@T: ALWAYS THE TWAIN

As readers are aware, I’m building a DIFFERENTIATION / CONTEXT ENGINE. Now whenever anyone originates anything it’s always advisable to check what competitors (or those providing broadly similar services) are doing. Then we can populate interesting competitive landscapes like this one I did on Chinese banks back in 2005:

Naturally, I’ve already put together a nice Porter-esque quadrant of the key players in the “smart ratings and recommendations” space and this appears in my business plan (which is available only under NDA).

Anyway, it may be interesting to gain some insights specifically into Hunch, whose Chief Product Officer happens to be Caterina Fake of flickr fame and which was publicly launched in June 2009 after a private preview period that started in Dec 2008, reached 1.2 million users by Feb 2010  and raised US$12 million in Series B financing in March 2010. At the time of writing there seem to be 12 people on the team. Notably, most of the engineers are MIT graduates and its board members include Gideon Yu (former CFO of Facebook) and Jimmy Wales (founder of Wikipedia).

Ok so this is the team I’ll be competing against when my wee tools finally launch — what  odds, LOL!

Now here’s what happened when I road-tested Hunch today — initially optimistic that MIT engineering graduates would have all the answers to everything, that the UI would be slick and AJAX-y (which it is, btw) and that it would be able to “capture my taste graph” and “personalize the Internet” experience for me.

There’s even a phrase on the site which claims :

(This all sounds like the type of phrase tarot card readers and astrology gurus deploy all the time, btw. Plus Hunch obviously needs an editor / copywriter to spellcheck their copy because it should be “astouNding”.)

The first question I didn’t screen-capture but it was simply about whether I lived in the suburbs / city / countryside and I’ve seen multiple choice questions like these before from Chris Hughes (former co-founder of Facebook, organizing guru at mybarackobama.com and now of Jumo).

Here are 5 of Hunch’s 20 questions along with my highlighted answers to give readers a flavor of the types of questions Hunch uses to train its prediction engine and generate its taste graph:

See the “61% of respondents chose the same answer”? That corresponds to the question about whether I can do 10 pull-ups and the 61% gives a good indication that around 2/3rds of Hunch users answering the 20 questions are probably male — since we women don’t do pull-ups because we don’t want to bulk up our shoulders, upper arms or rear backs. We do crunches, aerobics and yoga to strengthen our abdomen and increase our flexibility.

The result of 33% being spooned is interesting because it seems to confirm that 1/3 of Hunch’s respondents to this 20-q survey are women (which is consistent with the question on pull-ups).

After the 20 questions, Hunch produced this screen:

‘Great,’ I thought, ‘So now Hunch must know what psychometric profile I am — maybe like in an MBTI / Belbin / Baron-Cohen way!”

Alas, it was not to be because this is the page which popped up after I’d clicked on ’10 Predictions Hunch makes about you……’

A-ha……….so Hunch isn’t that intelligent and reverts to binomial Yes / No (+1/-1) in its decision-making architecture.

It’s also offered a nonsensical question about the bathroom signs at the Science Fiction Museum which, in fact, we can’t answer “Yes” or “No” to because what the Hunch system should offer are three buttons to specify gender “Male”, “Androgyne / Android” and “Female”.

Well, for the sake of completing the road-test and humoring the process I went through to either confirm or deny (yes / no) each of the ten statements. For some statements where I contradicted Hunch’s predictions, this then produced another link ‘Answer this question’ which led to another page of a multiple choice question.

(In seriousness, what happened to one-click intelligence? How many people would lose interest in training the engine after the first 10 clicks — never mind another 5+ pages to correct the engine’s prediction mistakes?)

Interestingly, wrt the question of “Do you fold your underwear”, Hunch made the prediction of “No”. Given the choice between the “yes / no” buttons I clicked on “no” to negate their “No” prediction. The mathematicians amongst readers will know that -1 – (-1) = 0 and the English grammar aficionados know that a double negative creates a……neutral statement.

So it’s curious that Hunch interpreted my click as me agreeing with their “No” prediction. In fact, I do fold my underwear because it looks neater in the drawers.

Anyway, the end result of my click contributions to training Hunch’s engine so that it would product my personalized taste graph……….This is what Hunch’s engine, built by super-smart MIT graduates, elicited:

Hmmmn and LOL. Readers will remember what happened when I road-tested one simple question on Elbot:

The engineers work hard and take years to perfect their intelligent systems, run them through extensive beta testing etc. Then I appear with a few simple clicks and they…………..CRASH.

Smart algorithms obviously don’t like me testing them out — LOL.