Recent events

The first of these events was a recent panel entitled “Artificial Intelligence Theory and Practice: Hard Challenges and Opportunities Ahead” which was chaired by Eric Horvitz, President of the Association for the Advancement of Artificial Intelligence, at the Microsoft Faculty Summit. There were seven of us on the panel representing a reasonable range of AI fields, and I think it was a pretty interesting discussion (at the time of this writing the video of the panel has not yet been posted to the Web, but it should be by the time you’re reading this, so search for it, I think you’ll enjoy watching it). Scarily, at some point I realized that I was the person on the panel who’d been doing AI the longest (although edging out a couple of others whose hair was as white as mine because I started in my freshman year of college). I also realized that I’ve been working on more-or-less the same problems for my whole career, but strangely at different times I’ve been seen as anything from a mainstream AI researcher to, more recently, doing research outside the mainstream of the field. That’s one theme I’ll return to.

A second theme arises from the use of the term “AI-Complete Problem” on the panel – with several people giving examples of what they thought some were. The odd part is that, that age thing again, I remember when the term was first introduced, and people really meant it to mean a problem on which, to demonstrate significant capabilities, you would need to solve the whole range of problems in AI – ranging from vision and robotics to language and planning. Nowadays it seems to me that the term was being used for “very hard problem,” which is something very different. I’ll return to this as well.

Another event was a meeting of the investigators from one of the large DARPA-funded projects that I’m involved in. For those outside the US, or lucky enough to be funded from other sources, you may not know that in recent years DARPA has pushed for AI researchers to form teams with industrial partners, and try to solve hard “go/no-go” problems. In the case of this particular project, in the second year we had to run our system on a set of problems, run human subjects on the same set of problems, and show our system outperformed the humans. The problems were in a relatively complicated domain, and DARPA set some pretty difficult ground rules – the main one being we couldn’t build in a lot of domain knowledge to solve the problem. Rather, we had use a set of learning technologies (mostly related to explanation-based learning) to accomplish the task. Amazingly, we passed.

Tying it together

So what is it that brings these three themes together? My answer is that all three of them relate to a narrowing of the goals of the AI field over the past decade or so.

To start with, in my career a large part of my work has always been focused on scaling of knowledge-based inferencing in one way or another. The reason is not that I think this is critically important for applied AI, although I do, but rather because one of the things that is clear to me when I interact with my computer is that we each have very different kinds of memory. My computer never really forgets anything I tell it (well, there was that hard disk crash a couple years back, but that isn’t what I mean), while my memory is pretty porous. However, with the exception of some kinds of purely statistical inference, my computer also doesn’t seem able to put things together the way I would. I may forget the details of some particular restaurant I ate in, or what the name of who my middle school math teacher was, but I sure can integrate a lot of other information about restaurants and middle school math in ways that my computer still can’t. This is similar to the theme of my earlier letter “Computers play chess, humans play go” [[insert ref]], but in this case I’m emphasizing that we humans do something really amazing with our memories, that computer models still don’t come near. We also seem to be the entities with the most sophisticated symbolic reasoning capabilities that we know, a reason why I’m as yet unconvinced that all the success with probabilistic models is getting us nearer to understanding human intelligence.

And speaking of human intelligence, I think that takes us to the second theme. The original concept of “AI Complete” included the idea that solving the problems would teach us something about intelligence writ large. That is, while there were always engineering goals in AI, and one of the reasons this magazine started (back when it was called IEEE Expert) was to reflect that, there tended to be a general feeling in the field that the goal of AI included an understanding of intelligence. Not necessarily human intelligence in the sense of cognitive modeling, but just as we know a lot more about how birds fly from having learned the aerodynamics of making planes fly faster, looking at the difference between computers and humans solving problems situated in the real world and needing a lot of knowledge, seemed like a way to learn more about humans and thought. I also remember that I used to hear at early AI conferences, but rarely if ever hear any more, that looking at problems that humans were much better at than computers was a good way to get inspiration as to what AI problems to attack – the challenge made them compelling, and steps towards solution could, again, help us understand more about what intelligence was.

