An inside look at Stephen Wolfram's computational paradigm | Hacker News Reader

An inside look at Stephen Wolfram's computational paradigm

venturebeat.com

75 pointscodelion41 comments

The problem with covering claims of changing paradigms is: who is qualified to evaluate such ambitious claims? Realistically, not many folks.

I think the best thing for a reporter to do is to connect the appropriate experts together to build a balanced story. Yes, spend time with Wolfram, take good notes etc.--but then find the right experts to respond to and evaluate what he said.

To me, this story reads a bit too credulous to be really useful in understanding what Wolfram is claiming. I read Wolfram's blog post already, and this story seems about the same in tone and content.

What I want to see is the equivalant of this [1] article: a cranky review by an expert in the field...with citations.

[1] http://vserver1.cscs.lsa.umich.edu/~crshalizi/reviews/wolfra...

WolframAlpha (the current site/service) is pretty impressive as it stands now.

It's good enough that I did the first few Project Euler[1] puzzles on it, and given that it isn't really a programming tool I think that's a decent demonstration.

I'd be inclined to believe he has something new, that really can do things that are difficult with other tools. Whether or not it is useful is another question.

[1] http://projecteuler.net/

Isn't that what hacker news comments are for?

Wolfram is a master at this game. You can't deny the brilliant stuff he's done; throw in new terminology for unoriginal concepts, and it becomes very hard to call a bluff (or a delusion). Edit: or confirm.

I am really excited about this. Combining their API with the NLP stuff we are building cool things would be even simpler.

Our engine turns things like "How many people are living in China right now" in to { "action": "query", "project": [ "population" ], "model": "geopolitics", "payload": { "country_code": "CN" } }

We can do more complex things, but I figure pasting huge code examples in to HN is frowned upon.

Combining our ability to create structured queries with Wolfram's ability to do amazing visualizations could result in some amazing things.

We are looking at basically having the Star Trek computer in terms of being able to say. "How many times more men live in China than in the US" and have our data create a query that Wolfram would visualize by taking the number of men in china and dividing by the number of men in the United States.

Or taking it a step further... "let me know" our version of "If this then that" could be really impressive with this kind of data. "Let me know if there is rain anywhere on my route from San Diego to San Francisco Tomorrow"

We have already started to do some of this at http://plexivoice.com but ties to Wolfram will make it even more surreal.

If it's not too much trouble, could you put some of these huge code examples in a blog post or something?

I get the sense that a lot of people (myself included) are still a little confused as to what "Wolfram Language" is exactly, and maybe some code from someone who sees value in it would help clarify things.

I will get a blog post up soon, and will update this thread.

I think part of the problem is Wolfram named it poorly.

Think "Ultimate Mashup Tool with a relatively easy interface for building the mashups"

I'm excited because with my NLP stuff it can be "Mashups built with your voice" or "Mashups your Grandma can build"

Count me in this bucket...

i bet it's dope, but I bet it's not "it".

Sidebar:

Argh.

I've played with alpha several times and each time gave up in frustration.

Just now I wanted to know "What's the history of the closing price of the Dow Jones Industrial Average in ounces of gold over the last 50 years?"

Spent 15 minutes fooling around. Finally got the pricing data in working separate queries, but couldn't combine it. So I asked to download the files -- only to be bugged to sign up.

Must be a totally awesome system. I can't make heads or tails of it.

Meh.

Was your intention to compare gold and DOW as investments for long term?

In that case you should ask question: "What is the price of Dow Jones Industrial Average dividends reinvested in ounces of gold?"

In long term, the majority of value in stock markets comes from dividends. That does not show in stock price graphs.

No. I wanted to see how the stock market change would look if, instead of being priced in dollars, it was priced in some form of commodity, like gold.

So if one day the index closed at 1000, and gold was 200/ounce, it would have closed at 5 ounces.

Thanks for the help, though.

If not even nabla9 can understand what you were really asking, it shouldn't come as a surprise that an NLP program couldn't ;)

"Use NLP to convert questions into code, which can not only retrieve data, but also do parametrized computation on it."

This sounds to me to be the core here, simplified as far as possible.

Quite an undertaking, but if it gets rolling properly, it can actually end up challenging Google. Imagine being able to ask the following question and get an answer with summary and graph and maybe options to drill down, all computed in response to the question, not pregenerated:

"How many people were living in China from 1950 to 1980?"

(Or don't imagine and just click: http://www.wolframalpha.com/input/?i=How+many+people+were+li... )

It's interesting computation, but how often is it useful to quickly estimate how many people were living in China between 1950 and 1980?

Not to pick on that one example, but I do think this is part of the reason that Wolfram Alpha has not taken off at consumer scale. It seems to me that the things it can do that Google cannot, while technologically impressive, are not particularly valuable to most people.

After completing their schooling, most people don't need to solve equations every day. If they do, they already have ways of doing it (Mathematica, for example).

