Computer, Heal Thyself

Intelligent machines that can learn and fix themselves are becoming a reality

Back in the Fall of 2001, Paul Horn was wrestling with the computer industry's demons. The former University of Chicago physics professor had been running IBM's (IBM ) research labs since 1996. During that time, the Internet, wireless technologies, and open-source software had exploded in popularity. But the complexity of all the new technologies threatened to overwhelm customers, both corporate and individual. He sat down to bang out what he terms ``a call to action,'' a paper entitled the ``Autonomic Computing Manifesto.'' ``Things were getting crazy,'' he says, ``and the only way to fix it was to have the entire community working together on the problem.''

His idea was simple. Scientists needed to come up with a new generation of computers, networks, and storage devices that would look after themselves. The name for his manifesto came from a medical term, the autonomic nervous system. The ANS automatically fine-tunes how various organs of the body function, making your heart beat faster, for instance, when you're exercising or stressed. In the tech realm, the concept was that computers should monitor themselves, diagnose problems, ward off viruses, even heal themselves. Computers needed to be smarter. But this wasn't about machines thinking like people. It was about machines thinking for themselves.


Now the results of Horn's epiphany are beginning to arrive in force. IBM just released a trio of software products that monitor computing systems, restart them automatically after a power outage, and shift processing jobs around bottlenecks the way cops detour cars around accidents on the highway. Cisco Systems Inc. (CSCO ) has lined up a series of so-called application-oriented networking products capable of peeking inside the packets of data that move across the networks and making decisions about how to best route them. And next year Microsoft Corp. (MSFT ) plans on delivering smart technologies in the new version of its Windows operating system -- one that will sense when a disk drive is about to fail and warn people so that they can save vital documents and photos.

Tech companies that can deliver such sophisticated technologies will get a jump on their rivals. In the $800 billion corporate computing market, prices are steadily dropping for run-of-the-mill PCs, computer servers, and networking gear. But products that make computing cheaper or easier still command a premium. That's why there's so much pressure on tech companies to make machines that can think for themselves. ``It's one of the only ways they still have to differentiate themselves from the competition,'' says analyst Joe Clabby of market researcher Summit Strategies Inc.

Initially, only a handful of these smart technologies will be visible to regular consumers. In addition to the Windows trouble-spotting software, several companies have come out with technologies that allow employees to plug laptops into company networks without fear that they'll infect the network with viruses. The products halt the PCs from initially gaining access, search them, and kill off viruses or patch them with software that puts them in compliance with company security policies. It's the digital equivalent of being strip-searched by police at the airport.

Most of the early smart technologies are geared to help corporations manage their sprawling networks and data centers. That's where computing has become such a tangle that the average company spends $9 on managing technology for every $1 it spends on creating applications. It also leaves them vulnerable to hacker attacks and makes it difficult for the tech industry to deliver on one of its promises of the 1990s: to make computing as simple to tap into as the electrical grid. ``This is very important,'' says Nicholas G. Carr, author of the book Does IT Matter? ``Automation addresses the complexity problem, which is the biggest problem in computing.''

A few nervy corporations have volunteered to be guinea pigs. ``It's early days, but I want to advance the cause of autonomic computing,'' explains Dennis Callahan, chief information officer of Guardian Life Insurance Co. in New York City. His company uses software from IBM to spot problems in a computing system that helps insurance agents put together sales proposals for clients. It has pared the time required to fix things by 90%.

IBM has made autonomics one of the centerpieces of its technology vision. Horn has assigned more than 100 research scientists to the project, and the company has woven more than 475 autonomic features into 75 software and hardware products. A cool one: a technology called eFuse that senses problems in a computer chip and automatically cuts off problem transistors so that the device can deliver reliable number crunching.

Some of the industry's smart products are essentially digital detectives. Data-storage leader EMC Corp.'s (EMC ) Smarts software creates diagrams of complex networks and, when a glitch occurs, sorts through dozens of symptoms to discover the cause. A few months ago a huge European telecom network run by COLT Telecom Group PLC (COLT ) began to behave sluggishly. Within minutes, Smarts discovered that the problem originated in a single data-routing machine in Madrid. Not only that, but it figured out that the router had simply overheated during a sweltering summer day. Smarts sent out an alert, and a technician moved the machine a few inches to improve air flow. Problem solved.


The next step is to create machines that heal themselves as well as diagnose problems. Think of them as digital doctors. So far, only a few self-healing technologies have made it out of the labs. One, Hewlett-Packard Co.'s (HPQ ) ``virus throttle'' technology, spots evidence of suspicious virus activity in a network, gradually squeezes off connections to the affected machines as it investigates. If real trouble is found, cuts off those machines from the rest before major harm can be done.

Techies are wrestling with how much intelligence they can inject into computers. Ozalp Babaolu, a computer scientist and professor at the University of Bologna in Italy, for instance, believes that it will be too difficult to design self-healing capabilities into large networks of computers. Instead, he believes, machines need to be designed so they are capable of learning through interactions with each other and adapting just as living organisms do. ``We have to do some of the things in computing that biology does,'' says Danny Hillis, a computer scientist and principal at tech consultancy Applied Minds Inc.

That vision for computing is still a long way off. Horn's quest, despite the recent product introductions, isn't over. He warned in his manifesto that much work has to be done, and quickly, to shore up information technology ``before it collapses on itself in a jumble of wires, buttons, and knobs.'' While IBM, Microsoft, and others have made progress, the industry has to make a lot more before it can heal itself. ``I thought this would be hard, but it turned out to be even harder than I expected,'' says Horn.

By Steve Hamm

Before it's here, it's on the Bloomberg Terminal.