FoS II · Midterm 2

Lecture 11 — Brain-computer interfaces and neural function

A BCI does not repair the nervous system. It reads signals from a part of the pathway that still works and routes them around the part that doesn't.

The normal motor pathway

  • Motor cortex plans and initiates a voluntary movement; neurons there fire in patterns that encode intention and direction.
  • Signals travel down axons through the brainstem into the spinal cord.
  • Motor neurons in the spinal cord send signals out along peripheral nerves.
  • Nerves release a chemical signal at the muscle, and the muscle fibers contract.

Break any link and movement fails, even though the other links may be perfectly healthy. In a cervical spinal cord injury, the motor cortex still generates normal intention signals and the muscles are still capable of contracting; the connection between them is interrupted. Quadriplegia means all four limbs are affected.

This is exactly why a BCI can work: it taps the intact cortex above the injury.

Motor pathway with a spinal cord injury and a BCI bypass The pathway runs from motor cortex through the spinal cord and a peripheral nerve to muscle. An injury in the spinal cord breaks the link, so no signal reaches the nerve or muscle, though the cortex and muscle are both intact. A BCI records from the intact cortex, decodes the signal, and sends it around the break to an external device. The injury itself is untouched. Motor cortex intact: still plans movement Spinal cord injury link cut no signal below Peripheral nerve Muscle intact: can still contract BCI record → decode External device cursor, keyboard, robotic arm Bypass, not repair The injury is untouched, and the nerve and muscle still receive nothing.
Bypass, not repair. A BCI taps the intact cortex above the break and routes the signal around it to a device. The injury is untouched: that is device control, not restored function.

How a BCI works

  • Record: electrodes detect electrical activity from neurons in the motor cortex.
  • Decode: software interprets the firing patterns as an intended action, such as "move the cursor up and to the right."
  • Act: the command drives an external device (cursor, keyboard, robotic arm, wheelchair).
  • Calibrate: the user imagines or attempts movements while the system records what the patterns look like, building a model it can then use. Recalibration is often needed as signals drift.

Recording methods

Noninvasive (EEG on the scalp)Electrodes on the brain surfaceIntracortical electrodes (inside the cortex)
SurgeryNoneYesYes
Signal qualityLow: blurred and attenuated by skull and scalpModerateHigh: activity of individual neurons or small groups
Control achievableSlow, coarseBetterFast, fine-grained
RisksMinimalInfection, bleeding, surgical riskSame, plus tissue reaction and electrode movement over time

The central trade-off: better signal requires more invasiveness.

BCI recording methods by depth A cross-section of scalp, skull and cortex. Scalp EEG sits on the scalp: low, blurred signal, no surgery. Surface electrodes sit on the brain under an opening in the skull: moderate signal, surgery required. Intracortical electrodes go into the cortex next to neurons: high signal from single neurons, surgery plus tissue reaction. Deeper means better signal and more risk. scalp skull cortex Scalp EEG signal: low, blurred surgery: none Surface electrodes signal: moderate surgery: yes Intracortical signal: high, single neurons surgery + tissue reaction signal deeper: better signal, more risk
Better signal costs invasiveness. Each step closer to the neurons sharpens the signal and adds surgical risk.
  • Flexible threads carrying many electrodes, intended to move with the brain and cause less tissue damage than rigid probes.
  • A surgical robot places them precisely while avoiding blood vessels.
  • Signal processing on the implant extracts neural spikes; data is sent wirelessly.
  • Decoding software converts activity into cursor movement, and the user trains with it.

The PRIME reading, and how to read it

The study's stated aims are to evaluate the safety of the implant and the surgical robot and to assess initial functionality for people with quadriplegia controlling external devices.

The instructive part is what went wrong: in the weeks after the first participant's surgery, a number of threads retracted from the brain, leaving fewer working electrodes and lowering his performance, measured in bits per second of cursor control. Neuralink responded by making the recording algorithm more sensitive to population-level signals, improving the translation of signals into cursor movement, and improving the interface, and reported that performance recovered and then exceeded his earlier scores.

What this does and does not show:

  • It does show that useful control can be recovered through software after hardware degrades, and that decoding quality matters as much as electrode count.
  • It does not show that the implant is stable long term, that the spinal injury improved, or that any biological function was restored.
  • It is a safety and feasibility trial with very few participants and no control group. Results from one or two people cannot establish general effectiveness.

Three levels of claim, kept separate

  • Device control: the participant can move a cursor, play a game, use CAD software. Demonstrated.
  • Independence / digital autonomy: the participant can do daily digital tasks without another person's help. A real benefit, and a different claim from control.
  • Repairing the nervous system: the damaged pathway is restored. Not demonstrated, and not what a BCI does.

When a question describes a BCI improvement, decide which level it supports. Technical improvement (better decoding, more electrodes, faster calibration) belongs to level 1 or 2, never level 3.

Three levels of BCI claim Level 1, device control: demonstrated. Level 2, independence or digital autonomy: a real benefit and a different claim. Level 3, repairing the nervous system: not demonstrated and not what a BCI does. Technical improvements such as better decoding, more electrodes and faster calibration point to levels 1 and 2 only; the arrow to level 3 is crossed out. LEVEL 3 Repairing the nervous system Not demonstrated; not what a BCI does LEVEL 2 Independence Real benefit, but a different claim LEVEL 1 Device control Demonstrated: cursor, games, CAD Technical improvement better decoding more electrodes faster calibration
Better technology tops out at level 2. Better decoding, more electrodes or faster calibration can support control and independence claims, never a repair claim.

Check yourself

  • Why does a BCI work for spinal cord injury but would not help someone whose motor cortex itself was destroyed?
  • Why is thread retraction a biological problem as much as an engineering one?
  • A participant's performance doubles after a software update. What exactly improved?

Further reading

Outside sources, opening in a new tab. The exam is on the course material; use these to check your understanding, not to learn new facts.

Tip: the ← and → arrow keys move between sections.