MODERN SCIENCE FICTION, which was launched in this country under the banner of dramatizing the scientific wonders of the world to come, has always had trouble reconciling its twin ideals of scientific plausibility and artistic ambition. The commercial magazines which shaped sf for most of its history subordinated both ideals to the gospel of telling a good story, although the magazines' heyday of the early '40s saw serious efforts by some writers, notably Robert Heinlein and the young Isaac Asimov, to accommodate as much of both as their talents encompassed. The '60s, a cauldron of controversy in sf as elsewhere, saw the polarization of these camps, and for some years afterward serious sf was marked by a relative reluctance to engage its seminal themes of scientific breakthrough or the strangeness of the universe.

Gregory Benford comes closer than any other writer to healing this rift, as his much- admired Timescape has shown. For Benford and many younger sf writers of a high-tech, post-Space Race generation, the science/art dichotomy was always a spurious one, and the hard details of a future world, its subatomic and interstellar particulars, are valid not simply as metaphors with respectably literary purposes but for having their own textures and poetry. Bruce Sterling, Greg Bear, William Gibson, and the company under review share the (broad) agenda of using the natural and technological surprises the future is likely to serve up as occasions for stranger and more exacting work than contemporary fiction or genre sf usually affords.

Artifact (Tor, $16.95), Gregory Benford's latest and longest novel, deals once more with scientific research, although the field has shifted, at first glance anyway, from the particle physics that informed Timescape to the more mundane world of archeology. Claire Anderson is overseeing a Greek dig into a Mycenaean tomb when the eponymous artifact is discovered, a cubic meter of black rock that appeared to hold unusual significance to the Mycenaeans and soon exhibits unusual behavior to scientists. Suspicious that we may be offered an artifact of alien design (by no means Benford's first) naturally arise, but the cube proves to have a natural explanation, and we are not so far from particle physics as we thought.

Deteriorating political conditions in Greece lead to the artifact's surreptitious removal to Boston, whereupon the novel shifts gears from exoticism and political intrigue to the world of scientific research. As before, Benford effectively dramatizes the excitement and procedures of discovery, and his evocation of academic research, its protocols and rivalries, is impeccable. Less assured are his descriptions of Boston, which have a worked-up, guidebook quality to them, as though Benford knew the locale less well than he does Mycenae. When Greek operatives steal back the cube, which destroys their boat and sinks in Boston harbor, the tone shifts again, to international thriller, as U.S. security forces, alerted by now to the artifact's destructive potential (and aware that a kindred phenomenon was left behind in Greece) take charge of things. Benford handles the location and recovery of the cube with the authoritative ,elan of one who loves the intricacies of procedures, but the novel's climax, an extravagant shoot-out on the Greek coast, seemed a final, forced shift into highest gear, and held less excitement for this reader than did the moments of discovery in the Boston labs.

Benford, unusually for a modern writer, acknowledges a division in his writing between serious novels and what Graham Greene calls "entertainments," and it is instructive to remember that Benford published two other novels the same year as Timescape, each essentially an adventure novel. Artifact, despite its wealth of detail and the careful speculation that Benford has put into the nature of his artifact, has more in common with these efforts than with Timescape, and is at its weakest when it imitates the conventions of the international thriller. Benford remains unsurpassed in describing the artistry of a theoretical insight, but his occasional desires to fold these into the yeastings of the popular novel (Shiva Descending, a near-disaster novel written with William Rotsler, had similar ambitions) can hobble his best efforts.

VICTOR MILAN's The Cybernetic Samurai (Arbor House, $15.95), a novel of artificial consciousness and power politics, takes on what bodes to be the quintessential sf theme of the mid-1980s. Like William Gibson and Bruce Sterling, Milan knows that Japan looms large in any cybernetic future, and it is competition between the Japanese zaibatsu -- dynastic corporations -- that prompts development of the first genuine artificial intelligence. For Milan, consciousness requires volition as much as self-awareness, and he makes much of the point that a cybernetic being must necessarily be not only self-programming, but willful. TOKUGAWA, the eponymous artificial intelligence, must be educated not programmed, and shows early willfulness in infantile tricks that bespeak his enormous power. Dr. Elizabeth O'Neill, the American scientist who develops TOKUGAWA for Yoshimitsu TeleCommunications, finds modern Japan deficient in its traditional virtues, and so instills in TOKUGAWA the samurai ethos of bushido. The rather hapless TOKUGAWA is drilled to protect the corporation as a warrior would his lord: a cybernetic samurai.

