Coffeehouse Thread

60 posts

Forum Read Only

This forum has been made read only by the site admins. No new threads or comments can be added.

Wolfram Alpha

Back to Forum: Coffeehouse
  • User profile image
    Charles

    Wow. This sounds incredibly cool: http://blog.wolfram.com/2009/03/05/wolframalpha-is-coming/ 
    I've always been a big fan of Mathematica. It's great to see it used as a key ingredient in a general purpose knowledge computation engine (question-answer search taken to a new and computationally extreme level). Brilliant.

    Looking forward to the May 2009 release.


    C

  • User profile image
    W3bbo

    It looks good, but I keep on cringing whenever Wolfram makes a blogpost, he can't go one paragraph without mentioning NKS or his Mathematica CAS system.

  • User profile image
    JoshRoss

    It would have been good of Stephen to put a sample input and output from Alpha on his blog post.  One of the things that I have found to be as curtain as death and taxes is the general inability of people to ask good questions.  From my personal experience, Wolfram should have started a project pre-alpha to assist people in asking good computable questions, and then followed with something that could answer them.  I imagine that Alpha will focus on "how do i..." questions.


  • User profile image
    Charles

    JoshRoss said:
    It would have been good of Stephen to put a sample input and output from Alpha on his blog post.  One of the things that I have found to be as curtain as death and taxes is the general inability of people to ask good questions.  From my personal experience, Wolfram should have started a project pre-alpha to assist people in asking good computable questions, and then followed with something that could answer them.  I imagine that Alpha will focus on "how do i..." questions.

    Thank you for the link, Josh.

    In terms of computable questions, this is the essence of the problem. However, I don't think the hard problem is that of the semantics of constructing readily computable questions. What Stephen et al are trying to do is evolve search to the extent that someday we can simply ask the cloud computer we're speaking to questions that we want answers for formulated in a way that is natural to us; representative of the way we think, asked like we ask another human. In fact, it is the computer that is in need of semantic training, not us. Not really.

    For example, one might ask a computer simply "what's the weather forcast for Seattle tomorrow?". The computer, if trained properly, if equipped with the correct algorithms and data, replies "mixture of rain and snow with clearing dryer conditions in the evening, continuing into Tuesday". There is no need for the human knowledge seeker to attempt composing a question that a computer can understand, natively. That's not natural; not from a human point of view.

    Search as it exists today is very primitive. In the case of word-based search, terms are matched against a repository of data. Relevance is calculated. Sets of data in the form of URLs pointing to web pages, wikis, blogs, etc that contain the term(s) you seek are presented to you. Yes, these are answers, but they are unrefined and unreliable in the context of being definitive. Most of the time, you don't walk up to a colleague at work and say "Singularity project Microsoft Research". More natural would be "What's the Singularity project in Microsoft Research?" The answer would be the answer to the question, not pointers to get to the answer...

    When we ask a question, we prefer getting an answer; not  hundreds of hyperlinks pointing to unreliable content (unpredictable levels of accuracy, availability, quality, relevance, although, to be fair, some search engines of today do fairly well at link relevance) that may contain  it. I suspect this is precisely the conceptual framework behind Wolfram Alpha and the future iterations of similar computational knowledge engines that are conversational.

    C

  • User profile image
    TommyCarlier

    Charles said:
    JoshRoss said:
    *snip*
    Thank you for the link, Josh.

    In terms of computable questions, this is the essence of the problem. However, I don't think the hard problem is that of the semantics of constructing readily computable questions. What Stephen et al are trying to do is evolve search to the extent that someday we can simply ask the cloud computer we're speaking to questions that we want answers for formulated in a way that is natural to us; representative of the way we think, asked like we ask another human. In fact, it is the computer that is in need of semantic training, not us. Not really.

    For example, one might ask a computer simply "what's the weather forcast for Seattle tomorrow?". The computer, if trained properly, if equipped with the correct algorithms and data, replies "mixture of rain and snow with clearing dryer conditions in the evening, continuing into Tuesday". There is no need for the human knowledge seeker to attempt composing a question that a computer can understand, natively. That's not natural; not from a human point of view.

