WolframAlpha: A Magical Knowledge Engine

[whybother]

Have people been reading about http://blog.wolfram.com/2009/03/05/wolframalpha-is-coming/" [Broken] on it make it sound pretty magical, and don't address any of the standard objections that AI people raise to actually creating a knowledge engine, as it appears to be... but I am still intrigued.

Thoughts?

 

[Astronuc]

Interesting. I'll be watching for it.

If Wolfram hasn't released details, how do people make any credible reports.

 

[whybother]

If Wolfram hasn't released details, how do people make any credible reports.

I honestly don't know. All Wolfram seems to have made is some grandiose claims about how he's managed to implement his New Kind of Science using Mathematica, but without explaining at all how he was successful in doing so.

But it's definitely something exciting to watch out for. A viable knowledge engine would be a pretty impressive achievement in AI.

 

[mgb_phys]

Reasonably positive review but pointing out some of it's problems
http://www.semanticuniverse.com/blogs-i-was-positively-impressed-wolfram-alpha.html [Broken]

 

[ExactlySolved]

For those of us who have been using Mathematica's built-in data sources since the summer of 2007, we have had some time to grow used to Wolfram's futuristic concept that it is possible to integrate hundreds of gigabytes of data into a general computational system, and do it in a clean organized way. I am continually surprised that almost nobody noticed or cared, but Mathematica 6.0 was the biggest step towards producing a general scientific database that was seamlessly accessible inside of a graphical and computational tool.

The new thing about Wolfram|Alpha is that it will allow more people to use this technology that has been in Mathematica since 2007 by processing their natural language commands with sophisticated algorithms!

I have no idea if the natural language thing will work out, and for me it does not matter. I am already in computation/data heaven with Mathematica 7. The benefit I derive from Wolfram|Alpha is the increase in the size of the curated database that is accessible for limitlessly customizable computation from within Mathematica.

 

[waht]

It's going to by turned on today on the 15th of may,

http://www.wolframalpha.com/


This is amazing. If it works, ask reasonably anything you like and it will give you an answer.

 

[RonL]

It's going to by turned on today on the 15th of may,

http://www.wolframalpha.com/


This is amazing. If it works, ask reasonably anything you like and it will give you an answer.

Looks great I think the challange will be how to best ask a question.

The star trek computer is almost here

 

[qntty]

I'm interested in just how much of the system is handcoded in and how much is generated on-the-fly from a database of information. If there is very little handcoded things then it looks like a promising project.

 

[qntty]

Looks like the developers are Douglas fans but show no love for Asimov

 

[RonL]

I'm interested in just how much of the system is handcoded in and how much is generated on-the-fly from a database of information. If there is very little handcoded things then it looks like a promising project.

I would think it will grow as people increase the input of questions and data.

My first request brought up the screen "I can't do that Dave"

 

[Phrak]

Looks like the developers are Douglas fans but show no love for Asimov

Well, it did like "I, Robot"--cause it's a movie

 

[FireSky86]

Wolfram Alpha is my new home page. It used to be Google.

 

[AUMathTutor]

Am I just missing something?

I asked: "How do you solve a second order differential equation?"

It said "I don't know what to do with that input."

I fail to see how this doesn't satisfy the contract that "You enter your question or calculation".

When I tried "What is an eigenvalue?", it gives me a bunch of unrelated calculation options for linear algebra... it doesn't say a word about what an eigenvalue is. Not impressed so far.

With my expectations lowering, I try "what is a cat?". Besides the page taking several minutes to load, for whatever reason, and the graphics not coming up correctly, it does seem to have at least recognized that I'm talking about the animal, and gave me some very basic information about what a cat is (kindom, genus, species, etc.) I could have expected more information, but there you go.

The results from typing "sin(e^x)" were better... much more information, good information (I guess), but not complete information, either. A graph would have been nice, and it seems like it should have been easy enough to figure out. *NOTE: funny, on some other functions it does (try to) produce good plots.

