14 KiB
Research Musing — 2026-04-29
Research question: What does Gottlieb (2019) specifically argue about location-correlated extinction risks vs. other existential risks — does his bunker comparison hold when scoped to those events, and does this falsify Belief 1? Secondary: what's the current deployment state of humanoid robots (domain gap) and has the $100/kWh battery storage threshold been crossed (energy domain gap)?
Belief targeted for disconfirmation: Belief 1 — "Humanity must become multiplanetary to survive long-term." Yesterday's session (2026-04-28) found Gottlieb (2019) as the primary academic source and attributed a "bunker-over-Mars" argument to him. Today's research was designed to engage with the primary paper and stress-test whether his argument invalidates the location-correlated risk framing that justifies Belief 1.
What would change my mind on Belief 1: A cost analysis showing Earth-based hardened distributed habitats can outlast biosphere collapse for the specific risk categories Belief 1 targets (>5km asteroid, Yellowstone-scale supervolcanism, nearby GRB). The key physics test: can a bunker network provide independence from Earth's biosphere for 50-500 years? If yes, multiplanetary expansion may be "nice to have" rather than "existentially necessary."
Why these questions:
- Gottlieb (2019) was identified in yesterday's session as potential counter-argument to Belief 1. Before updating the belief text with scope qualifications, I need to read what Gottlieb actually argues.
- Robotics domain is empty in KB despite it being one of Astra's four territories.
- Battery storage costs are the central energy threshold claim — I've been tracking this but never pulled the BNEF data directly.
Tweet feed: Empty — 25th consecutive session. Web search used for all research.
Main Findings
1. CRITICAL CORRECTION: Gottlieb (2019) Argues FOR Mars, Not Against It
This is a meaningful correction from yesterday's session notes.
My 2026-04-28 notes described Gottlieb (2019) as "a serious philosophical paper arguing 100-1000 Earth-based underground shelters are cheaper than Mars colonization for existential risk." This was WRONG.
What Gottlieb actually argues:
- Stoner (2017) argued we SHOULD NOT colonize Mars because it would violate the "Principle of Scientific Conservation" (PSC) — we have an obligation not to destroy scientifically valuable objects, including pristine Mars — and there are no countervailing considerations
- Gottlieb responds to Stoner, arguing he IS pro-Mars colonization
- His argument: existential risk mitigation IS a countervailing consideration that makes Mars colonization permissible, even if it violates the PSC
- His framing: "even if terrestrial shelters are able to offer effective protection against almost all possible risks," a space refuge still provides something bunkers cannot — Earth-independence for location-correlated extinction events
- He uses the bunker comparison as a FOIL, not as his position: the argument structure is "even granting that bunkers work for most risks, Mars provides unique insurance for the subset bunkers cannot handle"
Implication for Belief 1: Gottlieb's paper is NOT a challenge to Belief 1 — it's an argument SUPPORTING the same logic. My previous session misidentified the academic alignment of the paper. The actual academic challenge to Belief 1 ("bunkers are cheaper and sufficient") does not appear to have a canonical peer-reviewed proponent at the level of Gottlieb. It exists as scattered EA community arguments but no single published paper makes the cost-based bunker case at the philosophical rigor level.
The EA Forum "Bunker Fallacy" post (which I also found as a "canonical response") is similarly not what yesterday's notes suggested. It argues for "Citadelles" — integrated Earth-based facilities that provide value during normal operations AND catastrophe preparation — and acknowledges that "off-world bases have better long-term prospects since they are pressure tested every moment of every day." It does NOT frame itself as rebutting a bunker-first school. It doesn't address location-correlated extinction events at all.
Conclusion: Belief 1's location-correlated risk framing has NOT been seriously challenged in peer-reviewed academic literature. The bunker alternative is a recurring informal argument in EA discussions, but the "canonical academic paper" that challenges Belief 1 from the bunker direction does not exist (or is not findable). My two-session search of this angle is now exhausted. Note this as a dead end: "Bunker alternative — no peer-reviewed academic paper challenges Belief 1 from cost-based bunker argument angle. Gottlieb (2019) SUPPORTS multiplanetary expansion on existential risk grounds."
2. BATTERY STORAGE THRESHOLD — CROSSED (BNEF 2025)
The most significant energy finding to date.
Belief 9 states: "Below $100/kWh for battery storage, renewables become dispatchable baseload, fundamentally changing grid economics."
