Continued from yesterday’s post
It’s an interesting idea, that sensor testing might be able to do what both of those profiling systems just drew on, with the input an objective, measured assessment. One problem: related sensor based aspect testing just isn’t there yet. It complicates things that when we taste foods we’re really sensing two different things (aside from feel aspects like astringency, and noticing aftertaste, which is something else). We pick up basic flavors through our tongue (sweetnees, salt, sourness, bitterness, and umami, or savory taste), and also detect aromatic compounds through our nasal passages, a version of the sense of smell.
Electronic tongues and noses can replicate some of both, but there are gaps in getting them to cover our range as effectively, and even broader gaps when it comes to interpretation. It’s all early in the research phase. If there was a strong commercial driver for machines to taste things as effectively as we do they probably would be a lot more advanced, even though most of the current simpler forms of sensing equipment, testing process, and analysis (AI function) have only been developed in the last 20 years or so.
Even the description of the research into that subject is a bit much:
Tea quality prediction using a tin oxide-based electronic nose: an artificial intelligence approach
In this paper, we have (analyzed using a metal oxide sensor (MOS)-based electronic nose (EN)) five tea samples with different qualities… The flavour of tea is determined mainly by its taste and smell, which are determined by hundreds of volatile organic compounds (VOC) and non-volatile organic compounds present in tea. Tea flavour is traditionally measured through the use of a combination of conventional analytical instrumentation and human organoleptic profiling panels… The methods are also inaccurate because of a lack of either sensitivity or quantitative information. In this paper an investigation has been made to determine the flavours of different tea samples using an EN and thus to explore the possibility of replacing existing analytical and profiling panel methods…
I’d love to say more, and I did include a lot more detailed research findings citations in that recent post, but that long summary was still only a limited sample of the range of looking into the potential. You can kind of see where it’s going, and a bit about how it works, but it’s all very complicated, technical, and at present very limited in function. I’ve talked to the Teapasar vendor about it and there’s more to pass on about what they’re doing, but filling that in will require more research.
Images provided by author
Ultimately I think that taste is very subjective. What one person likes can be different from another. But yes, identifying areas like salty, sweet etc, using AI is the next wave of progress. I’m both fascinated and ambivalent about applying that to tea, this most ancient herb. Remember I’m a purist so technology isn’t always welcome in my world! I very much appreciate this interesting information however. I look forward to hearing more about it over time.
I can completely relate to that. The first question that comes up is why to do it at all. That subject exploration started with discussion of to what extent testing could verify quality level, which according to current potential relates to checking markers, not replicating tasting. But after considering that it’s natural to look into to what extent machines could do that, and what it would involve. An AI program would need to learn what human subjective preference involves to master it, and we’re a long way from that. Mapping out roughly what a person would experience in terms of aspects could be done now, with the right collection of current capability. There just isn’t an economic drive to do it, so development occurs bit by bit instead, study by study. It’s not just basic flavors (sweet, salty) that testing / sensors can pick up now, current potential also relates to aromatic component mapping (eg. which floral aspect is being sensed, at what level). It’s not even close to being mapped out and practical as an overall output, but most of the pieces to that puzzle have been roughed out, just not quite all, and it hasn’t been combined. Given much of what has been developed occurred in the past 10 years 10 more should see a lot come together, even without a clear purpose driving it.
This will adversely affect those with golden noses who work within the perfume industry.I understand those are some very well paid jobs!
Eventually, yes. But for the next 20 years they would be even more in demand for teaching the machines (AI programs). After that who knows what the world will be like. I suspect machines will never be able to completely duplicate our judgment of subjective preference, because they’ll only be communicating what they learned about patterns in the past, not well suited for experiencing and evaluating new patterns and combinations. That kind of more sentient AI is a long way out, probably. In the movies AI programs teach themselves to learn but we’re not at all close to replicating that.