AI / Wanderings 2026
Scent
AI is learning to read the chemistry in the air. The interesting part is not perfume. It is the food we waste, the faults we miss, and the people whose health data may soon be floating around a room.
A food-production line still has a slightly medieval moment in it. Somebody opens a package, smells it, and decides whether something has gone wrong.
That person is doing difficult work. A smell is a crowded chemical signal, mixed with temperature, humidity, the last thing in the room, and a lifetime of memory. Electronic noses use sensor arrays that react to volatile molecules, and software that can learn their pattern.
That is already useful. Electronic-nose systems are used for food and beverage quality control, packaging faults, contamination, oxidation, and product consistency. Alpha MOS documents these uses in production. In medicine, the FDA lists a breath-VOC test whose algorithm helps assess a narrowly defined risk of heart-transplant rejection. It supports a clinician; it does not replace the biopsy. That distinction is in the FDA record.
The near future gets interesting when smelling moves from a laboratory instrument into the places where people make decisions.
Waste
Food waste is often a timing problem. A retailer knows a date on a label. It rarely knows the condition of this particular crate of berries after a warm lorry, a cold warehouse and a delayed delivery.
A sensor at the receiving dock could compare the volatile profile of the fruit with the profile of fruit that still has a day or two of life in it. It could give the buyer a useful condition assessment and a reason to move it today, discount it, turn it into jam, or stop it entering the shop at all.
That is a small decision with a large number behind it. Better sensing can keep good food in circulation and make bad stock visible before it becomes a customer complaint.
Care
Breath, skin, wounds, medicines and rooms all have chemical signatures. A future care-home system could notice that a resident's room has changed and ask a carer to look in. A hospital could use breath chemistry to decide who needs a closer examination.
The technology should remain in its lane. A pattern can raise a flag. A human should decide what the flag means. The NHS, insurance companies and every enthusiastic founder in health tech need to remember that a helpful early warning can become a very efficient way of making people anxious.
Consent matters here. So does restraint. Air in a bedroom is intimate data.
Industry
Factories have always had smells. A hot bearing, a leaking solvent line, a batch that has started to spoil, a wastewater process that is drifting. Experienced operators know some of them by heart.
AI can give that experience a memory and a clock. It can record a chemical pattern, compare it with previous incidents and alert the person who can fix the pump or stop the line. For known life-safety hazards, certified single-gas detectors will still do the job. A broader chemical pattern becomes valuable when the fault has not yet announced its name.
The same applies outside the factory fence. Cities argue about bad air because residents smell a problem while the evidence arrives later, if at all. Networks of sensors will not settle every dispute, but they can give a community timestamps, locations and a record that is harder to wave away.
Machines
Robots see well enough to avoid a chair. They hear well enough to answer a question. The next useful ability is noticing that a battery is heating up, that a bin is fermenting, or that a cleaning process has failed.
A household robot with that ability would have a more useful connection to the physical world. Wisdom is not required.
Rules
The obvious risk is that smell data becomes another extractive data stream. A sensor in a workplace can reveal smoking, solvents, food, illness and who was there. A sensor in a home can reveal far more. The fact that data comes from air does not make it public.
We should set the rules before this becomes cheap enough to disappear into every ceiling. Clear consent. Local processing where possible. Short retention. A human explanation for every health or employment decision. Independent testing for accuracy across humidity, seasons, buildings and real people rather than a nice tidy lab.
The future of AI will be more physical. Machines will see, listen, touch and smell. The good applications will help people catch problems earlier and waste less. Build those systems with the same care we expect from the people who use them.