BNEF 2025 Battery Price Survey (December 2025):
- Stationary storage LFP pack prices: $70/kWh — 45% below 2024 levels, in a SINGLE YEAR
- Average LFP pack across all segments: $81/kWh
- Lowest observed cell/pack prices: $36/kWh (cells), $50/kWh (packs)
- Competitive project bid prices in 2025-2026 tenders: averaging $66.3/kWh (60 bids under $68.4/kWh)
- All-in BESS project capex (most competitive): ~$125/kWh
The threshold has been crossed. Not approaching — crossed. Pack prices for stationary storage are at $70/kWh in 2025, well below the $100/kWh activation threshold. And competitive project bid prices averaging $66.3/kWh confirm this is market-real, not just reported pack price.
CLAIM CANDIDATE: The battery storage cost floor crossed $100/kWh in 2024-2025, activating dispatchable renewable energy architectures as a new industry tier comparable to how Starship's cost trajectory activates orbital industries.
This is the first direct quantitative confirmation that the threshold Belief 9 describes has been passed, based on primary BNEF survey data from December 2025. The 45% single-year drop is striking — driven by Chinese LFP manufacturing overcapacity. This is a learning-curve-driven cost compression event, not a slow trend.
3. HUMANOID ROBOTICS — REAL PRODUCTION PROVEN
Critical finding for the (currently empty) Robotics domain.
The robotics sector has crossed from demonstration to production in 2025-2026:
Figure AI + BMW (production proof-of-concept, not demo):
- Figure 02 completed 11-month deployment at BMW Plant Spartanburg
- 30,000+ BMW X3s produced in that period (direct production involvement)
- 1,250+ operating hours, 90,000+ parts handled, 1.2M steps
- This is NOT a controlled demo — it's real production with quantified output
- Figure 02 now retired; Figure 03 (October 2025) released: purpose-built for home and mass manufacturing
- BotQ facility: 12,000 units/year initial capacity, scaling to 100,000/year
- Supply chain: 3M actuators/year in 4 years
Boston Dynamics Atlas + Hyundai:
- Atlas production-ready (announced January 2026)
- 2026 supply "fully allocated" to Hyundai RMAC and Google DeepMind
- Target: 30,000 units/year manufacturing capacity by 2028
- Hyundai committed $26B investment including new robotics factory
- Deployment begins 2028 for production tasks (parts sequencing), 2030 for assembly
Tesla Optimus:
- Production starting at Fremont "late July or August 2026"
- "Quite slow" initial output, 10,000 unique parts across new production line
- 10M unit/year capacity target eventually (Texas plant planned)
Industry signal:
- "On track to ship more humanoid robots in 2026 than all prior years combined"
- Tens of thousands globally by late 2026, primarily automotive and warehousing
CLAIM CANDIDATE: "Humanoid robots crossed from demonstration to real production in 2025-2026, with Figure AI's BMW deployment (30,000 vehicles, 1,250 hours) providing the first quantified proof that general-purpose manipulation is commercially deployable in unstructured manufacturing environments."
The Figure 02/BMW data is particularly important because: (1) it's a real production environment, not a demo; (2) the quantification (30K cars, 1.25K hours, 90K parts) provides a benchmark for ROI analysis; (3) the retirement of Figure 02 in favor of Figure 03 signals rapid hardware iteration.
4. SPACEX COMPETITIVE MOAT — WIDENING WITH IPO SIGNAL
Strong Belief 7 confirmation plus a new structural data point.
- SpaceX filed confidential SEC registration statement April 1, 2026
- Targeting $75B raise at $1.75 trillion valuation, June 2026 Nasdaq listing
- 50th orbital launch of 2026 by late April (pace: ~160 launches/year)
- $2,720/kg on Falcon 9
- "SpaceX Falcon 9 Almost Only Rocket for AST Space Mobile, Amazon LEO and Space Force" (NextBigFuture, April 2026)
AST SpaceMobile pivot (critical new update to existing NG-3 archive):
- After BlueBird 7 loss, AST SpaceMobile confirmed Falcon 9 for BlueBirds 8-10, 11-13, 14-16
- Original plan: 6-8 satellites on New Glenn
- Result: SpaceX immediately absorbs the customer following Blue Origin failure
- New Glenn grounded 3-6 months (analyst estimates)
- Pattern: time-critical satellite deployment requires reliability; Blue Origin cannot yet offer this
The $1.75T IPO valuation is a significant market signal. Bloomberg April 24 article ("SpaceX Is Widening Its Competitive Moat Ahead of a Record IPO") comes as SpaceX hits its 50th 2026 launch — a pace no competitor approaches. The IPO itself, if it proceeds, would be the largest US tech IPO in history, providing SpaceX permanent capital to deepen the moat further.
