Monday, January 25, 2010

Who You Think You’re Fooling?

The excellent Smartdogs’ blog has cited an interesting — if ill-advised — new way the government is wasting our money: building robots to train animals without human involvement.


Mind you, I do differ with Janeen in one important way: I’m not so sure you can’t, given the amazing robotics advances on their way in the next 10 to 20 years, build a robot that can train a dog. Given that the Japanese are starting to build robots with facial expressions, and even pheromone signals are beginning to make sense, I’m not so sure that this task is beyond near-future technology.

Now, I agree with Janeen in that the project is hopelessly naïve: at best it will take much, much longer than its creators realize (partly if they waste time with an all-operant box rather than starting by building a robot that looks and smells like a human trainer) [1]. But one important thing that a robo-trainer can have is perfect timing. Never a cue too late, never one too early. Never tired. Never distracted, pissed off, sinus-infected ...

As th’ better half is fond of pointing out, accurate communication with the animal is a fundamental — and good timing of such can produce results in the face of muddy-headed methods, unsavory personality, and much else that’s wrong, wrong, wrong.

I think it could work.

I’m already a bit off-message: my point today has to do with building a machine that looks and smells like a person, at least to a dog. Before you say “could never happen,” recall the cuckoo.

This is the bird that lays its eggs in other birds’ nests, getting free child-rearing in the process (often, the cuckoo chick will kick the lawful denizens of the nest out, so the mama bird is actually losing her babies as she raises the changeling). But it’s all the more amazing in that the cuckoo chick doesn’t look anything like the chicks of the birds it parasitizes: it’s bigger, uglier, just a different, um, bird entirely.

How in the hell can mama bird not tell the difference? Could it be that easy to fool dogs as well? Granted, most dogs are smarter than most birds (though consider the amazing intelligence of the corvids and the psittacines, and don’t be too sure of yourself), but this seems to be well within the ability of a bird to sort out.

Except if they tried, it would be a disaster.

Today’s entry, courtesy Diazaburo Shizuka and Bruce Lyon at UC Santa Cruz, explains why. Not content with basking in the permanent sun we’re told they enjoy out that way, these folks took a look at coots — no, not volunteer dog handler/firefighters pushing 50, but the water birds. Coots are a special case because they sometimes parasitize the nests of their own species.

What the California Dreamers discovered is that mama coot actually does a fair job of kicking changelings out of her nest, even though they look a lot like her own. How? She uses the first-hatched chick as a template, and boots chicks that don’t look enough like it to the curb.

Why does this work? Because, of course, you don’t parasitize an empty nest — you sneak an egg into one that’s already got eggs in it. You therefore start out with a younger egg than the rightful owner’s. And because of that, the first-hatched is likely to be a chick who belongs there, and not an interloper.

Why can’t it work for the birds that cuckoos parasitize? Because cuckoos grow fast and big, the better to muscle out the competition. There’s a good chance that the first egg hatched in a cuckoo-parasitized nest will be a changeling — and if mama uses it as a pattern of whom to keep, fewer, not more, of her babies will survive.

Bird-brained indeed. Maybe I was hasty in rejecting Data’s cat; getting a dog to accept a robot trainer may have less to do with how convincing it is than with picking the right cues, and understanding how dogs think.

[1] Another thing: as I’m fond of telling people, technology is likely to put us dog handlers out of business, but not by creating a robotic searcher. Why? Because by the time they could develop one, it’ll be too easy to find people in other ways, such as reading the location of their GPS-equipped cell phone. (Though this, too, isn’t quite as easy as you’d think.)

Friday, January 1, 2010


So on our way back from Jersey, visiting the Sicilians [1] for the holidays, we decide to take a little detour across the river into New York City to visit my old friend (and co-Best Man at my wedding) Mike Gelfman for an afternoon.

We met at the Metropolitan Museum of Art, which was my clever way of getting spousal brownie points, hanging with an old buddy from the Before Time, and getting to look at cool old armor and weapons all at the same time.

In the event, though, we never got to the Met's amazing medieval arms exhibit because we got sucked into the far earlier pottery and artifacts exhibits from Egypt, Greece, Tuscany [2], and so forth.

I didn't feel cheated.

Tucked away in a case, one vase among thousands, was this customer [3]:

Heather immediately noticed that the descriptive signage, while full of interesting bits about the religious symbolism, left out an important, if puerile, fact: what seemed obvious to us was about to happen in the depicted scene.

We went back and forth over whether the writer was assuming that tidbit would be obvious to museum goers; while we often decry the rather pathetic state of the art in museum signage these days, the fact is the Met's signs mostly date from some time ago, and it seems far mor likely to me that the author, in a Looney Tunes kind of way [4], was presenting information that a proper, if educated and cosmopolitan, city slicker parent can read to a child while “getting” it at a more adult level.

'Course, that's an assumption, and assumptions can be dangerous. Leading us to today's entry, a study of how the smell-reactive structures of the mouse olfactory epithelium map to the sensations reported by human subjects coming from Yuichi Furudono and pals at, of all places, the Japan Tobacco company's [5] Science Research Center and the National Institute of Advanced Industrial Science and Technology in Amagasaki, Japan.

They found, interestingly enough, that the patterns of nerve-cell activation in the mouse olfactory epithelium caused by 12 different odorants matches quite well with similarities and differences in the smells experienced by their human subjects when exposed to those odorants. It's a remarkable finding in its own right, in that it's a kind of Golden Spike that verifies what we're learning about the brain's encoding of olfactory experience by explicitly connecting a series of events in the nose and brain.

But let's take a moment to pick apart exactly what they were doing: they were watching the patterns of nerve-cell activation in the mouse's nose, using that information to figure out how the brain transforms those signals into the smells that we experience — as if we were certain that what happens in the mouse nose reflects perfectly what happens in the human nose, and to some extent brain to brain as well.

Now, the authors are careful to point out that this assumption carries some dangers, and discuss the issue at some length in the paper. But what really pops for me isn't that they're making this assumption, but that the odds are so fat that it's likely to be a sound assumption: from everything we know about the remarkable similarities in the sense of smell among vertebrates, I don't think anybody's losing sleep over the formal possibility that something vastly different can be happening in the mouse brain vs. human to produce similar patterns.

How far we've come.

Anyway, let me wrap up by taking a moment to wish everybody the best New Year Possible — Lord knows, we all could use a better year — and a belated Happy Other Holidays. And please, consider donating some dough to Wikipedia so that this remarkable resource can be there for us all.

[1] Aka my relatives.
[2] The Etruscans, who are a new interest of mine and worth a check-out.
[3] It's a krater, not a vase, apparently. Whatever.
[4] There's a New Year's Day marathon on today.
[5] Check out the link, there's an interesting, if coincidental, connection with the dumpling poisonings in Japan in 2008.