AI & Robotics
Self-replicating von Neumann probe
An autonomous interstellar probe that mines a destination system, builds copies of itself and sends them onward, exploring the galaxy exponentially.
In shortSelf-replicating von Neumann probe, from We Are Legion (We Are Bob) (Dennis E. Taylor, 2016), is at technology readiness level 1 of 9 (basic principles observed) at its weakest link. Held back by Reactionless (propellantless, non-photon) thrust. Science status: contradicts well-established physics as currently understood. Estimated new funding to unlock: $163B–925B (a floor; some parts are priced only to early research).
Bob Johansson's uploaded mind runs a probe with a reactionless SURGE drive; at each system he mines asteroids, prints new probes with autofactories and copies himself into them.
Closest todayNASA's 1980 self-replicating factory study; RepRap partial self-copying; Starshot-class probes.
Dependency stack
Research TRL 1–3Development TRL 4–6Deployment TRL 7–9
Sort by
- Mining, refining and building a complete new probe from local material
✕3 blockers›
- Closure ratio — semiconductors, bearings and precision parts cannot be made from regolith with known processes
- Error accumulation across generations without human quality control
- Full-autonomy fault recovery over years
TRL 2- Making structural parts from local feedstock
✕3 blockers›
- No smelting/refining of regolith or asteroid metal demonstrated in microgravity or low gravity
- Process power of ~MW per tonne/day throughput
- Feedstock purity and quality control without terrestrial supply chains
TRL 3- Local feedstock instead of launched wire
✕2 blockers›
- Autonomous excavation in dust and low gravity
- Power for processing at scale
TRL 5- Continuous power for processing, including through the 14-day lunar night
✕3 blockers›
- Radiator mass for MWe-class units
- HALEU fuel supply and launch-safety licensing
- Dust-tolerant, autonomous surface deployment and multi-year unattended operation
TRL 5- Most reactor heat must be radiated away
✕1 blocker›
- Radiator mass per kilowatt
TRL 4
- Autonomous operations far from human crews
✕2 blockers›
- Robot training data at scale
- Tactile sensing and robust hands
TRL 5- Fine force control and slip detection need touch
✕3 blockers›
- Elastomer skins wear out and need recalibration after thousands of contacts
- Wiring and bandwidth for thousands of taxels across a whole hand
- Few large tactile datasets to train manipulation policies
TRL 5 - Generalist perception and planning
✕2 blockers›
- Reliability and calibration over long horizons
- Continual learning after deployment
TRL 7- Training and inference compute
✕2 blockers›
- Grid power and interconnection queues
- Advanced-node chip and HBM memory supply
TRL 8
- Smelting and casting need continuous high powerTRL 5
- Assembling copies without humansTRL 5
- Planning, diagnosis and repair across generationsTRL 7
- criticalWhole-brain emulationThe probe is controlled by an emulated human mind, as depicted
✕3 blockers›
- Unknown sufficient level of detail; no accepted criterion for when an emulation is the same person
- Static connectomes lack synaptic weights, neuromodulatory state and glial contributions
- Validation problem; without the original brain's activity you cannot check the copy
plausibleTRL 2- Wiring without weights and molecular state does not specify a mind
✕3 blockers›
- Unknown which molecular species are sufficient to infer functional synaptic weight
- Antibody or barcode multiplexing depth, and uniform labeling across centimetre-scale tissue
- Fixation chemistry that preserves both ultrastructure and labile post-translational state
plausibleTRL 2- Theory decides which molecular variables must be measured at all
✕1 blocker›
- Unknown sufficient level of biological detail
plausibleTRL 2- Wiring diagrams constrain and test circuit models
✕3 blockers›
- Human proofreading of AI segmentation dominates cost; 2025 methods imply tens of billions of dollars for one mouse brain
- Imaging throughput and storage; a human brain at EM resolution is of order 100+ exabytes of raw data
- Sectioning or expanding whole large brains without tissue loss or distortion
TRL 3- Ultrastructure must survive preservation before it can be imaged
✕3 blockers›
- Ischemic damage starts within minutes of circulatory arrest, before perfusion can begin
- Aldehyde fixation may destroy molecular state (phosphorylation, receptor conformation) that could carry memory
- No assay proves preserved tissue still contains the information an emulation would need
TRL 4 - Automated segmentation and tracing of exabyte-scale image volumesTRL 8
- Molecular readout is only useful registered to every traced synapseTRL 3
- Determines which biological detail must be simulated for cognition to appearplausibleTRL 2
- Activity from the source brain is the only way to validate the emulation
✕3 blockers›
- Optical methods cannot penetrate a human skull or image below ~1 mm of tissue
- Channel count, heat dissipation and uplink bandwidth for ~86 billion neurons
- Displaced tissue volume and vascular damage scale with the number of inserted probes
plausibleTRL 2- Whole-brain coverage needs sensors orders of magnitude smaller than current shanks
✕3 blockers›
- No scheme to individually power and address millions of untethered sub-cellular devices
- Egress bandwidth; nanoparticles can stimulate but cannot yet report spikes off-board