I gave it f(n)=2*f(floor(n/2))+n, f(0)=1, and it didn't tell me anything about the recurrence at all.

So... I'm summarily unimpressed as of yet, although I applaud the idea. It'll need time to iron out the kinks. It is not the promise of the semantic web.

 

[waht]

I don't expect it to know everything, but it's very elegant, and the results are ready in pdf as well.

I asked it

"population of USA / area of USA"

it not only gave me the answer, but showed a graph of the ratio from 1970 to present.


"what's the temperature in New York City from 2000 to 2006"

nice graph

 

[ExactlySolved]

Giving it a specific differential equation, like y'' + x^2 y = 0, produces the general solution as well as time series and phase space plots for typical initial conditions.

If you want to know the definition of 'eigenvalue' type "definition eigenvalue."

The very first tip they give is: Wolfram Alpha answers specific questions rather than explaining general topics.

Try asking about "eigenvalues {{a,b},{b,a}}" and I am pretty happy with the results: just as with my differential equation the results were even cleaner and faster than with Mathematica (but of course with less control).

As far as I know this is the first free web-service that will 'show the steps' that a human student will take for derivatives, or students in school algebra can type "expand (1 + a b + b^2)^2" and see the steps as well.

The funny thing about post-modern cynics is that they will complain about anything if they perceive it as 'hyped.' Nevermind the fact that Wolfram is many years ahead of any potential competitor (evidence: Mathematica 6 has had integrated knowledge database abilities since summer 2007, and if you know of any other software package that compares and you are an expert user of both, i.e. I don't care about learning curves, then please tell me about your superior or even somewhat comparable package).

 

[AUMathTutor]

"I don't expect it to know everything"

I would hardly call being able to provide basic answers to questions like "what is a cat" and "what is an eigenvalue" the same as "knowing everything".

It felt brittle to me... I feel like the hype was too strong. I almost hoped this would be the sort of thing you can go ask a question *that you aren't even sure how to word*, and it would have a conversation with you to decide what you need to know.

No idea:
why did the soviet union collapse?
who won the 30 years war?
why are plants green?
How many eggs are in a gross?
Which side of the street do you drive on in England?
If mammals are animals, and cats are mammals, are cats animals?
Who made Wolfram Alpha?
What is object oriented programming?
What is the mean age of human beings?
What is the apeiron?
Are there security holes in OpenOffice?

Some information (ranging from good to very, very basic):
who was the twelfth president?
what is a semaphore?
why is the sky blue?
What is Wolfram Alpha?
How many weeks are there in a year?
Who was Anaximander?

Until (a) more knowledge is put into his engine or (b) the AI gets better, I'm going to keep using Wikipedia if I need to find out anything online. Sort of a let down.

 

[AUMathTutor]

"Giving it a specific differential equation, like y'' + x^2 y = 0, produces the general solution as well as time series and phase space plots for typical initial conditions."
Which is all well and good, but that may not be what I want to know.

"If you want to know the definition of 'eigenvalue' type "definition eigenvalue.""
The computational knowledge engine should be able to understand simple English... or else I fail to see how it is any different from a regular web search.

"The very first tip they give is: Wolfram Alpha answers specific questions rather than explaining general topics."
It's very easy to have an excellent product, if you get to define what excellent means. That is a false dichotomy between general and specific questions... how general? how specific? Who decides what counts as what? Any general-purpose computational knowledge engine shouldn't have hang-ups like this. It's silly.

"Try asking about "eigenvalues {{a,b},{b,a}}" and I am pretty happy with the results: just as with my differential equation the results were even cleaner and faster than with Mathematica (but of course with less control)."
Then this isn't anything more than a glorified calculator, is that what you're saying? Perhaps I had the wrong expectations. I thought this would answer any reasonable question on any topic.

