it:asi_vs_fusion
the search for holy grails - super intelligent AI vs nuclear fusion power
see also:
Introduction
- both of these technologies have been holy grails for the past 50 years and both require enormous amounts of money, R&D and electricity
- in early 2024, AI leaders were confident that AGI - “general human intelligent” AI would probably eventuate in 2027 after $US10-100b training up-scaling of current large language model AIs
- in July 2024, the ITER multinational fusion energy project marked completion of its most complex magnet system at a cost of some $US22b, but a working nuclear fusion power plant will not be ready until at least 2035, and even then, this experimental version with an output of only 200MW of electricity may not even break even and would require a larger version to produce 2GW of power. 1)
- this page was created to give some comparative perspectives on these technologies
Comparisons
| ITER experimental nuclear fusion plant | AGI level AI | ASI level AI | |
|---|---|---|---|
| expected date of completion | 2035 at the earliest, but commercial plants may be a further 20-50yrs away | c2027 | c2030-2040 |
| cost | $US22b so far | $US10-100b per AI model | $US100-2000b per AI model |
| theory | D + T → He-4 + n + 17.6 MeV with 14.1 MeV resulting in neutron kinetic energy which when impacting the lithium seeded tube walls results in heat to boil water for steam turbine use plus replenishes tritium from the lithium | modified transformer algorithms trained using GPUs on synthetic and web data, optimised by human and AI reinforcement feedback systems | ?new AI algorithms and new AI chips |
| development goals | deuterium/tritium plasma temperature raised to 100-150 million degrees Celsius to allow fusion to take place (10x hotter than core of the sun as there is not the added extreme gravitational forces that are present in the sun) | > 2M context length LLM trained on trillions of tokens | ? |
| development resources | 87,000 kilometres of thin wire was needed to create 19 toroidal field coils to generate 41GJ magnetic energy; | >1,000,000 GPUs, ?0.5GW power | ?2.5GW power |
| outcomes | 200MW clean power with future 2GW clean power per plant | AI capable of doing most human tasks, provide adequate intelligence for robots, and potentially could escalate progress to ASI | super-intelligent AI which may be able to solve problems that are currently beyond our capacity |
| benefits | unlimited clean power for ever? | increased discoveries, improved worker productivity, better health care, provision of important virtual workers such as virtual clinicians for areas which currently lack these (rural, Third World, etc), etc | solve problems we currently cannot solve (longevity, ageing, disease, physics, etc), replace most human jobs and with the improved productivity, humans can live a very different life - perhaps |
| risks | may not be able to create a commercially viable plant; although there are extreme temperatures involved and magnetic forces (these equate to around 10 MRI machines), these plants are far safer than nuclear fission plants as there is no risk of reaction runaway, and there is minimal radioactive material production or waste; | general AI risks - deep fakes, misinformation (and potential loss of democracy), bias, bad actor usage, near total loss of privacy, job displacement, etc | major geo-political risks - 1st to ASI will have a major military tech advantage - may result in a global dictator who may not be as globally benevolent as we would like; major societal changes; possible existential risk to humanity |
it/asi_vs_fusion.txt · Last modified: 2024/07/02 07:37 by gary1