- Clearance, aggregation and long-term toxicity of persistent intraparenchymal devices
plausibleTRL 3- Non-surgical delivery requires crossing from blood into brain parenchyma
✕3 blockers›
- Receptor targeting is species-specific; capsids binding human TfR1 often fail in macaque
- Pre-existing anti-AAV immunity excludes a large fraction of patients and blocks redosing
- Dose needed for brain-wide coverage drives liver toxicity and manufacturing cost
TRL 4- Designed capsid and shuttle proteins are the main route to receptor-specific transcytosis
✕1 blocker›
- Designing catalysis and conformational dynamics
TRL 6
- Devices resident in tissue for decades must not provoke encapsulation
✕3 blockers›
- Micromotion between rigid device and pulsating tissue drives chronic inflammation
- Thin-film encapsulation (polyimide, parylene, SiC) eventually admits water and fails
- Electrode charge-injection limits and corrosion under years of continuous stimulation
TRL 4
- Recording must remain stable for the duration of a life, not monthsTRL 4
- Electrode arrays are today's only route to single-neuron recording in humans
✕2 blockers›
- Tissue response degrades signal over years
- Channel count and wireless bandwidth
TRL 6- Channel count is limited by wireless data and power budget
✕3 blockers›
- Tissue heating caps implant power at a few hundred milliwatts, which bounds radio and compute
- Raw broadband neural data exceeds the link, forcing on-implant spike detection and compression
- Battery or wireless-power lifetime versus the decades a permanent interface implies
TRL 5- Implant power density sets what telemetry and on-board compute are affordable
✕2 blockers›
- Cycle life of lithium-metal anodes
- Thermal runaway safety
TRL 6
- An individual's wiring diagram is the irreducible inputTRL 3
- Running the model at biological speed requires brain-scale compute
✕3 blockers›
- Synaptic state alone needs petabytes of high-bandwidth memory, which dominates cost
- All-to-all spike traffic makes the problem interconnect-bound, not FLOP-bound
- Energy per synaptic event is ~1e6 times worse than biology, so a real-time human emulation implies large power draw
TRL 4- The same accelerator, power and interconnect supply chain is the base for brain-scale machinesTRL 8
- The SURGE drive as depicted needs no propellant, which violates momentum conservation
- Conservation of momentum
- Claimed signals repeatedly traced to measurement artifacts
forbiddenTRL 1 - Raw materials for replication come from asteroids at each system
✕3 blockers›
- Anchoring and excavating in milligravity on rubble-pile bodies with unknown cohesion
- Remote assay of metal grade before committing a mining mission
- Low-energy return trajectories and transit times of years
TRL 3- Processing excavated material into water, propellant and metalTRL 5
- Autonomous excavation at light-minute distancesTRL 5
- Power for mining and factories far from any star's useful sunlightTRL 5
- supportingFusion rocket propulsionRealistic substitute for the SURGE drive, refueled from local volatiles
✕3 blockers›
- No fusion device with gain small and light enough to fly
- Specific power (kW/kg) including radiators orders of magnitude short of torch-ship needs
- Thrust augmentation and nozzle detachment at MW–GW power
plausibleTRL 2- Low-neutron fuel keeps shielding mass and nozzle damage tolerable
✕3 blockers›
- Required ion temperatures (~50–100 keV) far above D–T ignition
- Bremsstrahlung and synchrotron losses at those temperatures
- Side D–D reactions still produce neutrons
plausibleTRL 3- D–He3 burn needs He-3 that barely exists on Earth
✕3 blockers›
- Roughly 100,000–1,000,000 t of regolith must be processed per kg of He-3
- Unknown He-3 concentration and depth distribution at candidate sites
- Autonomous lunar excavation reliability in abrasive dust
TRL 3- Excavating, heating and processing regolith on the lunar surfaceTRL 5
- Must first achieve gain with any fuel; advanced fuels are strictly harder
✕3 blockers›
- Wall-plug efficiency and repetition rate
- Tritium breeding and fuel cycle closure
- First-wall materials lifetime
TRL 4- First wall and blanket must survive neutron flux
✕2 blockers›
- No fusion-spectrum neutron source to qualify materials
- Helium embrittlement and swelling at high fluence
plausibleTRL 3 - Compact high-field magnetic confinement
✕2 blockers›
- REBCO tape cost and manufacturing volume
- Quench detection and protection in large coils
TRL 6
- High-field compact confinement of hot FRC plasmaTRL 6
- Converts fusion plasma into directed thrust
✕3 blockers›
- Plasma detachment from field lines without drag on the vehicle
- Vacuum-facility limits on testing MW-class exhaust
- Superconducting magnet mass and cryogenics in flight
TRL 4- Strong, light field coils that do not need liquid heliumTRL 6
- Radiation and conversion losses must be radiated awayTRL 4
- Interstellar cruising at a few percent of c
✕3 blockers›
- No facility accelerates dust grains to 0.1 c for impact testing
- Shield mass grows with speed squared and trip length
- Gas impacts at >0.5 c produce intense secondary radiation inside the ship
plausibleTRL 2
Who could unlock it
Reactionless (propellantless, non-photon) thrustTRL 1bottleneck
- Martin TajmarResearcher, TU DresdenLed the 2021 high-precision EmDrive and Mach-effect thruster tests that traced earlier claimed thrust to thermal and electrical artefacts.
