Joho the Blog » WolframAlpha’s big problem

WolframAlpha’s big problem

After a day of poking at the awesome WolframAlpha and watching some of the reactions around the Web, a major problem has emerged. WA is fantastic if it has what you’re looking for. But if it doesn’t, it looks like it’s failed, as in: “What? It can’t tell me how much energy it would take to move Henry VIII one kilometer, expressed in cheeseburger-calories? What a piece of crap!”

Google doesn’t have this problem. If you get no hits, it’s almost always because you’ve so egregiously mistyped something that no one else on the planet has ever posted anything with that same typo. Or, it’s because you’ve put an odd phrase in quotes, which requires taking the special action of, well, putting things in quotes. Almost always, Google succeeds at what it does (find pages that contain particular text), even when it fails at doing what you want (find a particular answer).

WolframAlpha, on the other hand, is like a roomful of idiot savants. Each knows a scary amount about a topic. And, unlike a such a roomful, WA also knows how to recombine and compute what each of the savants knows. But if the room doesn’t have the savant you’re looking for, you get back nothing but a “Huh?”

The eclecticism of WolframAlpha is its selling point. But the delight that it knows things you would never have guessed at means that you can have trouble guessing what it knows about. The question is whether general users will go back enough times to be trained on the sorts of questions it can answer. If not, WA will remain an awesome tool for specialists but will not become the broad, general-purpose tool it wants to be.

It would, however, be a completely awesome addition to Google…a path I suspect Stephen Wolfram does not want to take.

[Tags: ]

10 Responses to “WolframAlpha’s big problem”

  1. My impression of WA is also that of mixed feelings.
    I found some simple queries working well, but some other totally bad – with no indication why.

    I tried some chemistry on it (having PhD in theoretical chemistry I could :-) ). Simple molecules expressed via simple query like C2H5OH are super, including nice 3D model (try: http://www13.wolframalpha.com/input/?i=C2H5OH)
    but when you take something more complicated but perfectly well known (like La2(CO3)3) it shows something that has almost 0 of knowledge value.

    More, when I typed absurd H9O – the answer was unnoticeably similar to some other perfectly sound molecules !!!

    I tried some concepts form neuroscience – I always got simple: “Wolfram|Alpha isn’t sure what to do with your input.”

    And what bothers me the most: that mysterious source reference: “Wolfram|Alpha curated Data 2009” – with no real explanation
    what it is, how it was build etc….

    So far the only value I got from WA is the fun deciphering of my first name: Mirek.
    Try it – it’s fun and you will understand my signature below :-)
    http://www.wolframalpha.com/input/?i=Mirek

    Well, that’s joke – there are some areas in which it behaves well. Almost all math formulas I entered produced superb, nice plots, formulas etc. So, as Wolfram Mathematica web extension it is perfect. I’m not sure about other fields.

    Perhaphs it is as David writes: “WA is like roomful of idiot savants”

    I hope it will improve. There is something in me, like small voice, telling me – there is some huge value in WA – but I still do not see it….

    Mireka aka 1K-1

  2. [...] – really, game-changingly important. But the material it has to work with is, currently, sparse. As one reviewer commented, it is: “like a roomful of idiot savants. Each knows a scary amount about a topic. And, [...]

  3. Well, I see WA at the development stage of a baby. You need to feed it with information (curate) from healthy food sources and it will grow bigger and become smarter over time.

  4. I just tried to find out the population of Chester, Montana and it insisted I wanted Chester, PA, while admitting Montana is a state.

    I tried to find the incidence of rheumatoid arthritis, asking it several ways, and it did not understand my question. I’ll try again later.

  5. Well, it’s useful having a public copy of Mathematica online.

  6. In the early days of Google, at the bottom of each page there was a question “didn’t find what you looked for? try these sites instead” with links to Yahoo, Altavista, Hotobot, Lycos etc.

    Wolfram should have this feature too.

  7. Even when one gets the right idiot savant, and he computes an answer, where is the reality check on that? The curation (as someone above noted, it’s a black box right now) has to be taken on trust – unlike with an open system in which some answers might disagree with others, prompting further review and exploration.

  8. Sorry guys to bother you again,
    but I came back to Alpha with hope that my favourite mathematical topics have some good coverage in “Wolfram Curated Data” (sorry for cynisism).

    So I typed:

    What it is Fermat conjecture?
    and
    what it is Fermat theorem?

    in both cases WA did not know what to do with my question….

    In desperation I asked “Who was Fermat?”
    and I got an answer, that was OK, but was a far cry
    from what we could expect from “knowledge engine”

    I’m sad, I’m disappointed, I do not know what to say.
    It seems Wolfram premature launch of the site made a harm for all of us who have faith in semantic web, weak AI and knowledge engines….

  9. http://www72.wolframalpha.com/input/?i=Is+Flanders+in+Belgium%3F

    This program hurts the feeling of over 6 million Flemish people…

  10. It is more dissapointing when you bought those apps for IPod and later you find out that those apps need to be connected to the Wolfram server and you realize that the server or servers are usually down!!! What a fraud! I wonder if there’s a way to sue them

Leave a Reply


Web Joho only

Comments (RSS).  RSS icon