Exploration has two stages: reconnaissance and field study. The goal of reconnaissance is to acquire a broad overview of the compositions, processes and history of a given region or planet. Questions asked during the reconnaissance phase tend to be general — for instance, What's there? Examples of geologic reconnaissance are an orbiting spacecraft mapping the surface of a planet, and an automated lander measuring the chemical composition of the planet's soil.

The goals of field study are more ambitious. The object is to understand planetary processes and histories in detail. This requires observation in the field, the creation of a conceptual model, and the formulation and testing of hypotheses. Repeated visits must be made to the same geographic location. Field study is an open-ended, ongoing activity; some field sites on Earth have been studied continuously for more than 100 years and still provide scientists with important new insights. Field study is not a simple matter of collecting data: it requires the guiding presence of human intelligence. People are needed in the field to analyze the overabundant data and determine what should be collected and what should be ignored.

The transition from reconnaissance to field study is fuzzy. In any exploration, reconnaissance dominates the earliest phases. Because it is based on broad questions and simple, focused tasks, reconnaissance is the type of exploration best suited to robots. Unmanned orbiters can provide general information about the atmosphere, surface features and magnetic fields of a planet. Rovers can traverse the planet's surface, testing the physical and chemical properties of the soil and collecting samples for return to Earth.

But field study is complicated, interpretive and protracted. The method of solving the scientific puzzle is often not apparent immediately but must be formulated, applied and modified during the course of the study. Most important, fieldwork nearly always involves uncovering the unexpected. A surprising discovery may lead scientists to adopt new exploration methods or to make different observations. But an unmanned probe on a distant planet cannot be redesigned to observe unexpected phenomena. Although robots can gather significant amounts of data, conducting science in space requires scientists.

It is true that robotic missions are much less costly than human missions; I contend that they are also much less capable. The unmanned Luna 16, 20 and 24 spacecraft launched by the Soviet Union in the 1970s are often praised for returning soil samples from the moon at little cost. But the results from those missions are virtually incomprehensible without the paradigm provided by the results from the manned Apollo program. During the Apollo missions, the geologically trained astronauts were able to select the most representative samples of a given locality and recognize interesting or exotic rocks and act on such discoveries. In contrast, the Luna samples were scooped up indiscriminately by the robotic probes. We understand the geologic makeup and structure of each Apollo site in much greater detail than those of the Luna sites.

For a more recent example, consider the Mars Pathfinder mission, which was widely touted as a major success. Although Pathfinder discovered an unusual, silica-rich type of rock, because of the probe's limitations we do not know whether this composition represents an igneous rock, an impact breccia or a sedimentary rock. Each mode of origin would have a widely different implication about the history of Mars. Because the geologic context of the sample is unknown, the discovery has negligible scientific value. A trained geologist could have made a field identification of the rock in a few minutes, giving context to the subsequent chemical analyses and making the scientific return substantially greater.