Milan is conversant in both Japanese culture and theories of artificial intelligence, and is able to dramatize both without stopping to lecture. His evocation of detail is very good, though his style betrays a dismaying recourse to clich,e and ready-made phrases. A more serious deficiency is Milan's weakness in characterization, for most of the principals -- the crippled and emotionally stunted Dr. O'Neill, the wise corporate paterfamilias, his spoiled son -- are stock figures, whose reactions and conflicts seem as predictable as rebounding billiard balls.

TOKUGAWA's commission to protect and, if necessary, avenge his lord is tested to the full in the course of the story, which after the first hundred pages becomes extremely violent. Corporate warfare, betrayal, and eventually nuclear war follow in swift succession, displaying Milan's skill in pacing but building up to a rather frenetic climax insufficiently underpinned by plausibility. The Cybernetic Samurai is an ambitious work with many merits, and while the Japanese virtues of proportion and understatement are not among them, one can imagine better things coming from Victor Milan.

PAUL PREUSS' Human Error (Tor, $14.95) also concerns artificial intelligence, but approaches the issue from the standpoint of biological innovation rather than the construction of souls in silicon. The ambitious Compugen Corporation develops Epicell, an organic biochip that runs on sugar, not electricity, and can replicate itself like a weed. Computers using Epicell can grow additional memory capacity indefinitely, and as the cell itself is no larger than a virus and programmed to grow and learn, any Epicell device will in time become a virtual supercomputer.

The rub, as most readers will guess early on, is that Epicell can cross the blood-brain barrier and lodge in brain tissue, where it promptly begins to take over. Users of Epicell computers develop colds, then retreat into conditions resembling catatonia. By the time Toby Bridgeman, one of Epicell's two developers, realizes what is afoot, his partner has succumbed and the virus is loose.

Preuss knows the worlds of Silicon Valley, academia, and hospital research well, and the sense of authenticity he imparts goes far toward making his story seem plausible. The suave Toby and his socially maladroit partner share a debility in dealing with people, and the novel effectively plays the Frankenstein theme against Toby's attempts to discover what is deficient in his own character. Preuss' scrupulous eschewal of melodrama pays off in giving the story greater conviction than Artifact or The Cybernetic Samurai possesses, and avoids their predictability.

Preuss' variation of the Frankenstein theme is audacious and unexpected, and the handling of his characters' personal problems, though awkward in places, shows laudable sensitivity and complexity. Human Error -- as with Frankenstein, Preuss could have subtitled his novel "The Modern Prometheus" -- is superior science fiction, and a fine novel on any account.

TWO THIRDS of the way through his Worlds trilogy, Joe Haldeman has published his second story collection, Dealing in Futures (Viking, $16.95). Most of the contents were published over the past five years, making them, unlike the stories in Infinite Dreams, the work of an already successful novelist. Haldeman confesses in his introductory notes to sharing the widespread prejudice that novels are more "serious" than stories, though he admits to knowing better. Given that, plus his incontestable observation that stories pay more poorly than novels, Haldeman continues to write them for a variety of reasons: the satisfaction of completing something relatively quickly, the ability to experiment without investing a year's work, an interesting assignment. Unlike the efforts of such writers as Harlan Ellison or Gardner Dozois, whose muses confine them to the shorter forms, Haldeman's later stories are essentially craftsmanlike productions, written with intelligence and wit if not passion.

Several are occasional pieces, commissioned by editors with sometimes a stipulated theme. A number of the others appear to have been written with some eye to the market, and though Haldeman claims that he saw how "Blood Sisters" was a Playboy story only after finishing it, the reader is likely to guess a lot sooner. The author's remarks about the genesis of each story prompt various observations (Haldeman writes as well on request as under his own impetus; humorous sf is rarely commissioned), and generally strengthen the impression of a good writer working competently within his chosen field, without trying to shake up things.

It is interesting to note that the only stories to contain genuine emotion are the two longest ones, "Seasons" and "You Can't Go Back." The rest are variously funny, suspenseful, or horrific, but none can be considered moving in the way Haldeman's novels -- and early stories -- are. Any fan of Haldeman's novels will enjoy this book, though it will be the pleasures of honest craft, not art.