Harry Chen Thinks Aloud

BusinessWeek reports Microsoft research scientists are using technology similar to spam blockers to attack HIV, the virus that leads to AIDS. This approach came about when Microsoft scientist David Heckerman discovered similarities exist between spam messages and HIV.

From Heckerman’s perspective, HIV is like a cagey spammer. After attacking a cell, it injects its own genetic material and proceeds (much like a spam jockey who has commandeered an unprotected computer) to manufacture thousands of copies of the virus. It’s a notoriously sloppy copier, but that adds to its vigor. Each mistake launches mutant viruses into the system. Many fail. Some, though, survive–and resist the drugs.

… the connections between spam and HIV boil down to mathematics. He analyzes both scourges by studying statistical relationships among their ever-changing features. Consider the word “Viagra.” Sometimes it shows up in legitimate e-mails. Often it appears in spam. If researchers study thousands of e-mails, they can calculate the percentage of e-mails with that word that are spam. That’s one clue. But the spam-filtering machine needs to know more than that. What other features in an e-mail signal that it’s spam? Are certain fonts particularly spammy? What about e-mail addresses or types of punctuation? The trick is to figure out which combinations of these features identify an e-mail as spam. Each decision can involve thousands of variables and millions of different calculations.

When he brought his plan to Bill Gates, the company chairman “got really excited,” Heckerman says. Well versed on HIV from his philanthropy work, Gates lined up Heckerman with AIDS researchers at Massachusetts General Hospital, the University of Washington, and elsewhere.

This is a classic example of technology cross-fertilization. Many hard to solve natural science problems involve statistical analysis. Since computers are extremely capable of performing this type of tasks, advanced computer science technology like spam blockers can be exploited to solve non-technology problems like detecting HIV. I think we see more of this kind of innovations in the near future.

Original Story: Using Spam Blocker To Target HIV, Too

SICgen is an AI software designed to create bogus scientific research papers. It’s rather amusing to read papers created by this software. They are properly formated for scientific publication — the papers are created with graphs, figures and citations.

Some random papers created by this software:

• Scatter/Gather I/O Considered Harmful
• Unstable, Reliable, Symbiotic Theory

Can machines write papers? That depends on how you define “paper writing”. SIGgen can certainly produce papers that look like those written by people, but the quality of those papers is no match to those composed by the human mind.

If you study papers created by SICgen closely, you will observe the following patterns: (1) SICgen is good at producing papers with a consistent style and format. Section titles, table of contents and citations are all properly formatted. It does a better job in doing this than most people. (2) SCIgen is poor at “writing”. Sentences in the same paragraph and paragraphs in the same paper are often logically disconnected and lack coherence. Careful readers will also notice many writing style issues (e.g., the use of uncommon abbreviations without first defining the terms) .

Though SICgen is not a great writer (at least not yet), it serves a special purpose. It’s a tool to prove that the review process of many scientific research conferences are lack of quality controls. The MIT students who created SICgen has been submitting machine-generated papers to different conferences. In few extreme cases, their papers were accepted for publication, and the authors were invited to present them to the research community. Read the full story here.

AideRSS is a RSS filter service that helps user to identify important news and blogs. This service implements the PostRank algorithm that analyzes the popularity of individual feed items on the Web, and assigns each item with a number score. Feed items with a high score are considered more important and worthy of reading.

How does it work? See AideRSS video demo.

I think AideRSS is a very useful service. Many people face the problem of information overloading. The few dozens RSS feeds that we subscribe to don’t really help to solve the problem. Sometimes I think the ease-to-use RSS technology actually contributes to my information-overloading anxiety. AideRSS’s “intelligent” filtering can help to reduce the amount of time that we need to spend on manually filtering out information noise and reading only news that matter.

Nevertheless, AideRSS is not a silver bullet. Because the PostRank algorithm relies on external information such as trackbacks, post comment traffics etc. to rank individual feed items, it could falsely assign low ranks to important news that hasn’t yet been widely discussed in major social Web sites (e.g., del.icio.us, bloglines and technorati).

