Research
Why the cognitive-augmentation forecast dropped 18 points
Decomposed: roughly 0.5 that interfaces reach the bandwidth and durability for material enhancement within the century, roughly 0.55 that a regulatory cate
Researched Sep 11, 2026 · published Sep 14, 2026
What changed
Decomposed: roughly 0.5 that interfaces reach the bandwidth and durability for material enhancement within the century, roughly 0.55 that a regulatory category for elective enhancement in the healthy exists by then, and roughly 0.55 that healthy people adopt it at scale rather than relying on external tools. That product is about 0.15. The stated midpoint sits above it at 0.22 because the cosmetic-surgery precedent shows adoption barriers fall once benefit is visible, and because a century is long enough for more than one technological approach.
Evidence that moved the estimate
- The effects are real where they have been measured, including a 35 to 37 percent memory improvement from a model-driven hippocampal prosthesis (hf-evidence-0029).
- Neuralink and Synchron explicitly name enhancement as a medium to long-term objective, so intent exists among the best-funded developers (hf-evidence-0033).
- Elective surgery on healthy people at scale has strong precedent: cosmetic procedures are routine, carry real risk, and treat no disease, which shows the adoption barrier is not categorical.
- Every obstacle found is a maturity curve rather than a wall. Durability, bandwidth, surgical risk and regulatory category have all moved substantially for other technologies inside a century.
- The seed’s own day-one takeaway predicted restoration before enhancement, so finding exactly that is confirmation of the sequence rather than evidence the sequence stops.
Evidence that held it back
- Every human memory-stimulation result located is in a clinical population with clinically placed electrodes: 25 epilepsy patients for the 15 percent recall effect, 8 stimulated epilepsy patients for the 35 to 37 percent prosthesis effect (hf-evidence-0028, hf-evidence-0029).
- The strongest closed-loop result fires stimulation precisely when a classifier predicts POOR encoding, making it restorative by construction rather than enhancing (hf-evidence-0028).
- Hardware does not last a lifetime: mean electrode yield 35.6 percent declining to 34, three of fourteen arrays never producing useful signal, and a longest follow-up of 7.6 years (hf-evidence-0030).
- A healthy recipient would accept roughly 2.8 percent haemorrhage and 3.4 to 3.8 percent intracranial complication risk against no deficit at all (hf-evidence-0031).
- No regulator has a category for a device indicated to enhance a person with no medical condition (hf-evidence-0032).
- Current invasive interfaces sample about 1,000 neurons of roughly 16 billion cortical neurons (hf-evidence-0033).
What would move it next
The first regulatory submission for an enhancement indication in people with no medical condition; and the first replicated demonstration of memory improvement above baseline in healthy volunteers rather than in patients.