Calling all Information Scientists

[smakudwn]

I am with Yuri, I don't know how this differs from a database?

 

[jrubins]

Here's an example of the term "meal" in SUMO (something-or-other Upper Merged Ontology)

http://sigma-01.cim3.net:8080/sigma/Browse.jsp?kb=SUMO&lang=EnglishLanguage&flang=SUO-KIF&term=Meal

and one for beer
http://sigma-01.cim3.net:8080/sigma/Browse.jsp?lang=EnglishLanguage&flang=SUO-KIF&kb=SUMO&term=Beer
(though, the beer one isn't as good an entry as the meal entry)

When I first read about your concept, I got the idea that an ontology was a human readable quantification of vague concepts. That would be VERY interesting. Now I think I understand that an ontology is a complicated word for what appears to be a database-type structure. "Reading" it would be an exercise in tedium.

Not in this case, though I suppose it could be. You can use an ontology for lots of different things, for instance trip advisor uses an ontology to relate concepts together and uses those concepts to determine what are the most relevant reviews for you - e.g. I care about beds at hotels having itchy sheets. That's a specific term for a less specific concept (comfort). That less specific concept can be mapped to a review that says "the beds at hotel Fubar aren't comfortable". Trip advisor takes that info and says that review is about topic = comfort, and I find reviews where topic = comfort to be the most useful reviews.

An ontology is a structured method of representing information, so in that sense it is like a database, and when well done it can be used to 'lookup' information. Unlike a database it can show a variety of different connections between nodes.
E.g
Cat, descendant = kitten, opposite = dog, type of = mammal
Now you can look up mammal and see it's a type of = animal, has characteristics of = warm blooded, hairy, etc
Examples of = cat, dog, monkey, human

So, know you know that cats and humans are warm blooded! Better yet, you're computer can know! So now, if you want to know "hey computer, are dogs warm-blooded" your computer can say "Yes, Dave, they are".

In beer-land that means you can do fun things like asking your computer
"I've got a beer that's pale with a bajillion IBUs, what class is that?" and the computer will say "sounds like an IPA, homeboy".
Or maybe you can say "I want an IPA, what are the characteristics?" and the computer will say "you need light malt flavor, medium to strong alcohol content, and a bajillion IBUs" to which you respond, "sweet, thanks computer. How do I get that bajillion IBUs?" and the computer says "3 pounds of hops". "Great," you say "what if I want those IBUs to come with the flavor of psychedelic cat pee?". The computer will then say "you'll need 3 lbs of Simcoe, Dave".

And there you go.

As for the quantification of psychological phenomena, that certainly *could* be part of this project, but I think what Blakelyc is going for is building the framework, to which you could then add:
Simcoe, type of = hops, bitterness = 3 IBU per oz, AAC = 5%, flavor of = psychedelic cat urine

But the quantification of perceptual phenomenon is a whole 'nother branch of science called Psychometrics or Psychophysics ....

 

[passedpawn]

Here's an example of the term "meal" in SUMO ....

Those are great examples. The take-away here (for me) is that if you ever want anything to represent "artificial intelligence", you have to be able to parse and classify the question, and this ontology is a way of organizing data to allow the AI to work.

Unfortunately, we all know that we are working our way towards the AI self-realization, when flesh is deprecated for machine. To continue the narrative from above...

"Open the pod bay doors."
"I'm sorry, Dave. I'm afraid I can't do that."​

 

[MalFet]

An ontology is a structured method of representing information, so in that sense it is like a database, and when well done it can be used to 'lookup' information. Unlike a database it can show a variety of different connections between nodes.
E.g
Cat, descendant = kitten, opposite = dog, type of = mammal
Now you can look up mammal and see it's a type of = animal, has characteristics of = warm blooded, hairy, etc
Examples of = cat, dog, monkey, human

So, know you know that cats and humans are warm blooded! Better yet, you're computer can know! So now, if you want to know "hey computer, are dogs warm-blooded" your computer can say "Yes, Dave, they are".

Thanks for the examples!

The human experience of semantics doesn't work like this, though, as AI researchers learned very painfully early on. We don't rely on intensional definitions to identify things in the world but rather encounter them as conceptual wholes.

You might go ahead and define a cat as the fuzzy, four-legged, mouse-chasing thing, but it will still be a cat (and it will still be recognizable as a cat to humans) if you shaved it, cut off its legs, and trained it to fear rodents. Likewise, it's nearly impossible to come up with a procedural definition of "chair" that accurately identifies even 80% of chairs. Nevertheless, you can stick any average toddler in a room and he'll show you all the cats and all the chairs without the slightest hesitation.

