SCIENCE HAS REACHED the halfway house on the road from helplessness to reliable earthquake prediction. But half a loaf, in this case, may be worse than none.

The last two decades have seen the first examples of effective, dramatically life-saving earthquake predictions. The most notable occurred in 1975 at Haicheng, in China, where thousands who might have died were evacuated from their homes prior to a fierce tremor that leveled much of the city. The very next year, however, brought news of a colossal earthquake disaster in Tangshan, 100 miles from Peking. In that instance, there were no warnings and an almost unimaginable number of people -- 600,000 -- perished.

The problem is that there is still as much sorcery to earthquake prediction as there is science. The only "hard" knowledge is based on common sense: Earthquakes happen along known geological faults, and the more time that has passed since the last one in a particular seismic area, the sooner -- or greater -- the next is likely to be. That is the "seismic gap" theory.

The "soft" knowledge is much more abundant: It is based on "precursors" to earthquakes that have occurred in the past. When there is an earthquake, scientists and earthquake buffs comb through seismographs and other recorded data for clues, in hopes of finding patterns that, if repeated, would enable the next quake to be predicted. Their researches have elicited numerous possibilities such as foreshocks (easy to spot afterward, but how can one know they are not the shock at the time?); unusual uplift or tilt in the landscape; seismicity patterns (such as a "swarm" of microearthquakes followed by a period of quiescence and then the big bang); raised levels of radioactive gas emissions in well water; sudden shifts in water levels in wells, and strange animal behavior (in the Haicheng incident, despite freezing February temperatures, snakes came out on land and froze to death in sufficient numbers to b considered a harbinger of coming upheaval).

But all of these precursors remain no more than tantalizing hints: None occurs with every earthquake, none occurs only before earthquakes. Plate tectonics confirmed the theory of continental drift. We know that the earth's crust is mobile, we know the plate boundaries and fault systems where that mobility is expressed, but we do not know with any certainty what triggers a particular nudge at a particular moment. The basic scientific understanding of earthquakes is not yet achieved.

As the scientist gropes for secrets hidden beneath the crust of the earth, his or her primary concern is science -- how does the earth's interior function? -- not predicting earthquakes. But whatever the purpose of their collection, the scientists' data impinge on society and on national policy-making as they bring the hope for reliable forecasting closer to fruition.

The scientist is responsible for the first step in earthquake prediction. At the present halfway point, there is evidence that the "hard sciences" will have to grapple with "soft" issues of social science to carry out that charge. How should one go about predicting earthquakes? Society is volatile and liable to panic under warnings of catastrophe, and the majority of earthquakes do little harm; the danger at this stage is that a clumsy prediction can wreak more harm than the earthquake it forecasts.

Strictly speaking, an earthquake prediction must pinpoint time, place and magnitude. There is an important distinction, however, between a prediction and a conjecture. A prediction, based on a wealth of evidence, defines an event expected to occur; a conjecture describes from an untested theory: the event it forecasts will occur if the theory is correct.

Profs. F.F. Evison and John L. Roberts of Victoria University in Wellington, New Zealand, make another useful distinction, between "wildcat" and "housecat" (or domesticated) predictions. A wildcat prediction emerges from outside the main stream of earthquake-related research and is likely to lack adequate official consultation, peer review checks and assessment of implications. Housecat or "domesticated" prediction tends to automatically take account of those dimensions; it stems from the realm of orthodox earthquake studies or from the earthquakes policy community (primarily, the U.S. Geological Survey or its funded projects in universities).

Dr. Brian Brady's assertion that the greatest earthquakes of all time would strike Lima, Peru, on Aug. 15 of last year -- finally rescinded at the eleventh hour on July 27 -- serves as a worst case for wildcatting. An untested hypothesis (or conjecture) -- the giant quake, had it occurred, would have been the first successful test of Brady's "inclusion collapse" theory -- became equated with a full- blown prediction. (Note that this bears no relation to the conventional studies of earthquakes precursors. Brady, employed by the U.S. Bureau of Mines, had done work on rock collapses and his work was relatively unknown to the seismological community.)

Numerous actors became involved in the international drama. The media brought the ''prediction" to light in 1980, ending three years of obscurity since its publication. Officials then arranged for Brady to have an audience with President Belaunde of Peru. The scientist's peers called him before the National Earthquake Prediction Evaluation Council (NEPEC) and cast doubt on the validity of the prediction. (The Peruvian coast is highly seismic, however, and therefore none dared assert with finality that quake action was out of the question.) Meanwhile many who stood to gain from the confusion used the prediction (while not necessarily believing it) for their own ends. Alberto Giesecke saw to it that Peru's seismograph array, which he oversaw, was beefed up; AID's Latin American relief program sought additional funds. Airlines seemed to profit as departures from Lima were claimed to be on the increase.

