Seminar Opener · Discussion, not lecture

What Is Neuroengineering?

Seven readings, eleven cases, and no two readings alike. An argument the field has not settled and this session will not settle either.

Press SPACE to advance · ESC for overview · S for facilitator notes

How this session works

  • You will write down a definition in the next two minutes, before you see anyone else's.
  • We will then run four cases at it. At least one will probably give it trouble.
  • Only then do we look at seven readings of the question, and at what the field has actually published about itself.
  • You leave with a better defended position, not a correct one. There is no answer slide.

The one rule

Attacking someone else's definition is easy and cheap. The useful move is naming what your own definition has to throw out, and saying whether you can live with it.

Do this now, before the next slide

Write one sentence: what makes something neuroengineering rather than neuroscience, biomedical engineering, or computer engineering?

Keep it. You will be asked to revise it at the end, and the revision is the point of the session.

2 minutes · silent · no phones

The two answers that usually come first

Applied neuroscience

Neuroscience discovers, engineering applies the findings.

Engineering for neuroscience problems

Engineering principles brought to bear on questions about the nervous system.

Show of hands. Is what you wrote a version of one of these?

Both are defensible. Both are about to have a hard four minutes.

Case 1 of 4

The cochlear implant

1960s to 1980s

Pursued over the objections of prominent auditory scientists who had principled reasons to believe crude electrical stimulation could never carry speech. It went on to give useful hearing to hundreds of thousands of people.

Does your definition let it in? If neuroengineering is applying neuroscience findings, which findings was it applying?

The honest complication: it did apply established tonotopy. What it defied was the view that fine temporal coding was required. So it applied the neuroscience that was right and ignored the part that was wrong, which you can only tell apart afterward.

Case 2 of 4

The voltage clamp

1940s instrument, behind Hodgkin-Huxley

The Hodgkin-Huxley equations of 1952 rest on measurements only the voltage clamp could make. The pattern repeats: single-channel recording did not exist until the patch clamp in 1976, and human functional imaging went from a few PET centers to a mass activity once BOLD contrast, described in 1990, removed the need for a radiotracer.

The instrument came first. If engineering only applies what neuroscience discovers, why are so many landmark contributions to neuroscience instruments?

Case 3 of 4

A neuromorphic accelerator

event camera for robot vision, never touches tissue

Takes organizing principles from nervous systems and applies them to computation. Published in the field's journals, taught in its departments. No nervous system anywhere near it.

The arrow runs backwards. If your definition is about serving the nervous system, this is out, and you have excluded a chunk of the field's own literature.

Case 4 of 4

Neuromarketing and neuroarchitecture

Both deploy real neuroscience toward practical ends: one to advertising, one to building design. Both have their own literature, conferences, and practitioners.

Few would call either one neuroengineering. But if "applying neuroscience to a practical end" is your criterion, on what grounds do you keep them out and the chip in?

The trap closing

On that criterion, whatever admits the chip admits the ad agency. Same move, same direction, different destination.

The squeeze

Read it narrowly

"Neuroscience findings applied toward biological ends."

Throws out the cochlear implant, the voltage clamp, and DBS. The founding achievements.

Read it broadly

"Neural principles applied toward any end."

Lets in neuromarketing, neurolaw, neuroeconomics, and neuroarchitecture. Stops naming a field.

No setting of this dial keeps the founding cases and excludes the ad agency. The readings that manage it do something else entirely, which is the next slide.

The problem is not any single case. It is one phrase carrying two incompatible meanings and borrowing whichever one survives the challenge.

Seven readings, built for argument

Not a survey of what practitioners believe. Constructed to span the space of defensible answers, then checked against the published record.

1Applied, broad

Any use of neural principles, whatever gets built.

2Applied, narrow

Neuroscience findings applied to a biological or clinical end.

3Methods for science

Engineering built to answer questions about the nervous system.

4Interface discipline

Building the link between a device and a nervous system.

5Clinical

Restoring, replacing, or augmenting function in a living organism.

6Constraint

Engineering where a living nervous system sets the requirements. Take it away and the design changes or makes no sense.

7Institutional

Whatever the field's journals and departments treat as their own.

Yours

Which of these is your slide-3 sentence closest to?

The four cases, now scored

Same cases, same order, positions now named. Read it as how seven readings split, not as an answer key.

