While the Earthlings are celebrating the one hundredth anniversary of the Normandy invasion, news from Mars casts a somber note. NASA’s initial attempts to integrate animals into its settlement on Mars have faced setbacks, for two chimpanzees and several poultry species failed to adapt to the harsh conditions and were all confirmed dead. The nursing robot tasked with their care was also unsuccessful in accounting for the unique negative impact of cosmic rays on the neural systems and DNA structures of animals, because it relied solely on Earth-derived big data. As a result, NASA’s ambitious project of the settlement on Mars now urgently requires a Plan B.

Fortunately, NASA has initiated a team to design Mars rovers that utilize technologies different from the current, big-data-based ones. This new project, named “Watney”—after the protagonist of the novel and its film adaptation The Martian—indicates that the cognitive architecture of this proposed rover would function in the resource-scarce and unpredictable Martian environment, where supporting information derived from Earth is of little help. Project supervisor Joseph Kim describes four priori constraints on designing “Watney”: (1) the uninhabitability of the Martian environment makes it impossible for human experts to accompany the rovers; (2) communication with Earth is mediated by the artificial orbiters near Mars, therefore limiting the transmission of large data sets necessary for big-data technologies; (3) the relatively limited data accumulated about Mars renders traditional big-data approaches impractical in the new scenario; (4) the existence of Martian dust and other contingent factors may compromise the proper functioning of silicon-based machinery. When these four constraints are taken into consideration, the designing approach that embraces “small dataism” and situational decision-making systems naturally reveals itself to be more tenable.

What is the purpose of designing Watney? As Kim explained, the first generation of Watney is intended to build underground Mars habitats to accommodate future settlers, many of whom may be carbon-based. The design philosophy of Watney’s hardware draws from the military concept of a “vehicle family,” which emphasizes shared components across different models that aim to reduce maintenance costs. In the case of Watney, this philosophy is implemented by standardizing components across diverse models that perform different jobs. This means that the specialization of every different Watney—whether tunneling, transporting, or reconnaissance—is partly “internalized” through a shared modular framework. This framework mirrors how humans draw on past experience to solve problems: instead of relying on domain-specific data sets, it enables adaptive decision-making by applying generalizable insights from previous exposures to new challenges. Hence, the need for permanent attachments, such as tunnel boring machines or transporters, can be replaced by devices temporarily installed on the robots. Additionally, despite similarities in their core algorithms, the Watney family robots can be highly diverse in appearance to further maximize their adaptive capacities. Some resemble androids, while others are modeled after animals like snakes, frogs, or centipedes. Moreover, each type of Watney can also transform between shapes to better handle different tasks. On a sidenote, designing a flying Mars rover poses some particular challenges due to the thin atmosphere on Mars, which makes traditional drones aerodynamically impractical.

However, despite the flying issue, the building of other forms of Watney family robots is still on the right track. Kim presented a donkey-like Watney rover, demonstrating that its artificial fur is sensitive enough to detect the approach of Martian dust storms, even when the storm’s eye is tens of miles away. He also explained that the older generations of Mars rovers, such as Opportunity and Curiosity, were vulnerable to rocks and holes due to their inflexible wheels. Instead, Kim’s team drew inspiration from donkeys, which carry supplies and traverse cliffs in mountainous regions with precision, thanks to their flexible hooves and balancing abilities. The smaller size of donkeys also makes them maneuverable in steep spaces.

Despite its donkey-like appearance, surprisingly, the Watney rover does not think like a donkey; instead, it thinks in a human-like manner. For example, when it perceives an incoming storm, it calculates the storm’s scale, assesses the risk of continuing its current task versus taking cover, and decides when to issue storm alerts to other robots. Additionally, the rover is capable of communicating with human riders by using standard English.

As to this rover’s reasoning module, it integrates fast-thinking and slow-thinking mechanisms described in Daniel Kahneman’s psychological theory, thereby enabling the system to deliver instinctive reactions when reaction time is insufficient and to make deliberative decisions in response to complex but nonurgent challenges (like how to distribute batteries among Watneys in an utilitarianist manner). While slow thinking could be based on big data sets, fast thinking is not. Therefore, a considerably large part of the rover’s reasoning module does not function by relying on big data, allowing the rover itself to survive in life-and-death situations where deep consideration is impossible due to time/information constraints.

Currently, conceptual studies for the second generation of Watney-type Martian rovers, known as “Watney-II,” are already underway. These rovers, new forms of embodiment notwithstanding, will inherit the experience of their predecessors on the software level. They will also acquire new knowledge related to the care for carbon-based life-forms on Mars. More specifically, as Kim explained, teaching Watney-II how to raise poultry on Mars would be a tenable start, as it would help the rover gradually understand the needs of human beings. On a sidenote, as Kim finally pointed out, Mars could provide a simpler and more suitable testing ground for developing terrestrially applicable artificial general intelligence (AGI), given that scenarios like home nursing pose great challenges to AGI due to the complexity of microsociological environments on Earth.

Jack Xu from The Mars Settlement Center, Houston