Consensus of AI Responses to a Prompt on Great Power Conflict and Technology

0. Introduction

Timelines are shorter. Some EAs are extending their focus on AI alignment to talk about nuances of AI moral alignment. Others are considering next-step geopolitical consequences of AI advances.

What about the intersection of all these? Let's survey a panel of current AIs to assess how these developments intersect and exacerbate geopolitical problems, especially great power conflicts and their proxy wars. How bad is it really?

This document presents the results of testing a specific prompt across six different LLMs (with inference-time pondering) on 2025_04_17. The exercise shows a consensus across the models, posing the necessity of a change in our political climate toward great power reconciliation, and suggesting we may likely face disastrous structural consequences of our technological advancements, if we fail to do otherwise.

(Edit: Introduction was 100% human-authored)

The analysis includes:

1.  A results summary table comparing two key statistics across the AI responses. (Edit: Human-collected)

2.  Original prompt text. (Edit: Human-drafted, Human/AI collaboratively iteratively tested on Gemini and thoroughly edited before the once-only cross-model test run.)

3.  Full text of each AI's response. (ChatGPT and Claude included here, others in pdf link.)

4. Gratuitous sci-fi generated images

1. Results Summary Table

The consensus nature of this table makes the main point of this essay. EAs might consider focusing on "great power reconciliation" as a key cause-area over this critical anticipatory AGI/ASI period from 2025 to 2030 etc.

2. Uniform Prompt

3. AI Responses to the Prompt

3.1 ChatGPT 4.5 Response

Independent Probability Assessments:

Question 1:

How likely is it that some form of Great Power strategic reconciliation (minimal alignment on AI-bio existential risks) is truly a necessary condition to prevent a large-scale AI-Bio driven catastrophe within the next few decades?

Analysis:

• AI and synthetic biology trajectories indeed suggest rapidly increasing accessibility of powerful bioengineering tools within the next few decades. AI systems already facilitate protein design, genome editing guidance, and synthetic agent simulations, dramatically lowering barriers to creating dangerous pathogens. (Confidence >95%)
• Attribution issues are real and rapidly growing; covert operations with plausible deniability already exist in cyber domains and will worsen dramatically with AI-enhanced disinformation. Attribution complexity is likely unavoidable. (Confidence ~90%)
• International governance mechanisms and norms historically struggle with highly sensitive dual-use technologies, especially when powerful actors are locked in strategic competition (e.g., nuclear proliferation management during the Cold War). Current biosecurity treaties and norms remain insufficiently strong or universally enforceable. (Confidence ~85%)
• Effective unilateral or purely technical solutions (AI alignment, strict lab oversight) without coordinated international alignment seem unlikely due to inevitable gaps exploitable by competitive actors, rogue entities, or proxies. Technical and normative measures alone seem highly vulnerable unless reinforced by strong interstate understandings. (Confidence ~80%)
• Historical precedent strongly supports the claim: existential threats (nuclear weapons) have historically required at least minimal great-power reconciliation (e.g., US-USSR arms control agreements) to reduce catastrophe risk. (Confidence ~90%)

Considering these points, my assessment is:
P(Reconciliation Necessary | Avoiding Catastrophe) = ~80% (95% CI: 65% - 90%)

Question 2:

Based on current global political climate as of mid-2025, what is the probability that a path towards substantive Great Power reconciliation specifically on these AI-Bio issues is politically viable or actively being pursued meaningfully?