Case 1
Applied
broad
2
Applied
narrow
3
Methods
4
Interface
5
Clinical
6
Constraint
7
Instit.
Cochlear implant ARG OUT OUT IN IN IN IN
The voltage clamp OUT OUT IN ARG OUT IN ARG
Neuromorphic accelerator IN OUT OUT OUT OUT OUT IN
Neuromarketing study IN OUT OUT OUT OUT OUT OUT
Neuroarchitecture IN OUT OUT OUT ARG OUT ARG

Column 1 keeps the accelerator, and pays for it by keeping the ad agency. Column 7 keeps the accelerator without the ad agency, but only by deferring to what journals publish.

Eleven cases, seven positions, no two alike

No two positions agree across the table. On four cases, no reading commands more than three of seven votes.

2 / 3 / 2Voltage clamp
1 / 3 / 3Grid-cell model
1 / 3 / 3EEG headband
3 / 2 / 2EPI sequence

in / out / arguable, out of seven

Live matrix

The reading page has all eleven as an interactive table. Click a cell for the reasoning, or add your own column and see which position your calls track.

What the field says about itself

Its own answer is a conjunction

An emerging interdisciplinary research area that brings to bear neuroscience and engineering methods "to analyze neurological function as well as to design solutions to problems associated with neurological limitations and dysfunction." Editorial board definition, in Durand, "What is Neural Engineering?", J. Neural Eng. 4(4) E01, 2007

Position 3 and Position 5, joined by an "as well as", with nothing said about how the two relate.

Then it hands the question back: the scope is "best determined by the scientists and engineers that practice it." Position 7, from the document stating the official definition.

And the record contradicts itself

The standard definition

Neuroengineering is a discipline within biomedical engineering using engineering techniques to understand, repair, replace, enhance, or otherwise exploit neural systems.

The founding conference

The first IEEE EMBS Neural Engineering conference, Capri, March 2003, was convened as a gathering for a field distinct from both neuroscience and biomedical engineering.

Both statements are in circulation. Neither has displaced the other.

Meanwhile the flagship journal's scope claims neuromorphic engineering, computational and systems neuroscience, and neuroimaging, wider than its own official definition covers.

What few would dispute

The definitions differ. These constraints are largely common ground.

The target pushes back

Encapsulation begins in days. Day 400 is not the population you recorded on day 1.

The controller is learning too

The decoder adapts to the user while the user adapts to the decoder.

Observability is rationed

No non-invasive way to watch one human neuron. No budget buys out of the trade.

Harm is often irreversible

One skull, one cochlea, one cord. The failure mode is not a product return.

The spec is often unknown

Nobody hands you the correct neural code. You co-discover it while building.

Timelines outrun careers

Cochlear implants: 1960s trials, mid-1980s approval. DBS: 1987 to 1997.

Why any of this matters

A definition is not decoration. Four things actually turn on it:

  • Curriculum. If neuroimaging is diagnostics and your definition is clinical, does it belong in the core?
  • Study section scope. Whether a proposal is reviewed by people who recognize the work, or by people who think it belongs somewhere else.
  • Hiring and promotion. Whether a candidate building neuromorphic hardware counts as a hire in the field or a hire adjacent to it.
  • What a student is told they are becoming. Arguably the one that matters most, and reliably the one discussed least.

Break out · 15 minutes · groups of three or four

Pick one and argue it

  • Which case embarrasses your position most, and can you accept that exclusion?
  • An accelerator and a prosthesis front end may share most of their circuit design. Should a definition separate them?
  • Every position here excludes something the field's own institutions embrace. Defect in the definitions, or in the institutions?
  • Should a definition exclude bad work, or only misplaced work?
  • Does the field need a definition at all?

Close

Take out what you wrote at the start. What would you change, and which case made you change it?

If nothing changed, say which case you are willing to exclude to keep it. Either answer is a result. "It depends" is not.

3 minutes · then two or three out loud

Take it further

Resource What is there
What Is Neuroengineering? All seven readings, each with its source or a note that none was found, plus the interactive matrix and the published definitions
History and Timeline The actual sequence of instruments and findings, which several arguments turn on
Grand Challenges Why the brain is hard to measure, in physical terms
Neuroethics Where the identity and agency constraints get worked out
Knowledge Base The full topic map the definitions are arguing over

Positions 1, 2 and 6 have no published source. The reading page says so on each card.