Harry Chen Thinks Aloud

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.

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/.