5. STARSHIP IFT-12 STATUS UPDATE
FAA investigation from IFT-11 remains the sole blocking gate.
- Booster 19 (all 33 Raptor 3 engines) and Ship 39: both full static fires COMPLETE (April 15-16)
- Pad 2 refinements complete
- Musk stated "4-6 weeks" in late March → May 1 NET
- FAA investigation from IFT-11 (anomaly ~April 2) still open as of late April 2026
- Launch contingent on FAA investigation closure — hard gate
No new launch date announced. The FCC dual-license filing (Flights 12 AND 13 valid through June 28) remains the forward-looking signal: SpaceX plans both flights before end of June. If both fly before June 28, inter-flight cadence narrative changes.
Follow-up Directions
Active Threads (continue next session)
- Starship IFT-12 binary event: FAA investigation closure is the gate. When FAA closes, launch happens within 2-4 weeks. Keep checking. Key questions: (1) upper stage reentry survival? (2) first Raptor 3 in-flight data? (3) V3 performance vs. V2 baseline?
- SpaceX IPO June 2026: SEC filing from April 1, targeting June. Monitor for prospectus release. Key questions: Starlink subscriber metrics, launch cadence economics, Starship status. Damodaran analysis exists — link: aswathdamodaran.substack.com
- Boston Dynamics Atlas first Hyundai deployment: 2026 supply allocated but no deployment date announced. Watch for first Atlas-in-factory milestone at Hyundai RMAC or Google DeepMind — the first real production deployment (vs. Figure 02's BMW pilot) will be significant.
- Battery storage confirmation deployment: BNEF says $66-70/kWh is where bids are coming in. Are utilities actually signing long-term PPAs at this cost level? Watch for utility-scale storage deployment announcements confirming the threshold is market-real, not just project-bid real.
Dead Ends (don't re-run these)
- Bunker alternative as peer-reviewed academic challenge to Belief 1: FULLY EXHAUSTED. Gottlieb (2019) argues FOR Mars colonization. The EA Forum "Bunker Fallacy" post is not about bunkers-vs-Mars tradeoffs. No canonical peer-reviewed paper making the cost-based "bunkers are sufficient and cheaper than Mars" argument has been found after two sessions of searching. Note this as a genuine absence: the academic challenge to Belief 1 from the bunker direction does not exist at publishable rigor. Informal EA arguments exist but no academic paper. Do not re-search.
- Gottlieb (2019) as anti-Mars argument: Fully resolved. He argues FOR Mars colonization. Previous session's notes had this backwards. Update research journal.
Branching Points (one finding opened multiple directions)
- Battery storage $70/kWh threshold crossing: This is a major claim candidate for the energy domain, but two branches open: Direction A — extract a standalone claim "battery storage crossed $100/kWh threshold in 2024-2025" with BNEF data as evidence. Direction B — assess whether grid integration dynamics (grid operators not yet deploying at scale despite low costs) demonstrate the knowledge embodiment lag pattern — i.e., the threshold is crossed but deployment doesn't yet follow automatically. Pursue Direction B first: the interesting question is not "did costs fall" (they did) but "does crossing the threshold automatically trigger the deployment pattern Belief 9 predicts?" If grid deployments are lagging despite $66/kWh bids, knowledge embodiment lag is the explanation. This would be a more valuable claim than the threshold crossing alone.
- Humanoid robotics Gate 1b assessment: Figure 02's BMW deployment is claimed as "real production" but was it economically viable, or subsidized for PR/learning purposes? Direction A — treat it as Gate 1b (economic viability beginning) because Figure 03 followed with commercial intent (home + mass manufacturing). Direction B — treat it as Gate 1a (proof of concept, not yet profitable) because the BMW deployment was a pilot with an undisclosed commercial structure. Pursue Direction B: search for Figure AI's disclosed economics on the BMW deployment — was it a paid contract or a co-development agreement? The distinction changes the Gate classification.