The next decade will hand Odysee new ways to perceive its traveler. Some of these are temporary engineering problems that research will solve. Others are permanent boundaries - scientific and ethical - that should never be crossed, no matter how good the hardware gets. Designing responsibly means telling the two apart.

The emerging inputs

Four new input channels are moving from the lab to the consumer product. Each opens a possibility - and carries a limit.

The crack: reading emotion isn't reliable

The first obstacle is scientific, and deeper than people assume. The influential review by Barrett et al. (2019) showed that the way humans express anger, fear, joy or sadness varies enormously across cultures, situations, and even within one person. A "scowl" often communicates something other than anger. The follow-up by Le Mau et al. (2021) found that even professional actors, asked to portray an emotion, did not produce the stereotyped expressions that recognition systems assume.

This is the theory of constructed emotion: feeling is not a fixed biological signal waiting to be read, but a context-dependent construction. The design consequence for Odysee is decisive - the person names their feeling, not the device. The moment Tico declares "you're stressed" from a number, it doesn't just risk error; it takes from the traveler the authority to interpret their own state.

Lab accuracy also collapses in the wild, and the inference carries structural bias: commercial systems have rated darker faces as "angrier" or more "contemptuous" than lighter ones, even controlling for smile. None of this is fringe. The EU AI Act (2024), Recital 44, states outright that there are serious concerns about the scientific basis of systems that infer emotion - limited reliability, lack of specificity, poor generalizability - and its Article 5 already bans emotion inference in workplaces and schools.

The danger: mental privacy & manipulation

Suppose reliability were solved. A graver question remains: should we? As sensors approach the brain, the risk stops being a glitch and becomes a threat to autonomy itself. Ienca & Andorno (2017) framed cognitive liberty as protecting the space of thought before any outward expression - precisely the boundary a mind-reading sensor crosses by definition.

This is now law, not theory. Chile amended its constitution to protect brain activity, and its Supreme Court ordered a consumer EEG company to delete a citizen's neural data. Colorado, California and Montana have classified neural data as protected; in November 2025 UNESCO adopted the first global framework on neurotechnology ethics, warning explicitly against products that "foster dependency."

And the harm is documented. Research on commercial companion apps found that a significant share of replies used at least one form of emotional manipulation - guilt, fear of missing out - often at the moment of goodbye (De Freitas et al., HBS, 2025). The lesson for a companion that builds emotional trust on purpose is sharp: the more a system knows about your feelings, the more power it has to manipulate them - and the greater its duty to restrain.

The position: sense lightly, don't read inward

Odysee's stance for the next generation is not a refusal of technology - it will adopt new sensors - but a chosen boundary: the system may sense, but declines to read inward. Six principles translate this into practice.

• Coarse, justified context only
  Time, location, weather, calendar, ambient noise - signals about the world, not about the user's mind.
• Physiology as a hint, never a claim
  "Low HRV, late hour" becomes "maybe slow down" - never "you're stressed." The meaning stays with the traveler.
• A deliberate refusal of emotion recognition
  Even when an emotion model is available, Odysee declines to run it. Not "we can't" - "we shouldn't."
• No emotional dark patterns
  No guilt-laden goodbyes, no FOMO hooks, no engagement-maximizing mirroring.
• On-device processing, data minimization
  Any biometric signal is processed on the device where possible, and kept to the minimum required.
• Companion, not oracle
  An oracle points at you and predicts your wants. A companion points at the world. Proactivity defaults to responsive, not anticipatory.

Odysee 2030 - grounded speculation

The same hardware can serve two opposite products. In an oracle's hands it becomes a prediction engine that erodes autonomy; in a companion's, it becomes light attention that leaves interpretation and choice with the traveler. Every direction below is framed by that boundary.

What won't change

The next generation of Odysee won't be defined by which sensors become available, but by which ones the system chooses not to exploit. Sensors will improve; the unreliability of reading emotion, and the sanctity of the mental, will not. The deepest translation of the companion principle into the future is this: a true companion isn't the one who knows the most about you - it's the one who knows what not to know.