And what about the success of the DARPA project? How could that tie into the narrowing of AI? I guess my reaction came from the fact that of all the researchers in the room, I was the only one who was really surprised that we were able to outperform the humans on this test. The point is not that the others were so blasé, but rather than in today’s AI milieu we are used to seeing AI programs outperform humans whether it be at searching for information, predicting traffic patterns, beating the world’s chess champion, or solving narrowly defined problems. My surprise in this case came from the fact that the problem we were solving was actually pretty hard, but not artificially so due to data overload, a bad interface or the similar

In the problem we attacked, there are many plausible solutions, but only a few good ones, and the restriction was that there was only a small amount of training data (in fact, one expert trace) for the computer to learn preferences from. This was the kind of problem that we once said we’d be able to get AI to do “someday,” and I was very glad to see, that at least for this particular case (and with an investment of a very large amount of time and money), that day had come. However, I was also a bit chagrined to realize how long it’s been since the last time I had that sort of thrill. While I’ve joined others in the field of AI in celebrating our successes, many of which I have pointed out in these pages over the past few years, this experience made me realize how rare it is that I feel the thrill I felt in the early days of my AI career when, I have to admit, it was a pretty amazing accomplishment when we got the computer to do anything that seemed “smart.”

The consequences

So why do these things bother me? After all, this seeming change in our direction has enabled AI to accomplish significant engineering advances and to become a stronger and better-understood technology. In fact, on the panel I mentioned earlier, one of my younger colleagues expressed how proud she was that when she first took at AI it seemed to be pretty ad hoc, and now when she teaches it the course is full of much stronger theoretical material. My rejoinder, grumpy old man that I’ve become, was that when I first took AI, I was hooked for life on the first day, when Roger Schank who was teaching the course, said something to the effect that almost nothing he would teach us was proven to be correct, and that any one of us was just as likely to come up with a key insight as the people whose work we’d be studying. I found this exhilarating; my colleague seems to have found it a weakness.

But here’s the thing! Despite all the major AI successes of the past decade, and the great strides we have made, what Roger said is still true! When it comes to really understanding the amazing symbolic processor that is the human mind, we still know very little. While I don’t mind that we have more techniques to teach our students, I think it is important that we don’t become enthralled by what Marvin Minsky has referred to as “physics envy.” We should admit, gloriously and deliberately, that when it comes to understanding intelligence our field is still in its very early days. The challenge remains, and it is one of the greatest intellectual challenges of our, nay of all, times – to understand thought, conciousness and intelligence. The best and brightest students go where the most exciting problems are, and we’ve got one of the all-time winners! Let’s not forget that fact.

So as my days as Editor-in-Chief wane, and you read this, my penultimate letter, I hope you will remember that although the focus of this magazine includes “systems,” with an emphasis on bringing AI theories into practice, it also includes “intelligent” and that’s something that we as a field mustn’t ignore. We’ve made a lot of progress, but the journey is far from over, and the original goal is still far over the horizon.

Happy Sailing,

Jim Hendler

A few “google” results reveal that cuil may use some clustering technique to build such hierarchies. It is interesting to think will such hierarchies indeed improve search experience. When I search “Semantic Web”, cuil recommends me to browse “Ontology (computer Science)“  and some of its sub category; it also suggests me to look at “James Hendler”’s homepage. I would say that it will be very useful for exploring.

Building meta data using machine learning technology is a cool thing. On the other hand, I believe that human intervention is also critical. When wikipedia knowledge is used in clustering,  I expect some gain in recall or preciseness. As “Ontology (computer Science)” is a wikipedia page, I guess that cuil may have already used wikipedia information in their results.

Also don’t forget the “network effect”. I have created a prefix-based, syntactical gmail label hierarchy for a while. I really like to share part of the hierarchy to my friends, so that when I send a mail labeled with “party”, then they don’t need to relabel it again. If millions of users can share their small hierarchies (not only on gmail, but also on flicker, youtube, twine, etc.), each is connected somehow to hierarchies of friends and family, eventually we will have a very large network of ontologies which may improve search much more than we can do now. Just a random thougt.

P.S. I found one interesting thing. Cuil caches my wiki page at Iowa State University. However, that page should be offline no later than May 2008, while Cuil was online officially only on July 28, 2008.  It seems its crawler has been alive for a while.

Jie Bao

I can’t help imagining that one day, when OCR agents get smart enough to pass the captcha test (I strongly believe that day is not far away), what test we will use to identify a human on web.  Math? That will be easy for a good program. Scrabble? maybe, but not that secure. Ask for a Shakespeare’s sonne? Or the end year of world war II? That looks more likely to succeed. But…There are two issues.

First,  an agent may have access to a knowledge base. With projects like Dbpedia, human knowledge has been KBized in a speed never seen before in the history. A query as ” the end year of world war II” may be answered by a semantic web agent fairly quickly. I can imagine that someday we will have to design increasingly hard questions (like art things)  to identify a human and fight spamming.