And most people don't need to look up census data, or historical GDP tables, or the location of the ISS. And if they do, they already have ways to do that.

Basically, Wolfram's approach seems like a great way to organize and make useful knowledge that is fully known. But the most value is pushing the edge of the unknown, and that means continuously taking in, indexing, and presenting new information. Facebook does this, Twitter does this, Stack Overflow does this, etc....and Google indexes all of that.

> And if they do, they already have ways to do that.

Disseparate ways that take awhile, with results from multiple sources that need to be manually combined to figure out what information is worthwhile. Some of the random queries I've tossed into WA:

"If the average mailbox size is 57MB, and I've only processed 65% of one in the past half hour, how long is it going to take to proce ss all 80GB?" Q: http://www.wolframalpha.com/input/?i=30+min+%2F+%2857MB+*+0.... A: ~1.5 months, better come up with an alternative

"How big is a gigasecond?" Q: http://www.wolframalpha.com/input/?i=1+gigasecond A: ~1/3rd of a century

"Just how fast are my 5m/s Kerbal Space Program landings?" Q: http://www.wolframalpha.com/input/?i=5m%2Fs+to+mph A: "I've been rear ended at that speed"

"How much precision will a double give me at the radius of pluto?" Q: http://www.wolframalpha.com/input/?i=orbit+radius+of+pluto+%... A: "millimeter, perfect"

Or work related; "I'm having trouble visualizing these 3d points, please plot them so I can get a better mental picture for why this right angle assertion is triggering" "Divide by b squared? That can't be right..." when canceling out terms in math code for efficiency, rearriving at one of the equations of http://en.wikipedia.org/wiki/Vector_projection#Vector_projec... . "I can't intuit the behavior of this math without graphing it out, lets see... bounds... period... yeah that makes sense!"

I also use it for simpler calculations: "How do the prices of gold and platinum stack up against each other?" "How long until sunset? Thanksgiving?" "When is 18 KST in my timezone?" "How much is that in dollars?" "How big is an acre?" "What does the compound interest on that look like?"

Some of these could be found through google... probably followed by trolling the results for a specific calculator. Others I could calculate myself. But most would be inconvenient enough without WA I wouldn't ask them. Some of these are idle curiosity, but others have saved me significant time doing my job. It's not a daily sort of thing, but it's a consistent way to answer these oddball questions for myself.

Google was a tool I had to learn to turn to with my problems. It didn't used to be my primary interface to all documentation, for example... slowly but surely I turned to it more and more frequently. And so it is so far with me and Wolfram Alpha.

Still, there are significant holes in the parsing engine - once you learn its quirks you can get around it, but it's a long way from natural. It can do, for example:

"estimated number of atoms in a liter of water" - http://www.wolframalpha.com/input/?i=estimated+number+of+ato...

and

"volume of a cup of coffee" http://www.wolframalpha.com/input/?i=volume+of+a+cup+of+coff...

but not:

"estimated number of atoms in a cup of coffee" http://www.wolframalpha.com/input/?i=estimated+number+of+ato...

What you're looking for is: "estimated number of atoms in 1 cup of water".

Similarly "distance to moon in terms of bill gates' height" does NOT work, while "distance to moon / bill gates' height" does work.

Quite true.

Still, the same holds for Google as well - sometimes you need to avoid relevant keywords because of the junk they'll bring in, other times you need to add seemingly tangential keywords, and yet other times you need to resort to google's syntax -- quotation marks, +/- and site: prefixes. (alack, poor +, we hardly knew ye)

Stumbling upon the right combination can take quite a bit of work until you've built up experience with both the search engine and the subject material's keywords.

It might be a good way to solve those stupid interview questions like "how many manhole covers are in Manhattan"

Slight tangent: If you haven't seen the ISS go overhead, you really should do so - it's pretty cool. To find on when to see it: http://spotthestation.nasa.gov/sightings

Interesting that that graph shows none of the effects of either the Great Leap Forward or the Cultural Revolution. In other words, it's quite inaccurate. Also note that neither of the two resultant graphs show the period requested (1950-1980). One shows 1970-2013, the other 1600-2013.

Maybe I'm spoiled by XKCD what-if, but that result doesn't tell me which of those people in 1980 are the same as in 1950, or new people. Or figure out how many people died young in between. If someone goes there for six months, is that living?

I think you're making the same "Facebook will kill Google because it's social" argument all over again, just from the mathematical, rather than social, perspective.

What I'm trying to say is Google incorporates many types of signals into its algorithm, it's not just one thing such as "social", which is why Facebook could never in fact "beat Google" at their own game (search engines), no matter how hyped up that was a few years ago. I think the same is true about Wolfram Alpha.

I don't see how this ends up challenging Google. It might win the nerds away but I think you'll find the top 85% of queries by volume are already handled well enough - and more importantly, the top 95% of queries by monetization/profit are already handed correctly.

Your example search is neat but the CPC on it will not be great. Wolfram could totally turn into a Thing, and a profitable one, just not on the same order of magnitude.

