Whole brain emulation, sometimes called mind uploading, is the idea of producing machine intelligence not by designing it but by copying an existing brain. The plan, as laid out in the 2008 Oxford report “Whole Brain Emulation: A Roadmap” by Anders Sandberg and Nick Bostrom, has three broad stages: scan a particular brain at high enough resolution to capture its structure, translate that scan into a computational model of how the neurons connect and behave, and run that model on hardware powerful enough that the emulation reproduces the original mind’s behavior.
What makes the concept distinctive is that it sidesteps the question of how intelligence works. If a faithful enough copy is run, it should be intelligent for the same reason the original was, without anyone needing to understand the principles. This contrasts sharply with the mainstream path of training neural networks, where researchers engineer the system from the outside. It also raises questions the engineering path does not: whether a copy would be conscious, whether it would be the same person, and how a society would treat minds that can be paused, copied, and run at different speeds.
The required fidelity is the central uncertainty: nobody knows how much neural detail must be captured for an emulation to work, which is why the roadmap frames it as a question of degree rather than possibility. The concept is taken seriously in long-term AI forecasting because it offers an alternative route to human-level and superhuman capability. Economist Robin Hanson built an entire projected civilization around it in “The Age of Em,” and which path arrives first - engineered AI or brain emulation - is a recurring theme in debates over how AI capability might take off.