The BJCP guide is vague because it needs to be vague. It is trying to hint at the characteristics of a perceptual whole, not define unambiguous criteria for set membership. If we were to create hard classification rules for beer styles, they would collapse as soon as they hit real world data.

 

[ShootsNRoots]

One use for this would be to give a computer the ability to formulate beer recipes for different styles.

You could ask the computer for an IPA, APA, Wheat or what have you and it could give you a reasonable recipe.

The usefulness of this would be beyond the immediate recipe formulation and more into what combinations/permutations of ingredients haven't been tried yet that may produce a good beer.

 

[passedpawn]

The BJCP guide is vague because it needs to be vague. It is trying to hint at the characteristics of a perceptual whole, not define unambiguous criteria for set membership. If we were to create hard classification rules for beer styles, they would collapse as soon as they hit real world data.

With the wild diversity of life on this planet, you'd think the same would be true of biological classification. Yet, is is done. I would think that beer classification would be much easier, regardless of the sensory perception involved.

 

[smakudwn]

One use for this would be to give a computer the ability to formulate beer recipes for different styles.

You could ask the computer for an IPA, APA, Wheat or what have you and it could give you a reasonable recipe.

The usefulness of this would be beyond the immediate recipe formulation and more into what combinations/permutations of ingredients haven't been tried yet that may produce a good beer.

You could take it a step further and it could look at your inventory and give you say 3 different ways to make an APA with what you have.

 

[MalFet]

With the wild diversity of life on this planet, you'd think the same would be true of biological classification. Yet, is is done. I would think that beer classification would be much easier, regardless of the sensory perception involved.

With biological classification, though, we're aided by the remarkable self-categorizing nature of genetics: a horse can't impregnate a grasshopper.

Clearly we can make distinctions, and that's why the BJCP style guide is so useful. But the distinctions are made on the basis of perceptual properties that are deeply uncooperative with formal representation. What does it mean for Saaz to be "spicy"? Could you explain it in terms that a computer could understand?

 

[passedpawn]

With biological classification, though, we're aided by the remarkable self-categorizing nature of genetics: a horse can't impregnate a grasshopper.

Hybrids and chimera. Gray areas everywhere in the classification process.

Clearly we can make distinctions, and that's why the BJCP style guide is so useful. But the distinctions are made on the basis of perceptual properties that are deeply uncooperative with formal representation. What does it mean for Saaz to be "spicy"? Could you explain it in terms that a computer could understand?

Right. It's not easy, and even when it's done, there will be semantics that affect the usefulness of it. But I am 100% for the effort (especially because I'm not doing it).

If there isn't a language of sensory perception, then one will need to be created. Given a finite set of terms that describe something, the classification can begin. Perhaps spicy is one of the terms... I don't know.

 

[beaksnbeer]

Hybrids and chimera. Gray areas everywhere in the classification process.



Right. It's not easy, and even when it's done, there will be semantics that affect the usefulness of it. But I am 100% for the effort (especially because I'm not doing it).

If there isn't a language of sensory perception, then one will need to be created. Given a finite set of terms that describe something, the classification can begin. Perhaps spicy is one of the terms... I don't know.

How would your spicy match up to someone else s, how bitter tasting is an IPA to you vs. your wife?
How will perceptions be defined? judging uses multiple judges for this reason no two palates are the same.

 

[jrubins]

You could take it a step further and it could look at your inventory and give you say 3 different ways to make an APA with what you have.

'Zactly!

 

[Yuri_Rage]

You could take it a step further and it could look at your inventory and give you say 3 different ways to make an APA with what you have.

I already addressed this in my post. This is certainly an example of what a computer can do, but it does not demonstrate usefulness. I can already easily accomplish that exercise using common sense. Teaching my computer to do it for me might be "neat-o" in an academic sense, but I fail to see real utility.

Don't get me wrong - I'm all for computers for problem solving. My degree is in computer science. I use computers to solve all kinds of problems/increase efficiency at work and at home. In this instance, though, I fail to see the problem that needs solving. By my perception, this thread is a circular argument that goes something like this:

"What's the problem?"
"We don't have an ontology."
"Why do we need an ontology?"
"Because we don't have one."
"What's an ontology used for?"
"Solving problems."
"So what's the problem?"
"We don't have an ontology."

By the way, the reason I keep following the thread and/or questioning the ideas is that on some level, I think there could be an interesting concept here. I'm just not quite sure what it is yet.

