A.I. something to build on: it's always hard to predict the future, but who would have thought that artificial intelligence's big contribution to education would be to help the teacher.

You may think of artificial intelligence as the precious brainchild of scientists in remote laboratories or the driving technology behind the renegade computer HAL in the movie 2001: A Space Odyssey. But computer programs that use AI have come out of the laboratory and down from the big screen, and, in all likelihood, they are already in some classrooms in your district.

In schools around the country, computers have been coaching--and sometimes grading--essay writing, as well as tutoring algebra, geometry and chemistry. Thanks to computerized programs that can learn the patterns of human reasoning and to the widespread use of PCs as a delivery system, a new generation of AI-inspired products has been attracting the notice of educators.

"In selected domains, such as medicine and education, artificial intelligence has come to fruition and is actually able to support--but not replace--a professional doing his or her work," says Scott Elliot, the COO of Vantage Learning in Yardley, Penn.

Vantage has made its mark with IntelliMetric technology, which the company developed five years ago to score essays automatically. After digesting almost 300 writing samples graded by human experts, the Intellimetric program developed algorithms for evaluating any other essays that come its way.

The new technology provides the brains for Vantage's MY Access, an online program that gives customized support to student writers, who choose from better than 50 broad topics ranging from "The Effects of Technology" to "A Memorable Childhood Event" to an assortment of literature comparisons.

Once students submit a finished essay, they receive an instantaneous response, scored from 1 (lowest) to 6 (highest) in five areas, including content, organization and mechanics. An explanation of what the grade means and how to improve it accompanies each score. The student writers also can see where they have made grammar mistakes and can view model essays on the same topic. They can rewrite as often as they want and store each version.

AUTOMATED GRADING FREES UP TEACHER'S TIME

For Nancy Bosserman, the director of staff development and accountability for the Whither Union High School District in Whither, Calif., bringing AI to essay writing has changed the approach of students and teachers alike. Bosserman piloted MY Access in remedial writing classes two summers ago and has since incorporated the program into all grade 9-11 classes at the district's five comprehensive high schools.

"Technology is something that our kids are remarkably comfortable with," she says, "and they almost attack this like they would a computer game. They want to beat the computer. We'll find kids revising five, six, seven times to get their scores up. So we get a lot harder effort out of students this way than we get in a typical English classroom."

"We also found in the summer school pilot that the kids were becoming wonderful peer editors. They were helping each other, looking over each other's shoulder. Their buddy would come over and say, 'Look at that. You only have two sentences in your second paragraph. You have to write more stuff. Why don't you give an example of something that happened to you.'"

"What happens is that you' re fostering an environment where students can write frequently and get frequent feedback," adds Elliot. "It dramatically changes the whole context for writing instruction."

Bosserman notes that English teachers who once staggered under as many as 150 student papers at a time have been able to transfer some of the heavy lifting to MY Access and have refocused their efforts on conferencing with students about the computerized feedback they have received.

"The students will print out the response that they got and sit down beside their teachers and they become collaborators together," Bosserman observes. "The teacher is able to say, 'Why do you think the computer told you this? What might you do to make this part of the essay better?'"

The Whittier Union district has been using the software not just to coach student writers but to formally assess their work. To save time and money, Bosserman has replaced the district's hand-graded pre- and post-writing tests for all 9th-11th graders with the automated versions.

Vantage estimates that hand-grading costs $4 an essay, compared to $1 each for artificially intelligent machine-grading. The company reports that its technology is used for large-scale grading of essays in Pennsylvania. Delaware, Virginia and Oregon.

The Educational Testing Service, meanwhile, has rolled out E-Rater, its own Al-inspired program, which is widely used in college placement tests. The push for automated essay grading has accelerated with the looming addition in 2005 of an essay section on the SAT.

FREEING UP TEACHERS' TIME

Tutoring, rather than testing, is the name of the game for the Al math products that have been taking root in schools. The Carnegie Learning Program, developed at Carnegie Mellon University in Pittsburgh, has gone to the head of the math class with its Cognitive Tutor products in algebra, geometry and integrated math.

These applications employ algorithms that model the various ways in which students solve math problems. They assess how well a student solves particular problems and select future items to reinforce weak areas and avoid unnecessary repetitions. They also provide questions and hints to help the problem-solving process along.

For the past four years, veteran math teacher Roland Connors has had his students use Carnegie Learning's Cognitive Tutors in first-year algebra, and more recently in geometry, at Denver West High School in Colorado.

"In algebra, there are multiple paths to the solution, and this program allows the kid to develop his own procedure, so long as it's correct," says Connors. "The power of that is that the student visualizes the problem and has developed a procedure that is correct and meaningful to him."

"We're trying to help students with 80 percent of their questions and difficulties they might have in understanding mathematics," explains Steve Ritter, the senior cognitive scientist for Carnegie Learning. "For the other 20 percent, they are able to call the teacher over and talk one-on-one. It frees up the teacher's time to talk about the high-level conceptual issues."

"It allows me to distribute my resources substantially better," agrees Connors, whose students use Cognitive Tutor in the school's computer lab. "I do an enormous amount more coaching and a great deal less lecturing than I might otherwise."