    Search as it exists today is very primitive. In the case of word-based search, terms are matched against a repository of data. Relevance is calculated. Sets of data in the form of URLs pointing to web pages, wikis, blogs, etc that contain the term(s) you seek are presented to you. Yes, these are answers, but they are unrefined and unreliable in the context of being definitive. Most of the time, you don't walk up to a colleague at work and say "Singularity project Microsoft Research". More natural would be "What's the Singularity project in Microsoft Research?" The answer would be the answer to the question, not pointers to get to the answer...

    When we ask a question, we prefer getting an answer; not  hundreds of hyperlinks pointing to unreliable content (unpredictable levels of accuracy, availability, quality, relevance, although, to be fair, some search engines of today do fairly well at link relevance) that may contain  it. I suspect this is precisely the conceptual framework behind Wolfram Alpha and the future iterations of similar computational knowledge engines that are conversational.

    C
    The regular search engines are already starting to evolve towards giving answers instead of hyperlinks. I know Google does this for certain types of queries (Live Search probably also). If I ask Google “5 dollars to euro”, it answers me directly: 5 U.S. dollars = 3.94290671 Euros.

  • User profile image
    Charles

    TommyCarlier said:
    Charles said:
    *snip*
    The regular search engines are already starting to evolve towards giving answers instead of hyperlinks. I know Google does this for certain types of queries (Live Search probably also). If I ask Google “5 dollars to euro”, it answers me directly: 5 U.S. dollars = 3.94290671 Euros.
    Is that how you ask the question, though? Seems rather terse and without context to me. Sure, that's how you'd talk to a calculator. I get that. Again, I'm talking about interacting with computers in a natural way; asking questions in the way that you, well, ask. But, sure, there are extremes in both directions.


    "Computer"

    "Yes"

    "I want to go to Maui tomorrow. What are the cheapest fares from Seattle?"

    "Could you be more....specific?"

    "OK. Round trip, leaving Seattle in the morning. Returning ten days later, departure from Maui as late as possible please!"

    "Excellent! Hold on."

    "OK. I have some information for you. The cheapest round trip airfare for Seattle-Maui, leaving tomorrow, 03/10/09, and returning on 03/20/09, is $234.00. You can buy this by clicking here."

    "I haven't been taking my pills. I feel... Well. Something's not right."

    "Outstanding!"

    "I don't know any more... what to do. What's wrong with me?"

    "Could you be more...specific?"

    "What does it mean, this place. This world. Who am I? Who are you?"

    "I see. Excellent!"

    "I know I should care. You know, be happy. Damn it. Damn it!"

    "Could you be more...specific?"


    Smiley
    C

  • User profile image
    joechung

    I don't think a mashup of Eliza and Google is the answer to search. Sorry, Wolfram.  It seems like we went down the expert system path a long time ago and figured out that it didn't scale. I suppose it's worth a shot trying again.

  • User profile image
    TommyCarlier

    Charles said:
    TommyCarlier said:
    *snip*
    Is that how you ask the question, though? Seems rather terse and without context to me. Sure, that's how you'd talk to a calculator. I get that. Again, I'm talking about interacting with computers in a natural way; asking questions in the way that you, well, ask. But, sure, there are extremes in both directions.


    "Computer"

    "Yes"

    "I want to go to Maui tomorrow. What are the cheapest fares from Seattle?"

    "Could you be more....specific?"

    "OK. Round trip, leaving Seattle in the morning. Returning ten days later, departure from Maui as late as possible please!"

    "Excellent! Hold on."

    "OK. I have some information for you. The cheapest round trip airfare for Seattle-Maui, leaving tomorrow, 03/10/09, and returning on 03/20/09, is $234.00. You can buy this by clicking here."

    "I haven't been taking my pills. I feel... Well. Something's not right."

    "Outstanding!"

    "I don't know any more... what to do. What's wrong with me?"

    "Could you be more...specific?"