If you want CPC:

"What is the highest-rated coffee machine with at least 100 reviewers below 150$?"

https://www.google.com/search?hl=en&tbm=shop&q=coffeemachine...

Eh... I upvoted you for weeding through the bullshit, but it is still bullshit. People shouldn't aggrandize like this. They should let the work speak for itself.

You don't get it. For any casual browsing a graph means nothing. But for people putting a powerpoint together or doing hw or doing research this graph is useful. Imagine you get a comparison chart from your query (Google already does this for very very very limited comparison queries). If you get the raw data out of Wolframalpha that's even better. Instead of having a data scientist to write code to get the graph or imputing this into excel or words to get a graph, let WA do it for you. Hey here is another idea for innovators: instead of screenshot a graph but don't we have a graph importer in powerpoint?

We are just looking at 2D plot, there is nothing fancy about this graph but there are more fancy stuff you can build. A graph is a generic term in mathematics and I hope you know that. Why do we go through social graph and knowledge graph? Why not keep them behind a firewall? Why bother building a JS interpreter in Go? If you can build a machine that can compute anything (universal turning machine, well, I know there are NP problems but we can do our best to compute a subset of the computable - approximation), you can do a lot of crazy things. I think WA is nailing down how the future of computing in relation to life is going to be. Ask a question and gives you back the data to you in both representational and raw form.

Here are examples:

http://www.wolframalpha.com/input/?i=Who+is+Mozilla%3F

http://www.wolframalpha.com/input/?i=Who+is+https%3A%2F%2Fne...

If somehow we can contribute to WA's modules (we write our custom programs and WA can adopt them by means of marketplace - select to use this module, for example), then we can utilize WA powerful machinery to do our query and benefit the mass. This is where Open Source can make a big difference! Sure APIs are nice but you know, not everyone has time to manage another new website. If we can get paid to write modules and get commission share based on the number of users ... So on top of API, direct module marketplace IMHO is the game changing. This is why I like Google App Script and App Engine.

http://www.wolframalpha.com/input/?i=Who+is+Facebook%3F

No you don't get it. It isn't sentient code. This isn't smart code. It's just plain old code that does plain old pattern matching like we've been doing for decades.

It isn't anything that hasn't been done before. It is just more of the same from WA - Hype and bullshit.

How exactly the work that WA is doing going to nail down the future of computing? This is crazy. I'm sick of the hype around WA. It's essentially just a search engine + mathematica at this point.

> It's essentially just a search engine + mathematica at this point.

Yeah and that integration is more than just 1+1. You'd actually think Google's search engine is entirely sentient code. I'd say half of the stuff we build is nothing more than some regex in the bottom and some string matching. You actually think WA is really just a system that understands some NLP and then sends the NLP strings to some Mathematica APIs?

> It isn't anything that hasn't been done before. It is just more of the same from WA - Hype and bullshit.

And half of the shit we see on HN has been done before. Quit your job because there is no innovation there. 90% of the security tool people build are nothing more than some stupid string comparison with some strings in the database. Very few is doing real sentient code.

I see nothing innovating about the way we browse website, the way we interact with technology or the way we build software today. We are still using browser. We are not automating the way we would like to be. Where are all the talking Siri integrated into our home?

WA is very different from Google search.

And if WA is good at dealing mathematical models, just wait and see what one could do with those APIs. You can't do this with Google easily.

From "define: sentient" on google.com.

    sentient
    ˈsɛnʃ(ə)nt/Submit
    adjective
    1. able to perceive or feel things.

From the query "dear google are you sentient?

    Big Brother Google becomes sentient at last | RiskHeads
    www.riskheads.org/big-brother-google-becomes-sentient-at-last/‎
    by Adam Bishop - in 40 Google+ circles
    Oct 25, 2011 - If you try the search above you'll notice you can view the
    sources from which Google gleamed the data and also let them know if 
    they got it right.

Yeah, sure this is cool. We probably can see this from Yelp in a limited way, from twitter or from facbeook. Everyone who work on recommendation algorithm probably have an idea how to implement one. I don't see that being impressive.

Its not very useful, eg it lists 40 sources for the data without really saying how they are used. Its kind of ok for an approximate answer.

Judging by past wolfram hypes, I don't have any confidence in this at all to be even a moderate success. I wish it would be true though, but we know it won't happen. PR/BS ratio is high.

Bravo, but I have one question:

What exactly is the "utterly new, insanely ambitious computational paradigm"?

The ability to create insane queries that leverage multiple data sets seamlessly and provide a visualization. I gave some examples in my comment.

Well, this is technically very difficult. I don't see how it is a new "paradigm" though. It's just a next step in the same direction technology has been heading, no?

It's _A New Kind of Search-Engine_

It's hard to figure out what the article is about underneath all the hype coating.

Hrm.

Wouldn't this be bad reproducibility wise ? I mean, the language spec is essentially changing on a rolling basis.

Reply on news.ycombinator.com