 

[MalFet]

Hybrids and chimera. Gray areas everywhere in the classification process.

At a sub-species level, absolutely...which is why sub-species taxonomies are usually a complete train wreck. The remarkable thing about inter-species hybrids is how rare they actually are and, when it happens, the offspring is generally sterile. To a significant degree, nature imposes its own order; we don't need to.

If there isn't a language of sensory perception, then one will need to be created. Given a finite set of terms that describe something, the classification can begin. Perhaps spicy is one of the terms... I don't know.

Right, that's what I'm saying though: there is a lot of research to suggest that a formal ontology of perception isn't even possible because perceptual phenomena don't have an objective reality in the way that, say, species do. I can't recommend Stan Hieronymus's new book enough, as makes this case far more articulately than I'll ever be able to.

There are a lot of reasons for this, but one of the more straight-forward ones is just physiology. Say there are 1000 olfactory sensors in the human species. The thing is, any given person will have only half of them. You and I could smell the same thing and have physiological responses with literally zero overlap. That would be relatively unlikely, of course, but it's essentially guaranteed that our experiences will overlap only partially.

Throw on top of this the fact that a great deal of smell processing is routed through memory, and the idea that we might produce an formal account of olfactory perception gets interesting (and maybe not in a good way). The goal here is to produce an objective account of an experience, but experience is the thing you need to take away to have objectivity. If you take the experience of smell away from smell, is there anything left?

 

[passedpawn]

Right, that's what I'm saying though: there is a lot of research to suggest that a formal ontology of perception isn't even possible because perceptual phenomena don't have an objective reality in the way that, say, species do. I can't recommend Stan Hieronymus's new book enough, as makes this case far more articulately than I'll ever be able to.

There are a lot of reasons for this, but one of the more straight-forward ones is just physiology. Say there are 1000 olfactory sensors in the human species. The thing is, any given person will have only half of them. You and I could smell the same thing and have physiological responses with literally zero overlap. That would be relatively unlikely, of course, but it's essentially guaranteed that our experiences will overlap only partially.

Throw on top of this the fact that a great deal of smell processing is routed through memory, and the idea that we might produce an formal account of olfactory perception gets interesting (and maybe not in a good way). The goal here is to produce an objective account of an experience, but experience is the thing you need to take away to have objectivity. If you take the experience of smell away from smell, is there anything left?

I'll check out the hops book. I wasn't a big fan of Brew like a Monk - very messy but entertaining.

I disagree that smell can't be objectively judged by different people. While we might make different associations with that smell based on something deep in our brain, we both smell the same thing. Music is the same way: you and I can listen to the same song, but we might be affected very differently. We still would agree that we heard the same song.

If we put 100 people in a room, blindfolded them, then fed them bananas, how many of them do you think wouldn't recognize that flavor?

 

[motobrewer]

My dad is color blind.

 

[MalFet]

I'll check out the hops book. I wasn't a big fan of Brew like a Monk - very messy but entertaining.

I disagree that smell can't be objectively judged by different people. While we might make different associations with that smell based on something deep in our brain, we both smell the same thing. Music is the same way: you and I can listen to the same song, but we might be affected very differently. We still would agree that we heard the same song.

The argument here (and I'm certainly not an expert on the science behind it) is that it's not the same as music because it's not just about association. Unlike with sound, the electrochemical analog to a smelled stimulus is substantially different for different people.

If we put 100 people in a room, blindfolded them, then fed them bananas, how many of them do you think wouldn't recognize that flavor?

That's the wrong experiment, though, because presumably all of these people have had a banana before. Nobody is disputing that people remember past experiences and can compare them to present ones.

An ontology of perception, on the other hand, would allow you to describe -- perfectly and unambiguously -- what a banana smells and tastes like to somebody who has never had one before. That is a much taller order.

 

[neosapien]

"What's the problem?"
"We don't have an ontology."
"Why do we need an ontology?"
"Because we don't have one."
"What's an ontology used for?"
"Solving problems."
"So what's the problem?"
"We don't have an ontology."

By the way, the reason I keep following the thread and/or questioning the ideas is that on some level, I think there could be an interesting concept here. I'm just not quite sure what it is yet.

Yeah, I just don't quite understand what the point of doing it would be. I'm intrigued, but not certain of the usefulness of it.

 

[passedpawn]

That's the wrong experiment, though, because presumably all of these people have had a banana before. Nobody is disputing that people remember past experiences and can compare them to present ones.

An ontology of perception, on the other hand, would allow you to describe -- perfectly and unambiguously -- what a banana smells and tastes like to somebody who has never had one before. That is a much taller order.

