A Defense Department panel of computer experts has concluded that a space-based defense against nuclear missiles is theoretically feasible, but that current plans for "Star Wars" are likely to fail because the Pentagon and its contractors have designed the project improperly and lack the creativity to apply exotic new technologies.

The government's Strategic Defense Initiative (SDI), for which the Defense Department is expected to seek nearly $5 billion from Congress this year, has put too much emphasis on weapons and hardware and not enough on the computer software needed to make the system work, the panel concluded. As a result, the "architectures" proposed by the Pentagon and its contractors during the first three years of Star Wars research are "not likely to be implemented successfully."

In addition, the panel of scientists from government, industry and universities drew a scathing picture of the military-industrial complex as "an industrial culture that resists change."

The arms companies and the Defense Department today are too hide-bound and bureaucratic to adapt necessary new technologies to the SDI project, the panel warned.

"The endless demands of project schedules, the lack of capable staff, the lack of capital equipment, the 'not-invented-here' syndrome, the conservatism in procurement decisions, and bureaucracy have created a culture that resists change and takes only naive risks," the panel said. "SDIO [the Pentagon's SDI Organization] must create a new culture that can adapt to changes more effectively."

Air Force Maj. David R. Audley, SDIO program manager for battle management, said contractors have accepted the panel's conclusions and are beginning to change their approach.

He said that rockets and lasers have not received too much funding, but that software -- the complex computer coding fundamental to Star Wars -- and battle management have been "grossly underfunded."

The panel completed its unclassified report in December. Danny Cohen, a professor at University of Southern California, was chairman of the study, with other participants from Princeton University, the California Institute of Technology, ESL of California and the Office of Naval Research.

David L. Parnas, a computer scientist at the University of Victoria in Canada, resigned from the panel last summer after concluding that its members were biased in favor of the SDI.

Parnas and many other computer experts have said that scientists will never be able to write the error-free, complex computer programs needed to make a space-based missile defense system work.

The Cohen panel agreed that error-free computer software will never be written, but said Star Wars could work anyway by decentralizing the system so that a failure in one program would not render the rest of the weaponry worthless.

The panel acknowledged that a decentralized system would require more weapons performing redundant missions, but said that is preferable to a " 'cheaper' system that simply does not work."

The SDI, which President Reagan kicked off early in 1983 with a call for research to render nuclear weapons "impotent and obsolete," is intended to find a way to shoot down Soviet nuclear missiles as they are launched, as they fly in space and as they descend toward their targets. Scientists envision scores of permanently orbiting satellites equipped with sensors and weapons such as high-speed rockets and laser beams.

Since the Soviet Union might launch thousands of missiles, each of which could disgorge scores of warheads or decoys, most contractors have postulated that "a Star Wars defense system would need to respond to an offensive strike as a single organism, coordinating perhaps millions of separate actions in a schedule timed in milliseconds," as the panel phrased it.

But the Cohen report, while agreeing that Star Wars decisions would have to be made quickly and perhaps without human involvement, disputed the "single organism" analogy. Instead, the panel imagines many small "battle groups," each comprised of several orbiting weapons stations that could function with some independence.

Thus, the panel concluded, the 10 contractors that were paid $1 million each to devise a preliminary "systems architecture" went astray by designing systems that "demand excessively sophisticated software and cannot be adequately tested."