Five of Denver's 10 high schools are currently using the Cognitive Tutor in Algebra I. District math coordinator Jim Loats is overseeing the program's implementation in the remaining schools and says he's been encouraged by the student and teacher response.

"We had really good success in summer school," he says. "And it was liked by the students, which is an accomplishment for summer school algebra. It also has helped very traditional high school math teachers make a change in their curriculum."

Another company, Quantum Simulations, in Murrysville, Penn., has brought digital tutors to high school chemistry. Students receive pointers and step-by-step assistance in navigating chemistry problems and concepts, from mastering the elements to balancing chemical equations.

So far, the early returns on student achievement have been matching the anecdotal responses to these AI products. In Denver's summer math program, for instance, the pass rate for algebra students using Cognitive Tutor was 84 percent, compared to 75 percent of students in classes that did not use the program. Likewise, 21 of the 25 writing students in Whither Union's summer pilot program raised their average scores on a 4-point scale from 2.0 to 2.8, a sizeable increase.

Still, for all the technological advantages offered by these programs--and the assurances that they are tools to help teachers rather than machines to replace them--they are prone to problems. In Denver, for example, the limit on computer lab space means that the district may not be able fully to implement the Cognitive Tutor in geometry.

The chemistry help provided by Quantum Simulations also can potentially do too much for unmotivated students. "When you have a system powerful enough to understand chemistry problems, it can do your homework if you want it to," explains founder and CEO Benny Johnson. "We've aimed this product at diligent students who want to learn."

The issues become more prickly in the higher stakes area of automated testing. The College Board has indicated that it will hand-grade the new essay component of the SAT because many students taking the tests might not have access to the required computers.

Critics such as FairTest, the Cambridge, Massachusetts-based organization that has for decades studied inequalities in standardized testing, say that AI, and the algorithms on which it depends, may not be completely reliable.

"Algorithms by definition are simplifications of complex realities. They have to be finite, simple, and quantifiable," explains Robert Schaeffer, FairTest's public education director. "It is hard to imagine an algorithm that views as great writing the different styles of Ernest Hemingway and James Joyce."

Vantage notes that its AI programs flag any writing styles that look unfamiliar so that teachers can take a look. But those who create these programs admit that they are no more perfect than the humans who design them.

Whither Union's Nancy Bosserman acknowledges that her district hit a "speed bump," when she discovered that MY Access had been normed to high-risk eighth graders. After the scores for Whither Union's students came in too high, Vantage had to recalibrate the program by incorporating a newly graded set of essays.

Bosserman says the changes were worth waiting for. "I've wondered for a long time, 'How is technology going to help education?'" she says. "It can't be just video games or little fun exercises to improve vocabulary. There has to be more to it than that. And this is the first time I've actually felt as though I understand how technology is going to help education.

Questioning Spellchecker

Programs that automatically check spelling and grammar have become a familiar sight on desktops, especially if you use Microsoft Word. However, recent research from the University of Pittsburgh has highlighted one additional mistake--the error of placing too much confidence in this software.

Researchers studied 33 undergraduates, nearly half of whom used Microsoft Word to correct a business letter containing three categories of problems: actual spelling and grammar mistakes; mistakes that the software missed; and perfectly good sentences that were identified as having mistakes.

For the students who corrected the letter without the computer program, the results varied. Those with higher SAT scores left an average of five errors in the writing sample. Their counterparts with poorer SAT scores averaged 12.3 errors. For the students using Word's grammar- and spell-checking programs, the higher achievers soared to an average of 16 mistakes, while the lower achievers increased to 17.

Although both groups benefited in places where the software correctly flagged problems, they stopped looking as closely for other errors--the ones the software was missing. And when the software recognized as wrong the sentences that actually were right, all students created additional mistakes by trusting the program's bad judgement.

"The results imply the need for careful use of intelligent agents, [which] will not substitute for user expertise," concluded the University of Pittsburgh research team in their report earlier this year.

AI's Cousin: Computerized Adaptive Testing 

Some standardized tests are becoming pretty smart although they work differently than their AI-driven cousins. Computerized adaptive tests do not try to fathom the reasoning processes of the students taking them. Instead, they instantaneously calculate the score of multiple-choice tests as an individual test-taker answers each question. The computer program chooses the difficulty of subsequent questions based on the interim scores. The questions come from a pool of more than 25,000 items in reading, math, language and science that already exist in a computerized database.

"If a child gets the first item on the test right, we move up the measurement scale a ways," explains Allan Olson, president of the Northwest Evaluation Association in Portland, Ore. "We do this based on a lot of experience. If the child gets that item right also, we move the child up another step. The third item the child gets wrong. So we do the math. We score the three-item test. We estimate the child's ability, and we go back to the item pool and select an item right at that estimated ability."

The process continues for the duration of the test. As with AI-inspired tests, students and their teachers get the results almost immediately.

NWEA notes that its tests are being employed in 37 states. This past academic year, Idaho used NWEA's tests in all its public schools and became the first state to implement computer-adapted testing at the K-12 levels.

Ronald Schachter, [e-mail], is a freelance writer based in Newton, Mass..