In the end, August passed quietly in Lima.

The Brady episode highlighted the difficulty of coping with such unofficial and unconventional predictions. There is a danger that in quashing troublemakers, future scientific conjectures might be discouraged. Wildcatting -- making the mental leap to a prediction prior to testing the hypothesis from which it springs -- has contributions to make to science. But earthquakes affect people, and wildcatters in this arena will have to bear in mind the furor their ideas might arouse.

If most of its effects of the Brady prediction were unfortunate, at least it served to focus attention on the need to clarify and make orderly the process leading to formalized earthquakes prediction. Had it been possible from the start to identify the Peru forecast as a conjecture rather than a prediction, and to make the distinction clear, many Peruvians would have been spared needless worry.

Other lessons emerge. Government agencies should have been informed of the prediction at least at the same time as it appeared before the public. From the start, the Peruvian government should have been consulted or, better yet, included in the research. When all affected parties work together, problems are much less likely to ensue.

The pinpoint timing of Brady's prediction was the main cause of all the trouble. His hypothesis required it as a scientific test of his theory; its audience was expected to be scientists. Brady's error was in at first ignoring the human beings likely to be alarmed by the potential consequences for them if the theory proved correct -- and then refusing until very late to shift or soften the blow when public concern did become a problem.

Hard upon the heels of the Brady prediction's demise came a best case; a "housecat" prediction or warning published in Science magazine Aug. 7, 1981. The forecast is again for a foreign nation: Nicaragua should expect a major earthquake greater than magnitude 7 off its western coast some time in the next few years or decades.

In this report, three U.S. Geological Survey scientists and one Nicaraguan scientist (from Managua's Institute of Seismic Investigations) present the traditional "seismic gap" theory and their reasons for believing that it bears on the data underlying their prediction. They then support those data by comparison with the findings of others and by reference to historic quake activity in the area. They further assess the implications of a quake for loss of life and for damage to local tructures.

The team is clearly accepting the responsibility to alert both the populace and the government to impending danger and is doing so with tact and diplomacy. By means of the binational research group, they avoid international friction. While the warning will and should lead to preparations for relief and hazard mitigation, it should not cause undue alarm.

Admittedly, the best case is also an easy case: The seismic gap theory is common- sensical, fewer lives are at stake than in Brady's case and the time factor is defused by being broadly defined -- so much so that Nicaraguans cannot be sure exactly when to start running. The right timing stated imprecisely is better than the wrong timing stated precisely. The report on Nicaragua will hopefully be but a first step, followed by more precise warnings about the impending quake's timing as precursory activity becomes evident.

But there lies the rub. There is no certainty at present that those final steps will be -- can be -- forthcoming, due to the scientists' lack of confidence in their abilty to predict timing on the basis of precursory phenomena. There is a clear need to know, but there is no knowledge, only guesswork. Is there a need to guess?

Scientific risk taking is necessary sometimes to save lives; there is a range of risks which must be taken. Scientists in southern California are today saying that the combination of earthquakes precursors there -- uplift in Palmdale, high radioactive gas emissions, changes in water level -- might have resulted in a strong warning by now if their country were China. Indeed, some in the know are quietly moving away. But American government officials, those with the final word on warnings, cannot help but be mindful of two things -- the shaky nature of the data (there is no clear cry for a prediction, no Eureka!) and the consequences of an erroneous prediction: anger, ignoring the next warning, going to legal battle over harm to business or tourism. So they wait.

Halfway to earthquakes predictions that we can trust, we have learned that any prediction with significant lead time will have to be handled with some but not too much caution and with clear steps from orderly notification of officials, to consultation (international or state and local), through scientific acceptance (or rejection), to public hazard warning.

Officials have to toe the line between failing to warn and crying wolf. The former is always the easier and more likely course. Scientists have to avoid endangering lives through careless statements while responding to the need for accurate precursory information.

The need is great and the knowledge small. We need officials with the vision to take risks in order to save lives. We need scientists with the vision to generate warnings and the advancing knowledge to increase their trustworthiness. We need warnings that are neither wild nor so domesticated that destruction predates prediction.