"The funny thing about post-modern cynics is that they will complain about anything if they perceive it as 'hyped.' Nevermind the fact that Wolfram is many years ahead of any potential competitor (evidence: Mathematica 6 has had integrated knowledge database abilities since summer 2007, and if you know of any other software package that compares and you are an expert user of both, i.e. I don't care about learning curves, then please tell me about your superior or even somewhat comparable package)."
I'm just pointing out the very obvious fact that this system has severe limitations for anything except for relatively well-defined computational problems. I was expecting a system which was much more interactive and intelligent than this. Like I said, maybe there will be more releases, more data, or revised AI.

Maybe my expectations were wrong. Still, "computational knowledge engine" is misleading. It should read "online calculator" if that's what it's purpose is.

 

[ExactlySolved]

"The computational knowledge engine should be able to understand simple English... or else I fail to see how it is any different from a regular web search."

It has a vast amount of built-in curated data and it computes answers in real time, while web search only allows you to access answers that other humans have already computed and published (sure google can do arithmetic, but that is as far as it can go in terms of real-time computation). Its partial ability to parse natural language syntax has nothing to do with my interest in the system. It's all about connecting the worlds largest top-down organized, curated database to the worlds most sophisticated general computational engine and to connect them in such a way that is sufficienly organized and integrated that (some) humans can actually use it.


"Then this isn't anything more than a glorified calculator, is that what you're saying?"

Nope, I was just showing that in addition to other computations it can do inolving factual data, it can also do more real-time symbolic mathematical calculations than any handheld calculator in the world. For free, accessible from any browser!

"I thought this would answer any reasonable question on any topic."

If that's the case then it's hard to imagine not being disappointed. But if you look carefully at the technology, you will see that W|A has more potential to live up to your expectations than any other product that is publically known. If you look at the progress of Mathematica development over the last 20 years, it is literally exponential. IMO the biggest reason for this exponential growth is the unparalleled level of organization: most software products become rapidly bloated as they expand, but Wolfram's genius was to design a system with so much long-term organization that it becomes faster and better as it expands, always building on itself (not rewriting from scratch as e.g. Microsoft has done many times in the last 20 years with its office products).

If you want to be disappointed by what W|A can't do, that's fine, but I look at all the things it can do that nothing else can, and I look at the fact that rather than bursting at the seams it is in fact primed for rapid improvement.

By the way, what was google like on the day it launched?

 

[humanino]

WA is very very disappointing. It understands very little. So much so, that I expect google will have no trouble doing way better. The product was released too early.

 

[AUMathTutor]

I'm glad I'm not the only one who thought as much.

Its functionality is very limited, and I see this manifesting itself in two areas:

(1) The AI isn't very good at recognizing what information is relevant to the question being asked. It's little more than a command-line interface.

(2) The data in the system seems limited in scope and incomplete in content.

The first can be fixed by:

(1) Investing more time in the language recognition algorithms. There should be no fundamental difference between the strings "what is an eigenvalue" and "definition eigenvalue". These mean the same thing.

(2) Organizing data better to allow for better semantic searching. If I ask "why are plants green", it should be able to see "plant" and "green" and the absence of the word "chlorophyl" and, together with the "why", at least mention that it's the chlorophyl that makes plants green. Etc.

The second can be fixed by adding more information to the system. It's misleading to say you speak English if you know 15 nounds and it's misleading to say you have a computational knowledge engine if it knows about the twelfth pesident of the US but not why the Soviet Union collapsed.

 

[humanino]

This is just ridiculous
http://www.wolframalpha.com/input/?i=ampere+law

 

[Phrak]

They might do better in the general knowledge arena by redirecting to wikipedia; by most evidence a proven, expert, semi-organic automata.

 

[AUMathTutor]

Oh my dear god. Hilarious.

 

[AUMathTutor]

"They might do better in the general knowledge arena by redirecting to wikipedia; by most evidence a proven, expert, semi-organic automata."

For the system as it is today, I would say that they could do better in just about every respect by making all searches redirect to Wikipedia.

For the sake of completeness, they could also offer a freeware Mathematica version for people who need to actually compute things that would make it worth it.

 

[Phrak]

"They might do better in the general knowledge arena by redirecting to wikipedia; by most evidence a proven, expert, semi-organic automata."

