemic Tools | Prediction markets, forecasting, verification systems |
| Coordination Technologies | Tools for large-scale cooperation |
Field Building & Resilience
Section titled “Field Building & Resilience”Growing capacity and preparing for challenges:
| Category | Description |
|---|---|
| Field Building | Growing the AI safety community |
| Resilience | Building societal capacity to handle AI disruption |
Evaluation Framework
Section titled “Evaluation Framework”When evaluating interventions, consider:
Importance (Scale)
Section titled “Importance (Scale)”- How much risk reduction if successful?
- Does it address a critical bottleneck?
Tractability
Section titled “Tractability”- Is meaningful progress possible?
- What’s the track record?
Neglectedness
Section titled “Neglectedness”- How much work is already happening?
- What’s the marginal value of more resources?
How Worldviews Affect Priorities
Section titled “How Worldviews Affect Priorities”Your beliefs about AI risk affect which interventions look most promising:
| If you believe… | Prioritize… |
|---|---|
| Short timelines (<5 years) | Governance, existing systems, immediate impact |
| Alignment is very hard | Technical research, fundamental breakthroughs |
| Misuse is the main risk | Access controls, monitoring, defensive capabilities |
| Racing dynamics dominate | International coordination, industry agreements |
| Institutions work well | Policy advocacy, standards development |
Getting Involved
Section titled “Getting Involved”Different backgrounds enable different contributions:
| Background | Potential Contributions |
|---|---|
| Technical | Alignment research, interpretability, evaluations |
| Policy | Governance research, legislative work, standards |
| Operations | Supporting research organizations |
| Communications | Public engagement, translating research |
| Funding | Grantmaking, donor advising |