Will whole brain emulation arrive before other forms of AGI?
Researchers differ over how likely it is that the first artificial general intelligence (AGI) will be produced via whole brain emulation (WBE).
There seems to be at least moderate agreement on how long it will be until WBE is feasible. For instance, Anders Sandberg predicts a median of 2064 for the year in which the technology for WBE will first be available. Robin Hanson guesses that it will be available within a century.
By contrast, there is substantial disagreement about the timelines for AGI. Some researchers expect AGI before 2030, while others don’t expect AGI before 2100. Hanson, Gary Marcus, are set on long AGI timelines whereas many researchers expect AGI before 2050.
An AGI built on WBE might have some desirable alignment properties, but would also present ethical and safety issues.
Another risk with researching WBE is that even if it is developed before other approaches, progress on modeling the human brain is likely to help with designing neuromorphic AGI — AI with a structure similar to the human brain but which is not an exact copy. It is likely to be easier to use insights about the human brain which were discovered while trying to model its components in order to copy them to a computer, than to work out all of the intricacies of copying a brain exactly. This suggests that WBE is unlikely to arrive before neuromorphic AGI[1].
As an analogy, imagine trying to assess in 1900 (before the Wright brother’s 1903 flight) whether the first flying machines that would be built would emulate birds with moving winds or would use a different design. One could intuit that understanding the aerodynamics of flight for birds could help in building the conceptually much simpler fixed-wings airplanes, and if you had a mostly-functional prototype of a mechanical bird you could adapt it into a functional airplane. ↩︎