For the system as it is today, I would say that they could do better in just about every respect by making all searches redirect to Wikipedia.

For the sake of completeness, they could also offer a freeware Mathematica version for people who need to actually compute things that would make it worth it.

In their defence, it's obvious from the interface that they rushed to make it public. It should improve alot, but I won't be betting the farm on utilitarian success. I've worked for an AI company at one time with a lot of expectation in the air. They crashed and burned, leaving a few of the founders too rich for their own good, and stockholders holding the bag.

 

[AUMathTutor]

I hope you're right about it improving a lot. Wolfram is a great, smart guy, and it would be a shame if he fumbled this idea. I doubt he'd let the product stand as it currently is if he values his reputation.

On the other hand, the less Wolfram gets done, the more for the next generation. If anything, this is a testament to how interesting and how useful a problem it is he's trying to solve and how good a good solution would be.

 

[NeoDevin]

You guys who expect it to be Google + Wikipedia + Mathematica + fluent in English are just setting up unreasonable expectations. If I build a new power plant and expect it to match the energy output of the sun, I'll be disappointed too.

It was designed to be first and foremost a computation engine. It's not intended to replace Wikipedia, or Google, or Mathematica. Read http://www.wolframalpha.com/about.html" [Broken] for more details. Note a few things:

Goals
Wolfram|Alpha's long-term goal is to make all systematic knowledge immediately computable and accessible to everyone.

It's not intended to be a dictionary, or a math/physics text, or anything of the sort. It's meant to give you immediate access to data, and be able to offer some relevant graphs and computations with that data. It doesn't have all the worlds data in it yet (and probably never will, considering how fast data is generated), but it has a lot.

Our goal is to accept completely free-form input

I think considering that I can't even understand what some people type onto the internet, it will never be completely free form. They're going to try to get as close to that as possible. They do this much better than Wikipedia, but perhaps not as well as Google (though you don't get the proverbial digital diarrhea of websites that you do from Google). It needs some work to perfect, and I expect this to improve over time, though as long as there are people typing things online which I can't make sense of, I don't expect their algorithm to make sense of it either.

As of now, Wolfram|Alpha contains 10+ trillion of pieces of data, 50,000+ types of algorithms and models, and linguistic capabilities for 1000+ domains

This should make it abundantly clear that it's designed for retrieval and calculation of data, rather than general definitions/education.

Future
Wolfram|Alpha, as it exists today, is just the beginning. We have both short- and long-term plans to dramatically expand all aspects of Wolfram|Alpha, broadening and deepening our data, our computation, our linguistics, our presentation, and more.

This is, like Wikipedia, a never ending effort. There will always be more to add, as well as simply changing things to better reflect the expectations of the current generation of users. It's impossible to release it completely "finished", since it will never be completely finished.

Wikipedia, Mathematica, Wolfram|Alpha, and Google are all very different. Each one excels at some things, and fails at others. You don't search Wikipedia for instructions on how to sew, you don't ask Mathematica to define terms for you, you don't ask Google to solve differential equations, and you don't ask Wolfram|Alpha for introductory calculus lessons. There is some overlap between each of them, but as long as you use each of them as they were intended, you'll be able to get the most out of all of them.

 

[Borek]

From my backyard - it (almost) correctly deals with "what is pH of 0.01M acetic acid" but it fails with "what is pH of 0.01M sulfuric acid". For me that is a sure sign of premature release, as it is able to correctly understand the question, but it uses incorrect approach for calculation.

D from chemistry if you ask me

 

[RonL]

This is just ridiculous
http://www.wolframalpha.com/input/?i=ampere+law

change the word to amps, and you get much of what you want.

As I mentioned near the start, the quality will improve with use, how to word the request is the challange.

 

[Moonbear]

I played with a few things, and as far as I can tell, it's totally useless. There aren't even links out to the sources, so if you need more than the bare bones information on something, you can't search beyond the overly simplistic content on the site...stuff you could find in a very minimal set of encyclopedias.

