My guess from having formerly been married to a chemist is that machine-based chemistry is one of the most difficult areas of science. Chemistry is still equal parts art and science in that the techniques that will give accurate scientific results depend greatly on the types of things being measured for. For detecting a particular substance, there may or may not be a cheap or viable detecting device that could be built, depending on whether that thing is normally detected by chromatography, reaction with something else, spectroscopy, etc. For example, carbon monoxide detectors are great and cheap. But I think it's probably just really case by case. Chemistry is just super hard to generalize or systematize. I used to ask her why they don't have Star Trek tricorder type devices yet that can universally analyze substances, and that was more or less her answer.

Edit: for the difficulty of machine chemistry, see also the Theranos debacle.

Carbon monoxide might be easier because it's one specific molecule, and the concentration has to be fairly high (comparatively) to be life-threatening. The paper mentions VOC (volatile organic compounds) which are probably harder to quantify. Detectors aren't always great either. I've worked in labs that didn't use helium detectors, because - at the time maybe - they weren't super reliable, but having them usually meant people would rely on the detectors instead of paying attention to the symptoms of a helium leak.

Smell/organic chemistry is weird, too. Some molecules have similar smells, despite being sometimes quite different. Many molecules have quite different smells, despite being not that different [0], edit: the esters table maybe shows that better, [1].

[0] https://jameskennedymonash.wordpress.com/2014/01/04/table-of...

[1] https://jameskennedymonash.wordpress.com/2013/12/16/infograp...

Oh yeah, I'd forgotten how weird smell in particular was. I seem to remember at the time that there were three competing theories for how smells are even perceived or what gives a substance a particular kind of smell, and each only explained a subset of smells.

For the helium leak, is one symptom a high squeaky voice, or is that at already really dangerous concentrations?

From what I remember, dizziness, headache, inability to concentrate, etc. You'd notice a big/quick release as the liquid helium boils of quite easily. A slow leak is more dangerous, in which case you might even get used to the higher pitch without noticing.

One symptom of a helium leak is that your iPhone might stop working: https://motherboard.vice.com/en_us/article/gye4aw/why-a-heli...

This happened also to the coworkers of a friend of mi e - occasionally everyone's iPhone in their office would stop working. Android phones still worked fine. Later (after this article was published) they found out there was an MRI machine downstairs.

Isn't carbon monoxide also a ridiculously simple chemical? Detecting complex molecules has to be a lot more complicated, doesn't it?

Also noses self repair. You can foul them with another scent for a while but eventually the receptors will recover.

We have a long way to go before we are even remotely as good at nanotechnology as biology is.

I'd give decent odds we'll sooner grow a dog nose in isolation and interpret the neural signals directly than build a machine as good as a dog's nose. (And I'm not saying that's easy, either.)

The mental image of walking through, say, a security checkpoint that contains an artificially grown dog nose that sniffs you as you go by seems as good a writing prompt for dystopic fiction as you could ask for.

There was a British company which attempted to develop drug and bomb-detecting tools powered by bees. They would condition bees to associate the smell of a particular drug or explosive with receiving a morsel of food. Then load the bees into cartridges, put a dozen or so of the cartridges in a wand format and hand them to security staff at airports.

It doesn't appear to have gotten off the ground, I can't even find the name of the company any more.

Progress is being made. Twenty years back, I worked for a company that put a bunch of live cells into microtiter plates and use computer vision to see how they were responding to environmental stimulus. One of the contracts was with the military to detect battlefield biohazards. I've not been involved in some time, but I assume that this tech has progressed in the last twenty years.

> but I assume that this tech has progressed in the last twenty years

or completely shelved and waiting for you to connect the multidisciplinary dots.

>connect the multidisciplinary dots

That's what I had done 20 years ago. So one reason that I returned to software after doing graduate work in neuroscience was the frustration of keeping cells alive while you poke at them. These titre plates full of thousands of different cells pose a problem orders of magnitude more challenging. Perhaps now it would be easier - engineer cells with the properties you need using CRISPR/Cas9.

Bear in mind that the sense of smell is different from the other four senses in that there isn't a separate sense organ in the nose. The Olfactory bulb is actually part of the central nervous system. So to "grow a nose" you really mean "grow a brain".

This is entirely incorrect. The nose has a surface called an olfactory epithelium, in which are buried the actual sensory apparatus. Olfactory receptor neurons in here respond to the odors themselves, and transmit the information to the olfactory bulb. The olfactory receptor neurons are essentially analogous to related structures for the other senses, such as sight (rods and cones) and hearing (hair cells).

I know progress has been made decoding the nerve impulses directly for vision and sound, to the point that we can engineer machines that generate the impulses (pretty much the hardest test there is), with Cochlear implants being standard, off-the-shelf medical tech now [1]. The eye equivalents are harder, but prototypes are being built. (I don't think they're off-the-shelf tech yet, though.)

I have not heard anything similar about the olfactory nerves. I suspect they're going to prove to be much messier. Perhaps not necessarily "complicated" in some sense, but messier. But it shouldn't be impossible.

[1]: https://en.wikipedia.org/wiki/Cochlear_implant

It’s intersting how the sense can get damaged too. Imaging for the problem (anosmia) can be very interesting. https://www.mayoclinic.org/symptoms/loss-of-smell/basics/cau...

I stand corrected. I shouldn't rely on 30 year old memories. This is interesting "There are approximately 1000 different genes that code for the ORs, making them the largest gene family." - wikipedia

I imagine it's probably that it's prohibitively expensive to make one with the qualities required, which is reliable, mobile, can alert others automatically, with a long running time.

Dogs (some breeds more than others) have some of the strongest smelling capabilities on the planet. So much so, that some people theorize that one of the reasons dogs don't traditionally do all that well on the mirror self-recognition test is that it focuses on vision, and as animals that rely so heavily on such a strong smelling capability, it may throw them off (I guess the human equivalent would be if the test was audio only, or audio with a very fuzzy image. If you weren't already familiar with hearing your voice played back to you, how likely would you be to recognize it as yours?)

The search space for which molecule or sets of molecules are being recognized is huge. There are at least 120 different receptor types, and the dynamic range of their detection thresholds is large too. So we're looking at perhaps a trillion different combinations in theory.

Plus the detection threshold can be as low as parts per billion, spread across 300 million receptors in a dog.

So just the mathematics of the search space seem like a pretty good reason why it is a hard problem in general.

It's not that we can't. We're just now beginning to characterize this in dogs now. Next we need to identify the chemical signal and develop hardware to pick up that signal.

If it's the strength of a particular chemical signal -- like dogs can normally smell it, but it's suddenly stronger -- then that seems like it's going to make false detections very difficult. Training a machine to have an innate sense of how smell works under changing environmental conditions seems like it's going to be hard.

Living things are machines, they're just incredibly sophisticated biological ones ;)