To mark the second anniversary of the launch of Wolfram|Alpha, I did an interactive webcast:

Here’s a transcript of my introduction:

[Note: here is what I wrote for Wolfram|Alpha's first anniversary a year ago.]

So, as of today, Wolfram|Alpha has officially been out in the wild for two years.

And I’m happy to say, it’s doing really well.

You know, I’d been thinking about building Wolfram|Alpha for more than 30 years.

And I’ve been working to build the stack of ideas and technology to make it possible for nearly that long.

At the beginning, I was not really sure that Wolfram|Alpha was going to be possible at all.

And I think if I look a year ago from now my main conclusion was that after a year out in the wild, we’d proved that, yes, Wolfram|Alpha was indeed possible.

Well, now that we’re two years out, I think my conclusion is: Wolfram|Alpha is even a lot more important than I thought it was.

This effort to make all our knowledge computable is really something very fundamental, that’s sort of inevitably going to be needed just all over the place.

So what have we been up to this year?

The most central thing is putting more knowledge, more data, more algorithms, more capabilities into Wolfram|Alpha, and tuning and tightening up what’s already there.

Wolfram|Alpha is just an absurdly complex object. From its underlying software engineering, with nearly 20 million lines now of *Mathematica* code, to all the different domains and capabilities that it covers.

But over this year I have to say that we’ve been extremely successful in building out all the systems that we need—automated software systems and human and management systems—to make it possible just to keep growing, keep scaling up Wolfram|Alpha.

We always run a portfolio of development efforts. From things that complete in days to weeks. To major framework changes that can take multiple years.

For example recently we’ve almost finished a major framework change that greatly enhances our linguistic processing capabilities.

We’d have never figured out all this stuff without being able to see a billion actual queries come through the system.

And it’s pretty fancy algorithmic stuff. But the result is that we’ve made it faster to process linguistic input, and we’re going to be able to handle much more elaborate linguistic forms.

And the bottom line is that we’ve still further increased the fraction of queries that we can understand—so it’s now up to about 95%, which I think is pretty extremely good.

Well, in terms of core content in Wolfram|Alpha, over the year we’ve added all sorts of new domains of knowledge.

There’s a list of those on the web. I think there was a blog that just went out containing those.

I think the main thing we’ve discovered is that, yes, after you’ve dealt with 2000 domains, the 2001 is easier. Not least because you get to use everything you know from the 2000 previous domains.

But it’s still sort of irreducibly difficult. I think we’re pretty clever, and we’ve done some pretty amazing automation.

But it’s really really clear that there just isn’t a magic bullet in all of this.

If you actually want to get right answers, reliably, there’s real algorithmic and human effort that you have to put in.

Well, part of the good news is that there are so many people who want to help us to make Wolfram|Alpha really shine in every different area.

World experts in almost anything. Also our wonderful group of volunteers who help with data curation in their areas of expertise.

Well, so, OK, when Wolfram|Alpha was first launched, it was a website.

But what we’ve learned is that really the system we’ve built is much broader and more significant than that.

It’s sort of an inevitable building block. It’s a generic enabler and accelerator of the knowledge economy.

Taking all that knowledge that our civilization has accumulated, and letting it be applied to every possible specific problem, whenever and wherever it’s needed.

OK. So a big theme of this year has been understanding just how to deploy Wolfram|Alpha.

Through what channels and mechanisms.

If you’d asked me two years ago how many of these different channels and mechanisms there would be, I would probably have said 3 or 4.

But now we’ve figured out there are probably at least 15 of them.

So let me give you a few examples.

One of the things that’s great about wolframalpha.com is that it ultimately has an extremely simple interface: just a single input field.

And from that input field you get all this power and computational capability and so on.

But the challenge then is to learn what you can really do with that input field.

Well, I think one of the things that’s gradually been happening is that people have a better and better mental model of what’s possible with Wolfram|Alpha.

And we can see from the query stream that people really use Wolfram|Alpha to do serious things. Things that really make sense.

We get some tourist traffic. But mostly we get people who really robustly want to use Wolfram|Alpha to achieve things.

But, one of the issues is that people can’t figure out all the things they can actually do with Wolfram|Alpha.

So what we realized is that in the modern app’ed world, we could make that easier.

So we’re building lots of specialized apps that are powered by Wolfram|Alpha, but that organize things in a very ergonomic way for each particular type of user or usage.

And actually, in the next 12 months, we expect to release at least 100 such apps.

We’ve got half a dozen out already. One series is “course assistant apps”. The concept is to create “an app for every course”.

Then there are professional apps. And reference apps. And lots more.

Perhaps in a bit I’ll demo some of these.

But anyway, apps are one deployment channel for Wolfram|Alpha.

Another is search engine integration.

You know, it’s interesting. I think over the course of this year, we can see a definite trend that people’s expectations about information on the web are shifting.

People really want answers. They don’t want to be pointed to a bunch of links, for example. They actually want whatever question they have to be answered.

