Male swordtails steered nine generations of virtual females toward bigger bodies

The Exeter tanks had one screen at each end ^[1], so every trial was still a choice between two options ^[6]. The range comes from the step between trials. Winning virtual fish were paired with winners from other trials, and their offspring became the next round's options ^[3]. Over nine generations ^[4], a run of two-way choices can move a stimulus a long way from where it started.

"Studying mate choices in animals is difficult because experiments can only present animals with a limited range of potential mates," said Kathryn Bullough of the Centre for Ecology and Conservation at Exeter's Penryn Campus ^[12]. "Our new approach allowed fish to 'design' their ideal partner, with genetic algorithms creating new sets of options for fish to choose from, allowing them to refine their preferences," Bullough said ^[13].

The choice of species was part of the design. "Green swordtail fish are good test subjects for this experiment as they respond to videos of other fish, and they provide no parental care for their young. This means their mate choice is relatively straightforward, as fish are primarily looking for good genes to pass on to their offspring," said Laura Kelley ^[14].

The headline figure is that virtual females were more than 24% larger on average by the final generation ^[9]. That number describes the population of stimuli. It does not directly measure how much a male prefers size. How far a genetic algorithm travels in nine rounds ^[4] depends on how it builds offspring from winners, as well as on who does the choosing. The account's explanation, that larger females can produce more offspring, comes with a "probably" ^[10]. The trials themselves recorded only where each fish swam ^[7].

A fish choosing at random would still crown a winner every round, and the virtual population would still drift. The thing this doesn't tell you is how much of the 24% exceeds that drift. The phys.org account does not report statistical tests or a random-choice baseline. The word "consistently" in the males' result ^[5] is the best sign in the summary that the shift is real. If each of the 12 males ^[6] drove his own lineage, a shared direction would be hard to get by chance.

One female's virtual mates grew steadily more female-looking. The team's tentative explanation is that she avoided threatening male characteristics, moving away from them and so toward female traits ^[11]. That idea rests on a single fish out of 11 females tested ^[6].

Kelley sees wider use. "A study like this using real fish would take years, and our new method provides a powerful tool to understand how wild animals choose their mates," Kelley said ^[15]. The evidence so far is 23 captive fish ^[1] reacting to moving images in a tank ^[6]. Whether males in the wild in Central America, where the species is native ^[16], favor females as large as the ones this algorithm bred is a separate experiment.