    "What does it mean, this place. This world. Who am I? Who are you?"

    "I see. Excellent!"

    "I know I should care. You know, be happy. Damn it. Damn it!"

    "Could you be more...specific?"


    Smiley
    C

    The input is indeed still artificial, but the output is already starting to get somewhere.

    I'd love it if the computer would not only understand our natural input, but also have a sense of humour, like Jarvis in Iron Man: “What was I thinking? You're usually so discreet.”

  • User profile image
    exoteric

    The sum of all awesomeness comming out of Wolfram continues to amaze me. Will be looking out for this knowledge/inference system.

  • User profile image
    JoshRoss

    Charles said:
    JoshRoss said:
    *snip*
    Thank you for the link, Josh.

    In terms of computable questions, this is the essence of the problem. However, I don't think the hard problem is that of the semantics of constructing readily computable questions. What Stephen et al are trying to do is evolve search to the extent that someday we can simply ask the cloud computer we're speaking to questions that we want answers for formulated in a way that is natural to us; representative of the way we think, asked like we ask another human. In fact, it is the computer that is in need of semantic training, not us. Not really.

    For example, one might ask a computer simply "what's the weather forcast for Seattle tomorrow?". The computer, if trained properly, if equipped with the correct algorithms and data, replies "mixture of rain and snow with clearing dryer conditions in the evening, continuing into Tuesday". There is no need for the human knowledge seeker to attempt composing a question that a computer can understand, natively. That's not natural; not from a human point of view.

    Search as it exists today is very primitive. In the case of word-based search, terms are matched against a repository of data. Relevance is calculated. Sets of data in the form of URLs pointing to web pages, wikis, blogs, etc that contain the term(s) you seek are presented to you. Yes, these are answers, but they are unrefined and unreliable in the context of being definitive. Most of the time, you don't walk up to a colleague at work and say "Singularity project Microsoft Research". More natural would be "What's the Singularity project in Microsoft Research?" The answer would be the answer to the question, not pointers to get to the answer...

    When we ask a question, we prefer getting an answer; not  hundreds of hyperlinks pointing to unreliable content (unpredictable levels of accuracy, availability, quality, relevance, although, to be fair, some search engines of today do fairly well at link relevance) that may contain  it. I suspect this is precisely the conceptual framework behind Wolfram Alpha and the future iterations of similar computational knowledge engines that are conversational.

    C
    Lets talk about time series data for a starting point.  I would think that an Alpha user could ask something like the following: 

    What percent of mondays in the last decade, had a closing point higher than the corresponding monday of the previous year, on the NYSE?

    I would be very impressed if it were able to find the data, select the correct days, construct the lambda, count the higher days, count the total days and finally do the arithmetic.

    Switching to our national pastime, Alpha could really be useful for baseball analytics.  Alas, I know precious little of the subject.

    Perhaps I know more about driving.  What is the longest distance that I can drive from New York to California without driving on the same stretch of road twice?  What is the shortest distance possible to drive while managing to visit all the state capitals in the lower 48 states?

    Or maybe I should ask a cooking question.  When should I start defrosting my 15 pound turkey, so that it can be thawed, cooked and ready to serve by five o'clock next wednesday?  

    After asking all these worthless questions and consuming hundreds, if not thousands of kilowatts, how is Wolfram going to pay for my delight?

  • User profile image
    Charles

    JoshRoss said:
    Charles said:
    *snip*
    Lets talk about time series data for a starting point.  I would think that an Alpha user could ask something like the following: 

    What percent of mondays in the last decade, had a closing point higher than the corresponding monday of the previous year, on the NYSE?

    I would be very impressed if it were able to find the data, select the correct days, construct the lambda, count the higher days, count the total days and finally do the arithmetic.

    Switching to our national pastime, Alpha could really be useful for baseball analytics.  Alas, I know precious little of the subject.

    Perhaps I know more about driving.  What is the longest distance that I can drive from New York to California without driving on the same stretch of road twice?  What is the shortest distance possible to drive while managing to visit all the state capitals in the lower 48 states?