The other issue is that a human may have NO access to a knowledge base. Many, many people in the world does not know “the end year of world war II”, even if they may be knowledgeable in other things. They may not even know where to find such a knowledge. Also, they can get bored when been consistently asked such captcha questions and quit — technically, that means they failed the test thus are not “human”. When captcha becomes increasingly hard (like art things), more and more people may fail in one reason or another (including boredness). That will also lead to the failure of the identification system.

Will semantic web help spamming by designing smart agents? :) Maybe, let’s wait and see.

Jie Bao

Some of our Tetherless World researchers including me have just written a short paper to sell the idea of constructing a “webtop” using semantic technologies. In short, a webtop is a desktop on the web, that does similar jobs such as managing files, doing word processing, managing contacts, scheduling tasks, emailing, etc. Please see some examples of webtops with pretty GUIs.

Almost one decade ago, there has been hot for a while for the concept of “network computer”. At that time, a network computer means some low-end computer with limited storage and computational capacity that relying on the network to get great power. The webtop idea reminds me of network computer as they, while are different in many aspects, share the same idea of powering users with networked infrastructure. Ten years ago, this vision was tested with physical computers but largely failed, while today, with the advance of technologies, is revived by allowing users to create virtual computers that only exist on the websphere. I have many reasons to believe this time it will not only survive, but also prevail.

One reason is from my personal experience. From about two years ago, I stopped installing many software that have been with me for many years: Encarta is replaced by Wikipedia.com, Outlook is replaced by Gmail, MS Street is replaced by Google Maps, MS Word is replaced by writing in wiki, Powerpoint is replaced by online latex writing with the Beamer package, among a long list of other things. Browser is the application I stayed for more than 80% of time when I’m on my computers. There is indeed a strong need for me to organize all such online applications and data — simply bookmarking is barely a solution. I need something that can organize them, enable me quick access to them, and last but not least, pretty and neat. A webtop does exactly those things.

How semantic technologies help in providing a webtop? Actually, long before the term “ontology” getting popular, users are already creating ontologies on daily bases: email classification, creating file folder trees, grouping contacts or naming a photo as “Wedding picture at Troy”, all those efforts are creating relations between things or annotating a “meaning” to an entity. With semantic technologies, those relations and annotations can be made explicit so that data can be more easily managed and queried. For example, I may query that “find all 2005 photos of my friends”, or “show all meetings (even if they are not called meeting, such as “briefing”) in the past month”. A webtop based on semantic technologies will make such an ability universal to any application on its top.

There have been controversies about semantic web ever since the term is coined. I think this is partly because the semantic web community as a whole, failed to provide enough end-user friendly tools that can do something helpful in daily life. I wish to see more tools to help daily web activities: semantic email, semantic blog, semantic calender, semantic abstract of news (a little more than RSS), tagging files (picture, mp3,…) with taxonomy, etc. Even more important, to survive, such an application should never ask users to learn RDF or anything needs more than 3 minutes to understand. Bring such applications together, it’s a webtop. I believe something like this is one of the killer apps the community has long been waiting for.

Jie

As a “web of data”, the Semantic Web, boosted by Linked Data efforts, presents web users a maze of RDF graph with billions of arcs (triples). To explore the maze, below are some html browser approaches I came across:

• Hyperlink based RDF browsers, such as hyperdaml, Longwell, OpenLink RDF Browser, tabulator
• Search Engine based browser, such as Swoogle navigation (and a paper behind it),  dbpedia

An alternative approach is graphical browser, which seem to be more intuitive to end users. An interesting blog Large-scale RDF Graph Visualization Tools covered a handful of useful resources including something I never encountered and even links to 28 visualization software packages. Of course the list missed some RDF viz browsers such as FOAFnaut, Welkin, and self visualization. It is notable that scalability is still bugging most of the visualization approaches due to the limit of memory size: my last experience was “Otter had a hard time when processing a graph with over 10,000 nodes”.

There are still many user interaction issues beyond the browsers (e.g. search engines, semantic wiki), and a well-designed UI component is probably the key to the Killer-App of the Semantic Web.

Li Ding

Digital Object Identifier (DOI), on the other hand, has fast growing instance data space in the publishing industry. Unlike DC, DOI requires more agreements including (i) more mandatory properties, (ii) more restrictions on the value of properties; and (iii) a federated metadata registration mechanism. These features ensure better structured and interoperable DOI instance data.