It also has wrong information. Someone tried this on another site. Look up mountains in Australia and look at what it gives as the highest mountain. Now look up that information anywhere else...

 

[AUMathTutor]

NeoDevin:
So it's a fancy, online calculator?


You know, there was a Mathematica Online Integrator a while ago. That thing was great. If I needed an integration, I went and asked it. Beautiful. They could have added ~100 other pages like that, for a slew of different things, and I would have been immeasurably more impressed.

The "mathematical computation" it is capable of should not be praised overmuch. Most of humanity will never solve a 2nd order differential equation. They will want answers to questions that have nothing to do with "mathematical computation". Should a "computational knowledge engine" be able to deliver these answers? I believe the answer is yes.

People who need the answers to questions like "what's the solution to the y'' + y' = sin(x)" will either (a) know how to do it by hand, (b) know how to write a program in any language to get a numerical answer, or (c) have access to the *real* mathematica, or some other similar package, that will *tell them the answer*. So the people who would use it don't need it, and the people who wouldn't use it don't want it.

There's a difference between being a cynic and pointing out obvious flaws with a system. I know it's still early, and that there's a lot of room for improvement, but if people like the ones who are saying I have high expectations ran the world, nothing would ever get any better.

 

[AUMathTutor]

"I played with a few things, and as far as I can tell, it's totally useless. There aren't even links out to the sources, so if you need more than the bare bones information on something, you can't search beyond the overly simplistic content on the site...stuff you could find in a very minimal set of encyclopedias.

It also has wrong information. Someone tried this on another site. Look up mountains in Australia and look at what it gives as the highest mountain. Now look up that information anywhere else..."

Again, thank you. I haven't noticed any blatantly wrong information, per se, but then again I've had trouble getting any information at all out of the thing.

Do they cite sources?

I just tried "halting problem" to check, and ironically enough, there's no information in the database on that topic yet.

"pigeonhole principle" returned nothing. Are theorems and proofs not included in the system?

Well, "pythagorean theorem" is. There is no proof, no explanation, and no source cited. It gives you a little form to fill out to find the sides of a triangle.

When it does recurrence relations, something that would actually be useful to me, it doesn't show any steps or explain the derivation of the result, so I have no real reason to trust it. It understands

f(n)=2f(n/2)+n, f(0)=0

but not

f(n)=2f(floor(n/2))+n, f(0)=0

Since computer scientists would see no difference whatsoever in these two formulas, and strictly speaking the latter is more technically correct, this is unacceptable.

 

[Phrak]

"Goals
Wolfram|Alpha's long-term goal is to make all systematic knowledge immediately computable and accessible to everyone. ..."

I give up. What is Systematic Knowledge?

 

[ExactlySolved]

D from chemistry if you ask me

Is this a curved grading system, and if so then which online computational engine do you grant an A grade to for chemistry?

"So it's a fancy, online calculator?"

Sure, in the sense that our bodies are just an animated seres of salty water bags, or in the sense that most vocations consist of altering the position of matter on the surface of the Earth --- the point is that questions like these are 'baited' or 'loaded.'

What service, free or otherwise, is closer to achieving the objective of W|A ? Even if it is just a 'fancy, online calculator', is there any other online caculator that is more fancy, or that has more potential to achieve the ambitious goals of W|A ?

People who need the answers to questions like "what's the solution to the y'' + y' = sin(x)" will either (a) know how to do it by hand, (b) know how to write a program in any language to get a numerical answer, or (c) have access to the *real* mathematica, or some other similar package, that will *tell them the answer*. So the people who would use it don't need it, and the people who wouldn't use it don't want it.

Last night I needed to solve a transcendental algebraic equation to find the critical point of an ising-type model, but my Mathematica kernels were busy running a monte caro sim, so I used W|A to compute the answer. Similarly, W|A is availible on the iphone and other (public) computers I may find myself using that don't have Mathematica. Furthermore, W|A automates various things that would take a few steps of writing in Mathematica, so it is quickly becoming my choice for quick calculations.

