What a roadmap is actually for
The grand-challenges roadmap is best read as a map of tradeoffs, not a schedule of arrivals. Its most honest axes are read/write information rate against invasiveness: every point on that plane is a bargain — a scalp EEG buys safety with low bandwidth; an intracortical array buys bandwidth with surgical risk and the longevity ceiling. No technology yet occupies the top-right corner of high bandwidth at low risk, and whether anything can is the field's defining open question.
The generalization–specialization axis
A theme cutting across every challenge is the generalization-versus-specialization tradeoff. A decoder tuned to one subject on one day is accurate but brittle; a model meant to work across brains, sessions, and tasks trades peak accuracy for robustness. Neural foundation models pretrained on many subjects are the field's current bet on breaking this tradeoff — but the universality and transfer limits are unknown: it is genuinely open whether a universal neural code exists to be learned, or whether inter-individual variability imposes a hard floor on transfer.
The grand challenges, named
Strip away the marketing and a stable list of grand challenges remains. Chronic stability: an interface that holds for a decade, dissolving the biocompatibility ceiling. Naturalistic bidirectional control: high degree-of-freedom output with high-fidelity sensory write, confronting the read/write asymmetry head-on. Robust generalization: co-adaptation without daily recalibration. Ecological performance: results that survive outside the shielded lab. These are engineering-and-science challenges — hard, but not forbidden.
How will we know progress? By a task-level yardstick, not a channel count. Fitts's-law throughput measures effective bits per second of control, capturing the speed-accuracy tradeoff a headline number hides. Today's best intracortical cursors and handwriting BCIs reach a handful of bits per second — approaching able-bodied performance on narrow tasks, and a sober benchmark against which every roadmap claim can be checked.
Fitts's-law throughput: over movement time MT to a target of width W at distance D, it reports effective bits per second. Honest, comparable, and hard to game — the kind of metric a mature field measures instead of counting electrodes.
Beyond the device: standards, translation, equity
The gating challenges of the next decade may not be silicon at all. Standards and benchmarking bodies are needed so results can be compared; regulatory science for adaptive AI decoders must catch up to devices that keep learning after approval; and clinical translation — the slow path through trials, reimbursement, and surgical training — is where most promising demonstrations quietly stall.