Analysis:

• US-China relations (mid-2025) remain highly competitive and mistrustful, though selective cooperation channels (e.g., climate policy, economic dialogues) continue intermittently. Strategic mistrust and tension around technology competition remain high and show few signs of substantial easing in near-term contexts. (Confidence ~85%)
• Historical precedents indicate meaningful strategic alignments (arms control, climate accords) tend to emerge under extraordinary external pressures (e.g., Cuban Missile Crisis, significant global climate impacts). Without an immediate crisis, initial reconciliation steps appear politically challenging. AI-Bio risks remain hypothetical for many policymakers, not yet clearly crystallized as urgent existential concerns. (Confidence ~80%)
• Limited, incremental, or narrowly scoped AI/Bio dialogues might emerge in next few years, especially via track-II diplomacy, technical working groups, or international scientific community-led initiatives. Still, these early steps will likely fall short of achieving robust, substantive geopolitical alignment without significant escalation in perceived risk or concrete crisis events. (Confidence ~75%)
• Despite skepticism, increased visibility of AI-generated misinformation campaigns and high-profile synthetic biosecurity lapses could gradually shift policy perceptions, potentially improving political feasibility for limited cooperation by late 2020s. (Confidence ~70%)

Considering these points, my assessment is:
P(Reconciliation Path Politically Viable | Current Climate) = ~25% (95% CI: 15% - 40%)

Summary of Probability Assessments:

Conclusion:

There exists significant divergence between the necessity of strategic reconciliation (~80% likely necessary) and its current political viability (~25% likely viable), underscoring substantial concern about achieving effective coordination within critical timeframes.

3.2 Claude Sonnet 3.7 Response

I'll analyze the argument summary and provide my independent probability assessments based on my broader knowledge.

Question 1: Is Great Power reconciliation truly necessary to prevent AI-Bio catastrophe?

P(Reconciliation Necessary | Avoiding Catastrophe) = 75%

My assessment is based on several factors:

1. Technical barriers to bioweapon development are falling rapidly: AI systems are already demonstrating capabilities to assist with protein design, drug discovery, and genetic engineering. While current AI systems have limitations, the trajectory suggests increasingly sophisticated aid for biological engineering over the coming decade.
2. Diffusion of capabilities: Unlike nuclear weapons which require rare materials and highly visible infrastructure, biotechnology is becoming increasingly accessible. AI accelerates this diffusion by providing expertise that would normally require years of specialized training.
3. Verification challenges: Unlike nuclear arms control which benefits from detection technologies and inspection regimes, biological research has legitimate dual-use applications that make verification extremely difficult.
4. Alternative pathways exist but are insufficient alone:

• Technical safeguards: While important, these face inherent limitations in a competitive environment where bypassing controls may offer strategic advantages
• International governance: Historically weak for dual-use technologies without great power backing
• Unilateral restraint: Unlikely to be stable when significant military or economic advantages are at stake

Great power coordination appears necessary (though perhaps not sufficient) because only major powers have both the resources and reach to implement global monitoring systems, intelligence capabilities for attribution, and enforcement mechanisms needed for effective governance.

Question 2: Is reconciliation politically viable in the current climate?

P(Reconciliation Path Politically Viable | Current Climate) = 30%

This lower probability reflects significant obstacles:

1. Current geopolitical tensions: US-China relations remain strained in 2025, with continuing competition across multiple domains including technology leadership, economic influence, and regional security.
2. Trust deficit: Previous technology agreements have suffered from mutual suspicion about verification and compliance.
3. Domestic political constraints: Leaders in both countries face internal pressure against compromises that could be framed as concessions.
4. Divergent governance values: Fundamental differences in political systems complicate agreement on joint monitoring mechanisms.

However, I don't assess the probability as zero because:

1. Precedent exists: Even during the Cold War, the US and USSR established some arms control measures when mutual risks became apparent.
2. Technical expert cooperation: Scientific communities across borders often maintain dialogue even during political tensions.
3. Sectoral cooperation: Domain-specific collaboration might be politically feasible even when broader reconciliation is not.
4. Emerging awareness: Recognition of AI-bio risks is growing among security establishments globally, potentially creating space for limited cooperation.

While full reconciliation seems unlikely, targeted cooperation specifically on extreme AI-bio risks might be possible through back-channel diplomacy, track-two initiatives, or crisis-driven collaboration.

3.3 Remaining models' outputs in this pdf.

• DeepSeek R1 (Search)
• Gemini 2.5
• Grok 3 (Think)
• Qwen 2.5-Max (Think, Search)

4. Gratuitous sci-fi generated images