Wolfram Alpha Launch - Dead in the Water

I've been hearing a lot about Wolfram Alpha and how it will replace Google, since it can parse natural language phrases and figure out the user's intent. Well, tonight I was watching the live broadcast of the preparations to launch Wolfram Alpha for the first time. I actually managed to load the page and run a few queries while the data center was still reporting no connections. (But the page only loaded after they announced they'd activated, about ten to fifteen seconds after, so I was not using the broken, 'back door' version.)

It was interesting to be in at the very start, but rather underwhelming once I tried it out. I admit, I tried a few queries I was sure would not work, just to see how it handled them. It wasn't sure what to do with them. Okay, fine. This is just the launch, and I was asking questions I really didn't expect it to get right. So, for my first real test, I decided to hand it a fairly easy one. I tried phrasing my test question a few ways, and it kept being unsure what to do with the question, so I decided to all but hand the computation engine the answer on a silver platter. After all, I don't know how it works yet, so it is only fair to tweak my question until it ought to be easy for the engine to figure out what I want.

I asked "What major events happened in London, England during the year 1666?" I defined 'London' in the question; I specified that 1666 was a year - and I still got the answer "Wolfram/Alpha isn't sure what to do with your input." I have a screenshot of this - it was hard to imagine, after all the hype, it would miss the Great Fire of London. Now it did at least list "More to Explore", so I clicked on "People & History" - which opened with "Harriet Tubman" filled in as a suggestion. This is when I really began to wonder what all the fuss was about. It has a place, a date, and all it can think of to suggest is a woman who lived on another continent nearly two centuries later?

I tried another question, this one admittedly a bit more difficult, but surely one with many potential answers or at least links. "How can I improve my fiction writing skills?" Wolfram Alpha wasn't sure what to do with that, either. Well, it is a less clearly defined question, so perhaps it isn't fair to judge the engine too harshly for shrugging. So I decided to give it something easy, something it could calculate. "Is India larger than America?" It didn't know what to do with that, either. So much for natural language.

I was wondering if it was simply not working, so I decided to try baby talk. "Who was Abraham Lincoln?" Well, that showed me it was working, sort of. It told me he was a head of state - but not of which country - it gave me dates and places for his birth and death, and that was that. No links, no details at all about the man. By this time, my hopes of a real alternative to Google, something that might let me gather information instead of wading through commercial links I have no interest in were fading fast.

Okay, I still wanted to see what it can do, so I figured I'd ask a question with a simple, short answer. "When did the Emperor of Brazil abdicate?" It didn't know how to handle that, either. I asked it an admittedly vague question next, one I hoped it might at least ask me to clarify. "How long would it take a rocket to reach Pluto?" To be fair, this leaves a lot of variables out - but it is exactly the kind of question a normal user might ask. I thought it might at least ask for details. No, it simply didn't know what to do again.

By now, I was desperate to find something it could do. I typed in "November 15 1889" (which, by the way, is the date the Emperor of Brazil abdicated) and it did figure out I'd entered a date, and gave me a few dull details, like how many weeks, days, and years ago this date was. It also told me nothing important happened on that date. Well, perhaps the designers simply weren't that concerned with South America. So I tried again. "November 15 1889 in Brazil". That ought to tell it what to look for. It almost worked. It listed "Republic Day (Brazil)" - the holiday celebrating the end of the monarchy. But it also said "(no known major notable events)". It has the holiday, but not the event it is based on. I might just put this down to the fact it is just a launch, if it weren't for all the other problems I've had, even when I tried to help the engine along with hints.

Still, I decided to give it one more chance to play to its own strengths. I typed in the date of a very major event, the start of the Great Fire of London, "September 2 1666". It offered me an 'input interpretation' that suggested this was a date in the Gregorian calendar - without offering any option for the Julian calendar, still in use in the English speaking world at this date. And it didn't mention anything about any events that might have happened on this day in either calendar.

I don't know. The idea might be sound. After another ten years of fiddling with natural language and letting it look at data, it might actually be able to answer a few questions that aren't mathematical. I deliberately avoided those; after all, Mathematica already exists, and works. If Wolfram Alpha is just another front end for Mathematica, it isn't very impressive at all. Yet of all the "Wow!" comments I read in the chat reports of other users' tests, they all seemed to be asking it to calculate something, and using pretty standard terms to do it. I guess the guys at Google can get some sleep tonight. Meanwhile, all of the researchers who were hoping for more can keep on dreaming, because Wolfram Alpha just is not going to make our lives any easier. Even with the spam, I can pull more information out of Google faster - and as for the suggestion just before the launch that people might forget how to think because Wolfram Alpha will do all the work for them - HA! I was thinking harder than I usually do, trying to lead it to the answers and figure out what kept tripping it up, and it still couldn't manage much.

Yes, these are one user's impressions, typed right after trying it out, and they are harsh. But I do understand every launch like this is someone's dream, and a lot of hard work has gone into it. I don't enjoy crushing dreams, not that my one post will have the power to do that. But if the Wolfram Alpha people ever read this, it will be discouraging for them, and I don't do that lightly. I honestly don't think the hype raised my expectations too high; I just think there isn't very much there to work with. In one way, this is a vindication for something I've long believed, that math and language are so inherently different that a computer will never be any good at language. Yet it is also a huge disappointment, because I wish I could see at least the potential for a useful new tool that might be helpful in research.

Footnote: I'm aware this isn't very well written. It is late, I'm tired, and I just wanted to get my first impressions down. I'd thought I was witnessing something historic, and the impetus of that carried me forward. Now, I'm just deflated.