WATCH / RESEARCH NOTES · CLOSED-LOOP SYSTEMS, INTERPRETATION & ETHICS

I looked at The hardest part of achieving closed-loop neuromodulation from TTP alongside Rosalind Picard’s Affective Computing, Wearables and Social Robots because they feel connected in a useful way. One focuses on the technical difficulty of interpreting biological signals reliably, while the other asks what happens ethically when technology becomes capable of sensing and responding to information about a person’s internal state.

Together, they helped me see that the difficult part of a bioresponsive system is not simply collecting data. The harder problem is deciding what that data means, what the system should do next, and how much control the person should retain.

The hardest part is interpretation

TTP describes one of the main challenges in closed-loop neuromodulation as translating complex, noisy biological signals into reliable therapeutic decisions. The sensor itself is only the beginning. The system also needs an algorithm that can distinguish meaningful change from noise, movement, artefacts and other competing signals.

This connects directly to my project because airport travel is full of confounding factors. Heart rate, HRV or EDA may change because someone is walking quickly, carrying luggage, concentrating, feeling excited, anxious, overwhelmed or physically hot.

That means the system cannot simply use a fixed threshold and label the person as “stressed”.

Key design implication: the closed loop is only as responsible as the interpretation layer between sensing and response. My project needs to communicate uncertainty rather than presenting an algorithmic inference as fact.

A closed loop needs more than sensing

Thinking about the TTP explanation helped me break the system into clearer stages:

sense → interpret → decide → respond → sense again

This makes the role of AI much clearer. AI could sit mainly within the interpretation and decision layer, identifying patterns across multiple biosignals and contextual information rather than relying on a single raw measurement.

For my project, that might mean comparing a traveller’s current physiological state against their own baseline, while also considering movement or journey context, before deciding whether a change is significant enough to offer support.

The important word is offer. The system does not need to take control away from the user in order to be responsive.

Picard adds the ethical problem

Picard’s talk broadens the issue considerably. She discusses the potential benefits of affective technologies in areas including autism, epilepsy, anxiety and depression, but also raises the problem of technologies sensing emotion or affective information without a person knowing or consenting.

That immediately feels relevant to my project. Physiological information is extremely personal. A system designed for accessibility could become uncomfortable very quickly if it began sharing or interpreting that data without the traveller’s control.

The technology may be capable of sensing more than the user has consciously chosen to communicate. That means the interface has to make clear what is being sensed, what the system thinks it means, and what will happen with that information.

From “can we detect it?” to “should the system act on it?”

Putting these two talks together changes the question for me.

The technical question is:

Can the system identify a meaningful pattern in noisy physiological data?

The design and ethical question is:

Even if it can, what should it be allowed to do with that information?

For my project, I do not think the answer should be automatic environmental control or an invisible intervention that happens without explanation. A more appropriate model would be:

physiological change detected → system expresses uncertainty → user is offered a response → user accepts, changes or dismisses it

That still creates a closed loop, but the autistic traveller remains an active part of it.

Why this matters specifically for autism

This feels especially important in an autism accessibility context because the project is supposed to increase independence, autonomy and control, not replace one kind of external control with another.

If an algorithm decides that someone is overwhelmed and automatically changes their route, contacts staff or alters their environment without permission, the system could become paternalistic very quickly.

Instead, the sensing layer could remain private and discreet, while the response layer gives the user options. The technology could say, in effect:

“Your physiological pattern has changed from your usual baseline. Would anything be useful right now?”

The user could then choose from options such as quieter navigation, haptic regulation, audio support, breathing guidance, finding a quieter space, or simply dismissing the prompt.

Privacy and consent are not separate from the interface

Picard’s talk also makes me think that privacy cannot just sit in the terms and conditions.

If my project ever became a real system, I would need to consider whether physiological processing happens locally on the wearable or phone, whether raw data are stored, whether airports or staff can ever access it, and whether the user can pause sensing completely.

Those choices are part of the experience design.

Transparency also becomes important. Instead of saying “You are overloaded”, the interface might say something more honest such as “Your signals have changed more than usual” or “I noticed a sustained change — would you like support?”

That language recognises the limitations of the technology.

What I am taking forward

These two sources feel connected because they describe the two sides of the same closed-loop problem.

TTP shows me that biological interpretation is technically difficult. Picard shows me that successful interpretation also creates ethical responsibility.

For my project, that leads to several principles:

• biosignals should be interpreted against personal baseline and context;
• the system should combine information rather than trust one signal in isolation;
• AI should identify patterns without pretending to know exactly what the person feels;
• uncertainty should be visible in the interface;
• sensing should be transparent and pausable;
• physiological data should remain as private as possible;
• support should be offered rather than imposed;
• the autistic traveller should remain part of the closed loop.

The strongest shift for me is that my project should not be designed as: body → algorithm → automatic intervention but more like: body → sensing → contextual interpretation → transparent prompt → user choice → response → sensing again. That feels much more consistent with the project’s focus on autonomy.

Sources