From the above study, we may raise the following hypotheses:
1. simplicity and flexibility can lower adoption cost, but they should be carefully enforced to avoid damaging interoperability
2. restrictions (e.g. the range of property value) can ensure data quality and thus promote interoperability
3. making more information interoperated among systems is preferred to making all systems interoperating
4. interoperable metadata should support non-trivial automated data integration, such as and reference resolution.

Further readings
[1] Beall, J. (2004), “Dublin Core: an obituary“, Library HiTech News, Vol.21, No. 8, pp 40-1,
[2] Jill Hurst-Wahl (2007), “Dublin Core?”, (the comment is more interesting than the blog) access on July 15, 2008
[3] Allan Cho (2008), “Dublin Core is Dead, Long Live MODS“, access on July 15, 2008

Li Ding

OWL Mobile is powered by Jena and Pellet, operating remotely, to provide speed and battery performance mobile devices users expect from their applications. Load one or more ontologies through the Load Ontologies tab. Supply a URL to a custom ontology or use the list of past ontologies. Once you’ve loaded an ontology, use the “Classes”, “Properties”, and “Individuals” tabs to browse through the ontology. Clicking on an item will expand it and give additional information about that particular object. Links which point to other members of the ontology will switch to the appropriate URI when clicked. External links such as web pages, email address, and phone numbers will open the appropriate application on iPhone (phone numbers won’t work on the iTouch) when activated.

Point Safari to http://onto.rpi.edu/demo/owlmobile2/ to try the application. Feel free to bookmark it or add it to the home screen for easy access.

Evan Patton

Like many of the other blogs on this site, the Tetherless World blog has a number of different authors who write our pieces.  On PlanetRDF, however, if we don’t sign the blog, you cannot tell who wrote it (except by guessing) unless you link over to the original blog site.  This is odd as the RSS feeds from most of these blogs use some form of author field.

I did a little mousing around, and best I can tell there seem to be a bunch of different ways the different blogs report the author, with the RSS author element not being the most used.  I wonder if either we the bloggers, or PlanetRDF’s keepers, might not think about fixing this somehow?

At the very least, perhaps we could use the low tech solution and sign our blogs.

cheers

The masked blogger

Although it may not be a new idea (for example, 3721.com, now part of Yahoo!, has provided a service of supporting urls in Chinese for years), having local names other than Roman characters is absolutely a good move. About 10 years ago, I was asked to teach one of my father’s colleague on how to use computers; it was a hard job because she didn’t know how to use keyboard, which in turn because she didn’t know what are characters “A”, “B”, “C”. My mom is better: she is now a daily web surfer and she knows Roman characters - but she can never remember English words like “Google”, not to mention google.com. What she does now is to set a hub page as her browser’s homepage, with a Google link on it (and of course, in Chinese). She uses baidu.com, a Chinese counterpart of Google, more frequently than Google, partly because the word “Bai Du”, which literally means “a hundred times”, is much easier for her to remember (on the other hand, Google’s local name “Guge” is almost meaningless).

We people in academia are so used to our (both language and technical) education and sometimes take many things for granted. Two weeks earlier at the Tetherless World Grand Opening, Wendy Hall, the ACM President-elect, had mentioned that in her recent visit to China for the WWW 2008 conference, she was surprised to learn that there is such a huge part of web that is only in Chinese. “Chinese may be the most popular language on the web in the future”, she said. This may or may not become true, but I agree that web technologies should be easier to use and consider internationalization even more.

However, “ease” means differently for different people. When my mom learned to use mouse, she had to use her both hands to control it :) — and she did not give up only because she wanted to use computers to communicate with me. Last weekend, I tried to teach my father-in-law to use computers, he also had a hard time to control the mouse: regular computer users have an _instinct_ to locally relocate the mouse so we never feel “the line is too short”, but he has no such an instinct.

I’m a little off the topic. But what I want to say is that computers should be designed not only for the youth, but also for seniors; not only for English-speaking people, but also for the other 3/4 of people in the world; not only for geeks, but also for grandmas.

As to the Semantic Web, we should also always keep our “users” in mind. Who gonna use semantic web? What things are on the top list we should support? I have been long thinking about this question: as most of our daily web activities are emailing, blogging, calendaring, searching, etc., why there is still no end user oriented semantic tools to help us for such activities? For example, I have tried many “semantic search engines”, e.g., Swoogle, SWSE and Sindice, none of them can be considered end-user oriented: I cannot explain most of their results in RDF to my mom, just for an example. Google is a killer app, as my mom can use it even if she cannot spell “Google” itself. We will need something like that.

Jie Bao