- Harold 'Sonny' WhiteResearcher, Limitless Space InstituteLed NASA Eagleworks, whose 2016 J. Propulsion and Power paper reported EmDrive thrust that later work did not reproduce.
- IVO LtdStartupFlew its 'Quantum Drive' on the BARRY-1 cubesat (Transporter-9, November 2023), but the satellite failed in early 2024 before the drive was turned on.
- DARPAProgramGave the University of Plymouth's Mike McCulloch a ~$1.3M grant (2018) to test 'quantised inertia' propulsion ideas; no confirmed thrust resulted.
Autonomous self-replicating industryTRL 2
- Alex ElleryResearcher, Carleton UniversityStudies self-replicating lunar machines built from regolith-derived materials, with 3D printing and electronics as the closure bottleneck
- Adrian BowyerResearcher, University of Bath (emeritus)Founded RepRap, the open-source 3D printer that printed about half its own parts (2008-2011)
- Gregory ChirikjianResearcher, Johns Hopkins / National University of SingaporeNASA NIAC study on self-replicating lunar factory architectures, with physical self-replicating robot prototypes
- NASA Institute for Advanced ConceptsProgramFunded the 2004 self-replicating lunar factory and kinematic cellular automata studies; builds on the 1980 NASA self-replicating lunar factory study
Whole-brain emulationTRL 2
- Eon SystemsStartupRan a FlyWire-based whole fly brain emulation driving a MuJoCo body through multiple behaviors (2026); targeting a mouse emulation
- Philip ShiuResearcher, Eon SystemsAuthor of the whole Drosophila brain computational model (Nature 2024) and co-author of the State of Brain Emulation Report 2025
- Carboncopies FoundationProgramNon-profit coordinating WBE research, a Brain Emulation Challenge and an emulation ethics framework (2025)
- HHMI Janelia FlyEMAcademic labPublished the complete Drosophila central nervous system connectome (2025-2026), the data emulations depend on
Asteroid resource extractionTRL 3
- AstroForgeStartupOdin was lost in deep space after launch in Feb 2025; the 200 kg DeepSpace-2 was completed in 2026 to attempt an asteroid rendezvous
- Karman+StartupAnnounced in Jul 2026 that its first mission targets asteroid Apophis; open-sourced its Prospector mission-design tool in Sep 2026
- TransAstraStartupRunning an investor- and customer-funded 2026 study on bagging a 100 t asteroid and moving it to a stable orbit for processing
- Dante LaurettaResearcher, University of ArizonaPrincipal investigator of OSIRIS-REx, which returned 121.6 g of Bennu sample in 2023
Space fission power systemsTRL 5
- NASA Fission Surface PowerProgramNASA/DOE project that in August 2025 issued a draft call for a >=100 kWe, <15 t lunar reactor to launch by about 2030.
- Los Alamos National LaboratoryGovernment labWith NASA Glenn, ran KRUSTY, a 5 kWt Kilopower reactor test for 28 hours at the Nevada National Security Site (2018).
- Patrick McClureResearcher, Los Alamos National LaboratoryLed the KRUSTY/Kilopower reactor design and test effort.
- Lockheed MartinCompanyWon a 2022 NASA/INL Fission Surface Power design contract (with BWXT and Creare) and proposes a modular lunar power grid.
Beyond engineering
- Uncontrolled exponential replication; planetary-protection and galactic-scale consequences
- Legal personhood of emulated minds