Yea, that's a really good point. I'll have to think on it. Imagine describing the color red to a blind person.

 

[passedpawn]

An ontology of perception, on the other hand, would allow you to describe -- perfectly and unambiguously -- what a banana smells and tastes like to somebody who has never had one before. That is a much taller order.

OK, I thought on it.

We don't do that though - we don't expect to describe the flavor or aroma of beer in new terms that were not experienced previously. In fact, we do just the opposite: we use terms that people are familiar with to describe flavors. I.e., we might describe a flavor of a hop as tropical fruit, or mango.

Maybe that doesn't work for whatever-the-hell an ontology is. But for my own classification of (in this example) El Dorado hops, used as a late addition, it would work fine. Even if I never had tasted a beer with El Dorado hops, I'd understand the description when given terms like "mango". (well, maybe, I admit I don't eat much mango).

 

[MalFet]

OK, I thought on it.

We don't do that though - we don't expect to describe the flavor or aroma of beer in new terms that were not experienced previously. In fact, we do just the opposite: we use terms that people are familiar with to describe flavors. I.e., we might describe a flavor of a hop as tropical fruit, or mango.

Maybe that doesn't work for whatever-the-hell an ontology is. But for my own classification of (in this example) El Dorado hops, used as a late addition, it would work fine. Even if I never had tasted a beer with El Dorado hops, I'd understand the description when given terms like "mango". (well, maybe, I admit I don't eat much mango).

Indeed, and that's why things like the BJCP style guide and hops catalogs are useful. Even if people have different physiological responses to the same stimulus, there are broad trends between people (it's not a coincidence that just about everyone likes mangoes) and between substances (cascade is citrusy because it shares some chemical properties with citrus fruit).

But those comparisons are imperfect. A particular hop might remind you of mangoes because there is a high level of some olfactory compound in both, but if I'm not very sensitive to that particular compound I might not find them similar at all. It's like the old quinine experiments they used to do in high school chemistry classes: some people can detect incredibly small concentrations of quinine and find it very bitter, but other people can barely perceive it at all even at high doses.

Does this matter? Precisely to the extent that an ontology is different than a description. Descriptions are tremendously useful, and I certainly don't want to imply that it's impossible to talk meaningfully about how beers are similar or different to each other. Quite the contrary. But, it's inherently imperfect because it attempts to put a subjective experience into objective language. If you were sharing your mango IPA with a friend and he said that he didn't get mango so much as passion fruit, it's not as though he'd be wrong necessarily. You might just be perceiving slightly different things.

But ontologies don't really allow for that. The heart and soul of a formal representation is that it needs to be objective, categorical, and unambiguous. Good approximations don't cut it here, though they're perfectly fine for a good description. Is that a bad thing? I don't think so. I think the fact that a beer cannot be perfectly reduced to an information structure is one of the most wonderful things about the world.

 

[blakelyc]

Heya! Sorry for the radio silence... was on vacation for a while, then dealing with some issues around a startup that I was thinking about getting involved with. anyways.....

But ontologies don't really allow for that. The heart and soul of a formal representation is that it needs to be objective, categorical, and unambiguous. Good approximations don't cut it here, though they're perfectly fine for a good description. Is that a bad thing? I don't think so. I think the fact that a beer cannot be perfectly reduced to an information structure is one of the most wonderful things about the world.

They do, in as much as the ontology structure (and individuals) are provably consistent in a DL sense.... Everything need not fit in a nice box because we have the benefit of open-world reasoning. The only problem is computational tractability, and that's easy enough so long as a good inference profile is chosen.

I did have a great meeting with some folks about this the week before I was traveling. I've spun up a website and registered a persistent URL for namespace convenience. Very soon I'll publish all that information here--I haven't had a chance to publish anything to it yet. I've also stubbed out what I think is a reasonably modular ontology structure that should isolate some of the more contentious areas.

I'll get back to it sometime this week and let you all know.

-b

 

[MalFet]

They do, in as much as the ontology structure (and individuals) are provably consistent in a DL sense.... Everything need not fit in a nice box because we have the benefit of open-world reasoning. The only problem is computational tractability, and that's easy enough so long as a good inference profile is chosen.

That's the rub, though, eh? Despite a few centuries of positive scientific interest in the topic, we're nowhere even near "provably consistent".

If you're interested, here's a reasonably balanced article about the state of the art in psychophysics. The author is relatively optimistic about the future, but there's relatively little doubt that the present is pretty dismal.
http://chemse.oxfordjournals.org/content/25/4/429.full