I think I will try out AideRSS service in the next couple weeks and see how it works. While I see there are obvious weaknesses in this service, but it’s definitely a major step towards solving our information overloading problem.

DARPA announced its prize money $3.5 million for its urban robotics race next November, CNet News reports. The agency will grant $2 million for first place, $1 million for second and $500,000 for third (see also the official announcement).

About DARPA Grand Challenge

Created in response to a Congressional and DoD mandate, DARPA Grand Challenge is a field test intended to accelerate research and development in autonomous ground vehicles that will help save American lives on the battlefield. The Grand Challenge brings together individuals and organizations from industry, the R&D community, government, the armed services, academia, students, backyard inventors, and automotive enthusiasts in the pursuit of a technological challenge.

In order to win the competition, vehicles must complete 60 miles within six hours. In addition, vehicles must obey traffic laws while merging into moving traffic, navigating traffic circles, negotiating busy intersections, and avoiding obstacles. The location of the competition is yet to be announced.

Read the rest of this entry »

The New York Times has an interesting article on the field of Artificial Intelligence (AI). It walks through the AI research milestones since 1950 (see timeline).

The article also talks about some recent development in AI.

• Stanford University is developing a robot that can use a hammer and a screwdriver to assemble an Ikea bookcase.
• Microsoft is working on the idea of “predestination”. They envision a software program that guesses where you are traveling based on previous trips, and then offers information that might be useful based on where the software thinks you are going.
• Since 2001, the voice recognition software of Tellme Networks has improved its accuracy from 37% to 74%. Tellme supplies the system that automates directory information for toll-free business listings.

One weakness of this article is that it gives a false impression about AI. It presents AI as the research of building intelligent robots (or human-like machines). The article begins with an introduction of the Stanford robotic project, and ends with a quote from an AI scientist “It’s time to build an A.I. robot. The dream is put a robot in every home.”

I strongly discourage people to think AI is all about robot building. AI is more than HAL or IBM’s Deep Blue chess playing program. AI is the science and engineering of intelligence machines, especially intelligent computer programs (see AI basic questions). In addition to robotics, other research topics of AI include speech recognition, expert systems, natural language processing, computer vision, heuristic classficiation, planning, logical inference, search, and the Semantic Web.

Artificial Intelligence Genealogy Project (AIGP) is an NFS-sponsored project to create an academic family tree for researchers in the field of Artificial Intelligence (AI).

The field of AI will mark its 50th anniversary in the summer of 2006… Having the academic genealogy of AI … will be symbolically valuable for the field.

In addition to the symbolic value of the academic genealogy, we believe that the data will be a useful resource for historians and social scientists studying the nature of science. Intellectual influence among scientists is an immensely rich and complex relation. The student-advisor relation is an important special case of intellectual influence, and one that is approximate by the formal structure of the academic genealogy. We intend for the academic genealogy to be a resource for future studies of intellectual influence in AI, perhaps along with analyses of publications, authorships, and citations, available through resources such as CiteSeer or Google Scholar.

Here is a snapshot of my AI family tree:

Read the rest of this entry »

From AAAI newsletter:

PBS will air “the Great Robot Race” on March 28th, 2006 at 8:00 PM (check local listings). The program chronicles the DARPA Grand Challenge last fall where driverless vehicles raced across the Mojave Desert, including those created by Carnegie Mellon University’s “Red Team,” led by Red Whittaker, and “Stanley,” the winning Stanford University entry.

An interview with Stanford’s Sebastian Thrun, as well as additional background and program information is available at http://www.pbs.org/wgbh/nova/darpa/.

I just discovered that IEEE Computer Society now offers the availability of the latest magazines and transactions content through RSS. Now you can monitor your favorite IEEE magazines and journals the same way you monitor news and blogs.

Some of my favorite IEEE RSS include

• Computer
• IEEE Intelligent Systems
• IEEE Internet Computing
• IEEE Pervasive Computing