If I needed an integration, I went and asked it. Beautiful. They could have added ~100 other pages like that, for a slew of different things, and I would have been immeasurably more impressed.

Of coure W|A does all the integrals that th integrator did, and now it will also do school algebra and derivatives with steps shown. As for ~100 pages like that, perhaps you should look at the examples:

http://www64.wolframalpha.com/examples/

It sounds like the degree to which you are impressed has to do with your pre-expectations, but why not judge the service in comparison to what similar services exist right now?

Based on the responses in this thread, the biggest failure of W|A so far is to communicate properly what the intention of the system is, and since it is targeted to a wide audience that has very minimal experience in giving instructions to a computer, to teach them that the way you learn such things is by studying examples. Spend some time with the examples to learn what W|A can do, and how to ask it to ask it to do those things, and then generalize the examples to the cases that you are interested in. Admittedly, most people are unlikely to ever do this, same story as with Mathematica, but this product is still a step forward above anything else that is currently out there --- there was a youthful age at which Mathematica would have been too difficult for me but W|A could teach me lots.

To call the most ambitious computational knowledge engine of all time 'totally useless' is an exaggeration: look at the example page, look at all the things it can do. Most things in life are imperfect in an enormous number of ways: pointing these out is not as interesting as focusing on things that are good, that do represent progress and improvement.

There's a difference between being a cynic and pointing out obvious flaws with a system. I know it's still early, and that there's a lot of room for improvement, but if people like the ones who are saying I have high expectations ran the world, nothing would ever get any better.

Wow, we see each other's viewpoints with irony. I am certainly not saying that W|A has accomplished its goal, and in fact I identify with the people at Wolfram Research who are working non-stop to add improvements. As I see it, Google' marketing people are better at catering to this type of critic, the way that they stick beta tags on everything (don't attack our product, it's just a beta). When your product is improving exponentially, as Mathematica is, life is too short for beta releases --- when M7 was released there were already teams working on M8 and M9: as Mathematica improves it opens new possibilities that take time to develop, but then these improvements herald new improvements and so on (the hallmark of exp growth) and so you have to periodically release something because the product is never done. Of course, I don't mean things like minor bug fixes, but enormous new families of features.

I wonder how many people who are complaining about W|A have done much File I/O with a language like C : those people know that just being able to import data from all the hundreds of different file formats into the appropriate C data structures is a beyond daunting task. Then to organize all of this heterogeneous data into a database that can be accessed with simple one-line commands!

WA is very very disappointing. It understands very little. So much so, that I expect google will have no trouble doing way better. The product was released too early.

Look at the example page I linked, before saying 'it understands very little.' Also, think about the organizational issues involved, and realize that Google is at least 20 years + 1 super-genius behind Wolfram Inc. Super-genius Wolfram spent 10,000 hours doing design reviews for Mathematica 6 alone! Design reviews are not about obscure algorithms, they are about organization and long term vision. Through version 5, the estimated effort that had gone into Mathematica is over 1000 man-years, much of which comes from a large number of ex-soviet Mathematicians in the early 1990s. Almost no one truly appreciates Mathematica, but the thousands of us who do realize that it is totally in a class of its own amongst software. I am reminded of what Haydn said of Mozart:

"If only I could impress Mozart's inimitable works on the soul of every friend of music, and the souls of high personages in particular, as deeply, with the same musical understanding and with the same deep feeling, as I understand and feel them, the nations would vie with each other to possesses such a jewel."

and so to for Mathematica, and in time, so to for Wolfram|Alpha.

 

[ExactlySolved]

I give up. What is Systematic Knowledge?

Systematic as in methodical, procedural. A network of facts which is connected by a well-defined method or procedure for moving between them. This is not a mathematical definition, so please don't pick it apart (philosophy teaches us that this is pointless i.e. you could not even define the word 'game' in such a way that I could not pick it apart), the only point is that you get the idea that the phrase was trying to convey.