Wednesday, May 20, 2009

Toying around with Wolfram Alpha

The makers of Mathematica recently released their much anticipated online tool - Wolfram Alpha.
For the uninitiated, Wolfram Alpha('WA' here onwards) is, as they call it a 'computational knowledge engine'. The primary interface - a single text box, looks similar to Google but WA is very different from Google. Google is like a peon at office who can search for the file you asked in the entire office and bring it to you diligently; WA is like an intelligent consultant who will try to understand what information you seek and will search for it in his own knowledgebase (rather than your office), find relevant data, correlate it if possible and present you with a succinct summary and a graphic visualization if feasible. WA does not depend on the Internet for information.

Here are some things I have tried out on WA -
Am not posting the screenshots, they would take too much of space and time.

First, I began with some silly 'Wh' questions - who are you ? where are you ? why are you ?
- WA failed on the why are you ? I was hoping to get something like 'I am because my creator created me.'
Then slightly different one - are you conscious ? and then, can you pass the turing test ?
- WA returned a satisfactory answer for the first, but failed the second

An inquiry about Pune gave some useful information but better is available on the wiki.Any two things can be compared by putting them next to one another like this.

Thus, contrary to the pithy phrase, one can very well 'compare
apples and oranges' ;)

I asked WA about meclizine and then asked it to compare meclizine and meclozine
- WA correctly recognized it as a drug ingredient and gave very useful info about it. This is where having a reliable, structured knowlegebase of one's own makes a huge difference. WA need not spend time and power trying to classify the word(query) from scratch. It already has the classification ready, it just selects a category and uses it to fetch specific info. The Internet hosts a prodigious amount of information, but to extract, classify and interpret it is IMHO a herculean task. WA does not try to do that.
WA failed in the comparison. Meclizine and Meclozine are the same compound, WA interpreted Meclozine as Meclizine but could not infer that there is no point in giving a side-by-side comparison of two same things. Poor logic IMHO, but easily fixable.

Tried out a conversion I often need to do - miles per gallon into km per litre. Google Calculator cannot handle it, but I was sure WA (with Mathematica backing it) could do it easily.
- The answer came out well, WA also showed some relevant relative comparisons for the number I'd entered. US dollar to Indian rupee conversion worked fine too, it gave me a lot of additional info I'd not asked for :)

Tried to get a truth table printed for a simple expression
- Worked nicely. WA showed the minimum forms and even the logic circuit!
Then I tried to check equivalence of two expressions. I tried to use the same syntax used in Mathematica and it worked. Although WA showed the truth table with all 'T's, it did not give a verdict whether the expressions were equal or not, so my question remained unanswered. This also should be easily fixable.

Attempted to get some information about P & NP algorithms, sorting, Hamiltonian cycle, Huffman code but WA seemed to be oblivious about them :( Mathematica is aware of some of them though.

This comparison I tried out proved that WA has excellent resources when it comes to chemistry and materials.

WA also has a superb knowledgebase of genomic sequences, I'd expected to see hardly anything when I tried this. It was a cool surprise! WA not only knew the gene by name, it also knew a lot about it even including nucleotide polymorphism frequencies!! The data although not exhaustive, comes handy at times.

Next, I tried looking for a note - A natural
- WA gave useful info including freq, notation and key location on a keyboard. Then I tried to search for C#. WA was not aware about C# as a language, it interpreted it only as a musical note.

Tried to check out WA's fluid mechanics by searching for Reynolds number. WA brought up an interface which allowed one to calculate the RN by giving inputs. It had tabulated the inputs and laid out the equation well. With raised expectations then, searched for worm gear. Sadly, nothing relevant turned out. Instead, WA completely condoned the important word 'gear' and displayed information about worms(animals). Disregarding user input is the last thing to do!

Overall, WA is an interesting tool, but it is hardly as useful as Google. With the limited data that it currently has, many of our questions go unanswered. It is good for a quick overview and comparison of objectively quantifiable topics and for doing maths and statistical calulations that google calculator cannot yet handle. Google won't be left behind, the folks are already working on Google squared. With the vaaast amount of data already indexed by Google, IMO, Google will definitely have an edge over WA when it comes to coverage. WA on the other hand, will be better in data correlation and compute power requirements owing to the structured nature of its data sources.

Here is Wolfram Alpha for you to try out -

Related Content

1 comment:

Anonymous said...

I want to quote your post in my blog. It can?
And you et an account on Twitter?