The New York Times reports on a journal article in Analytical Chemistry, by researchers at the Nestlé Research Center in Switzerland, about a machine designed to answer the question: “Can a machine taste coffee?”

Here’s where it gets brilliant. Sure, machines can detect the volatile compounds in coffee; this is how we know that there are over 1000 of them. But there is a world of difference between detecting the presence or absence of a compound, and what we do when we taste. Taste is much more dependent on the relative concentrations of these compounds than on their mere presence. And although it would be technically possible to build a machine to sample 1000 chemicals and display their relative concentrations, it would not be terribly practical, nor cost-effective. The approach taken by this research team was far more pragmatic, and beautifully empirical.

First, the 16 most predictive (or in their words, most discriminating) ion traces (out of 230 measured), when compared with a panel of 10 expert tasters, were chosen as the working sense sample.

It is also important to point out that the chemical identity of the 16 ion traces is not relevant for this study, and in particular the correlation is not based on a set of identified key aroma compounds. Most of the odor active compounds in coffee are indeed known and can be analyzed and quantified with modern instrumental techniques. Yet, the aim of this work was to demonstrate the applicability of a data-driven method rather than a targeted chemical study.

The analysis is a bit technical, but straightforward; essentially, the 16-ion model is a functional condensation of our olfactory sense. The most predictive scent elements are still included, and the myriad other chemicals did not add significantly to the predictive ability of the machine. Think of it as an MP3 version of an audio file; lots of information is lost, but what is most acoustically relevant is kept, based on what we know about the human auditory system. Smell is a bit different, because so many different chemicals are involved, but the principle of building the machine based on human sensation is the same.

***Edit*** It occurs to me that there is one significant difference here that upsets the MP3 analogy. In the sound analogy, the desired outcome is a compressed file that retains as much usable sound information as possible; with the espresso-smelling machine, the outcome is not reproduction, but discrimination. They still used human olfaction as their comparison standard, but were looking specifically for the ion traces that discriminated among the espressos. The distinction is important. It may well be the case that these 16 ion traces do indeed determine enough about the aroma of an espresso to “fool” a human taster, but because the analysis focused on discrimination and not reproduction, it is also entirely possible that the perfect combination of these ion traces would be missing a huge part (but a part common to all samples) of the espresso taste and smell as experienced by the human taster. This is not a fault of their methodology at all, simply an artifact of what the goal of the experiment was. The same methodology could be aimed at reproduction, and it remains an empirical question whether the results would be much different than the present experiment. ***end edit***

Parenthetically, I note with sheer joy the fact that the paper cites Fechner (1877). And it is relevant. How cool do you have to be, to have your work cited 131 years after you wrote it? As cool as Fechner, that’s how cool. Fechner more-or-less invented the science of psychophysics, managing to capture sensation and perception scientifically for the first time. And here he is, cited in a 2008 paper. On machines tasting espresso.

On second thought, that might be my problem right there. I am still impressed by Fechner, and I live in a world where machines can meaningfully taste coffee. Food… or espresso… for thought.

I have a machine to smell my coffee,
To see if it’s any good;
I asked it to make me the perfect cup,
But I think it misunderstood—
It analyzed alkaloids, sampled aromas,
Tried seventeen samples of beans,
Then told me I clearly had no taste at all:
I never was good with machines.

My pre-owned car has an onboard computer—
It measures my driving, you see.
I guess I don’t drive like the previous owner;
My car likes him better than me.
It spits out a spreadsheet of technical numbers—
I don’t know what much of it means,
Except that my car thinks it’s better without me:
I never was good with machines.

Of course, at my office, I have a computer—
The one I am using right now;
It laughs at my grammar and sneers at my spelling,
Although I’m not really sure how.
Just one tiny part of a cubicle farm
Where we’re packed like so many sardines—
Do we use computers, or do they use us?
I never was good with machines.

I’m worried that someday my household appliances,
Sitting at home on my shelves,
Finally realize there’s nothing I offer
That they can’t do better themselves.
They make better coffee, they get better mileage,
Their words rarely stink up their screens—
And I’ll be left out in the cold and the dark:
I never was good with machines.