Part of that shift of expectation has I think happened because of mobile.

The network speeds, and the screen sizes, make people want their machines to do more in automating things: in just getting to the result.

Well, that happens to be great for us. Because it brings people much closer in mental model to what Wolfram|Alpha has always been set up to do.

The question is: how does one get the best of both worlds—search plus knowledge?

We’ve learned a lot particularly from our partnership with Microsoft and Bing.

The constraints—both cognitive and engineering—for the search engine environment are different from the knowledge engine environment.

And one of the things we’ve done this year is really to develop mechanisms to handle that.

Very fast ways to triage a search engine query stream and work out which queries Wolfram|Alpha can expect to handle well.

Ways to bring up “summary boxes” incredibly fast.

Even in some cases to answer questions instantly—like we do with simple math right on the wolframalpha.com website.

It’s taken a while to understand this stuff. But we’ve got some great products for search engine integration coming out soon and you’ll see a bunch of work done with partners on that.

Well, there are lots of other channels for Wolfram|Alpha.

Another that’s coming soon is integration with user data. You’ll see that first in the Professional version of Wolfram|Alpha.

Which allows for customization of output, user-defined input shortcuts, extensive data downloading options—and also uploading of user data.

You see, eventually you’ll want Wolfram|Alpha to suck in all sorts of ambient data for you.

Your social graph. Images you’ve got. Spreadsheets. Things from your file system. And so on.

And be able to compute from those things, as well as from public data.

And eventually part of the goal is just automatically—preemptively—to get computations done, without any need to even explicitly request them.

With Wolfram|Alpha just figuring out from your “data environment” what you want to know.

Well, OK. There’ve been lots of new directions like this this year.

One I find very interesting is the integration we’ve done with *Mathematica*.

Even though Wolfram|Alpha is built on top of *Mathematica*, Wolfram|Alpha and *Mathematica* are very different things.

*Mathematica* is a precise programming language, in which one can build arbitrarily complex programs.

Wolfram|Alpha at some level is a sloppy, “drive-by” system, with extremely broad knowledge and capabilities, but with a very simple mechanism for giving inputs.

Well, one of the things we did this year is to combine these two. To make it so that inside *Mathematica* one can just use the free-form linguistics of Wolfram|Alpha. And have it automatically translate that free-form linguistics to precise *Mathematica* code.

So that you don’t have to be a programmer at all any more to be able to create precise *Mathematica* code.

You can just use plain language to do programming.

That’s something that went into *Mathematica* 8 that we released late last year.

And it’s having a quite transformative effect in many areas where *Mathematica* has long been used—for example in education.

By the way, the integration of *Mathematica* with Wolfram|Alpha is going the other way too.

Soon we’ll be launching Wolfram|Alpha Interactive, which uses *Mathematica*-based Computable Document Format (CDF) technology—to let results from Wolfram|Alpha be interactive directly on your computer.

Somewhat related to that, there are all sorts of things going on with Wolfram|Alpha in publishing, and in courseware development.

Our spin-off company Touch Press that’s published the best-selling highly interactive iPad ebooks this past year has used Wolfram|Alpha in all its titles, and has been experimenting with all sorts of interesting ways to pull computational knowledge into diverse kinds of books.

And we’re going to have various tools to make it really easy to insert Wolfram|Alpha into documents. Whether just through dynamic links, or with widgets, or by embedding automatically updating pods of information, or whatever.

You know, it’s interesting. There are all these different ways and places to deliver computational knowledge.

And one of the great things is that when you’re computing results, you just have a lot of freedom about exactly what to generate, what to output, how to structure things.

On wolframalpha.com, you enter a simple query and you get this whole report out.

And we’ve done all sorts of usability studies over the last year or so and the conclusion that keeps on coming back is really very satisfying: that what we’ve done really works well. People are able visually to find the information they want really fast.

And they also, perhaps more importantly, learn useful things that they weren’t already expecting. And they really like that.

Well, with Wolfram|Alpha in *Mathematica*, one of the main interfaces is actually different.

What one wants to do is to insert specific results into a whole sequence of results in a session. Being able to use one result to get the next result and so on. So one wants just a single result by default.

And there’s actually a whole spectrum of “short answers” to get from Wolfram|Alpha.

The kind of thing you’d want in SMS, or for responses from a bot. Or for responses in audio format and so on.

And that’s another direction we’re experimenting with.

Well, so, lots going on with Wolfram|Alpha.

I think at year 2 we’re really in gear. The technology and content side of things is really humming along.

And we’re now really understanding the product and business side of things. How to take what we’re building and deploy and distribute it as widely as possible.

You know, Wolfram|Alpha is ultimately a never-ending project.

And in many ways we’re still just at the very beginning.

But I’m pretty happy with where we’ve reached so far, and I’m looking forward to what we’ll be able to achieve in the years to come—and particularly in this coming year.

Well, OK, that’s enough of a speech. Let’s get down to some discussion here.

So, who has the first question?