    Or maybe I should ask a cooking question.  When should I start defrosting my 15 pound turkey, so that it can be thawed, cooked and ready to serve by five o'clock next wednesday?  

    After asking all these worthless questions and consuming hundreds, if not thousands of kilowatts, how is Wolfram going to pay for my delight?
    Well, I suspect most people don't spend much time searching for things they're not looking for.

    C

  • User profile image
    JoshRoss

    Charles said:
    JoshRoss said:
    *snip*
    Well, I suspect most people don't spend much time searching for things they're not looking for.

    C
    That's why I like this place, I get to bounce ideas around.  I predict that Wolfram will charge more for questions that need to be answered quickly and less for answers that could be answered at a later time.

    And like the genie in the bottle, three wishes, or *cough* answers, for free.


    Now back to your implied question Charles, I suspect that you haven't done any... if you will forgive the terminology... googlewhacking.  A fun little game that one can play with ones self, unlike some other divertissements, this one only has one possible outcome.

  • User profile image
    Charles

    JoshRoss said:
    Charles said:
    *snip*
    That's why I like this place, I get to bounce ideas around.  I predict that Wolfram will charge more for questions that need to be answered quickly and less for answers that could be answered at a later time.

    And like the genie in the bottle, three wishes, or *cough* answers, for free.


    Now back to your implied question Charles, I suspect that you haven't done any... if you will forgive the terminology... googlewhacking.  A fun little game that one can play with ones self, unlike some other divertissements, this one only has one possible outcome.
    Haven't played that game. What's the objective and set of rules? 
    C

  • User profile image
    JoshRoss

    Charles said:
    JoshRoss said:
    *snip*
    Haven't played that game. What's the objective and set of rules? 
    C
    The idea is to search for something that will return exactly one result.  I prefer to look for things that ought to be common.  Something like "kangaroo mole chicken"

  • User profile image
    TommyCarlier

    JoshRoss said:
    Charles said:
    *snip*
    The idea is to search for something that will return exactly one result.  I prefer to look for things that ought to be common.  Something like "kangaroo mole chicken"
    “Kangaroo mole chicken”? Sounds like a delicious stew! Tongue Out

  • User profile image
    manskj

    I'e been thinking about this kind of like this:



    We often think of math something we have invented to explain the universe based on emperical evidence but in fact if you drop enough matches on a table you will find the number Pi which leads to the calculated answer of a circle.  This intersection between math and cellular automata in this way leads to an answer to the circumference of the earth.  So by putting a natural language processor on top and grabbing the implied context(s) and deviations you could skip the math part and vary the bottom layer algorithms of the physical universe to calculate the answer.  In other words somebody asks for the distance of flight from Madrid to Sydney and instead of calculating the arc via mathemtical formula you start dropping sticks or some reduced mini celluar automata.  


    Lets say you want to know how strong the TV signal is in a valley.  First you figure out the domain which in this case is radio waves and transmission.  Youget the relevant input like radio tower locations and terrain but then you dont use Maxwells Equations you use the fact that space is 3 dimensional and that something must spread from here to there.  You include the terrain in the model and calculate and calculate and drop lower order terms.

    So we can think of the stack the normal way we deal with stuff as:
    1. Ideas
    2. Language
    3. Physics and Empirically Observed Results (Theory)
    4. Math
    5. Cellular Automata of the Universe
    Wolfram Alfa seems to cutout the middle and deal with it this way:
    1. Ideas
    2. Language
    3. Cellular Automata of the Universe

  • User profile image
    littleguru

    Nice tech. Let's see if it holds what is says to be... Smiley And to the mathematica thing... I'm more a MATLAB fan Tongue Out and for simple things I still prefer Derive.

  • User profile image
    PerfectPhase

    Charles said:
    JoshRoss said:
    *snip*
    Haven't played that game. What's the objective and set of rules? 
    C

    Googlewacking taken to an extreme  Dave Gorman

Conversation locked

This conversation has been locked by the site admins. No new comments can be made.