Spotlight Flexible Planning in Ravens? Jonathan Redshaw, 1,3 Alex H. Taylor, 2,3 and Thomas Suddendorf 1,3, * Across two different contexts, Kabadayi and Osvath found that ravens preferentially selected items that could be used to obtain future rewards. Do these results demand a rethink of the evolution of flexible planning, or are there leaner alternative explanations for the performance of ravens? Many complex and highly adaptive human behaviors can only be understood through our capacity to imagine, flexibly plan for, and actively shape future events. It has been proposed that this capacity is unrivaled by any other species and is a key adaptive strategy that enabled humans to fundamentally transform and control ecological niches across the planet [1,2]. Exactly how unique our powers of foresight are, however, is con- troversial [3]. Evidence for the presence of flexible planning in non-human animals would raise profound questions about the evolution of the capacity and what precisely is distinct about human foresight. In an apparently breakthrough study, Kabadayi and Osvath [4] recently reported that ravens (Corvus corax) can flexibly plan for future events in domains where they have no natural behavioral predispositions (tool use and bartering). If correct, this conclusion would indicate that the capacity must have evolved at least twice in distinct phylogenetic line- ages separated by 320 million years of evolutionary history. Over a series of four experiments (E1–E4), five ravens were given the opportunity to select a tool or token that could be used to obtain a future food reward. The birds selected the correct tool or token significantly more than distractor items in conditions in which the reward became available 15 minutes (E1) or 17 h (E2) later. They also selected the correct item significantly more often than a small food reward (E3– E4), suggesting that they could inhibit a current preference for food so as to obtain a greater amount of food in the future. Although the birds clearly preferred the appropriate items in these experiments, it is premature to conclude that ravens can flexibly plan. Before the first experi- ment the ravens learned how to use the functional tool and token to obtain food over five and 35 training trials, respec- tively, and they also had the opportunity to learn that distractor items were non- functional. All the ravens then selected the trained items on the first experimental trial, even though they had not experi- enced a predictable return of the reward situation and therefore had no reason to expect a specific future episode in which to use them. Thus, a lean alternative hypothesis is that the learned reward associations simply made the functional items more attractive than the distractors, given that the functional items had been associated with food more than the distractors. One could test associative explanations, for instance, by pairing distractors and functional tools equally with reward before test, or by visibly destroying the reward apparatus to check if ravens con- tinue to prefer the previously functional tool over the small food reward even when there is no future utility. One could examine flexibility by making the same objects functional in one future context and useless in others. Instead, through- out the studies the ravens were tested for selection of the same objects. The authors infer flexibility because the ravens selected the target object more when there was an immediate opportu- nity to use the items (E4) than when there was not (E3), but this may reflect carry- over effects because these experiments were not counterbalanced. That is, the birds could have learned to choose the functional object more often because of their experience in E3, leading to their increased performance in E4. This could have occurred in two ways: (i) memory- mediated reinforcement [5] during E3, where birds remembered their prior object choice when they were presented with the apparatus 15 minutes later, and paired this memory with the subsequent outcome (functional item leads to reward; small food leads to no reward), or (ii) the association between functional item and reward could have been stronger in E4 as a result of repeated pairing of reward and tool in E3, meaning that the functional object was more attractive in E4 than E3, and was therefore chosen more often. In fact, E3 was the second experiment in the series and occurred before E2 ([7_TD$DIFF][1_TD$DIFF]see supplementary materials of [4]), and the ravens therefore had even more opportu- nity to learn the association between functional item and reward before E4. One way to test if the behavior of the ravens was truly affected by temporal dis- tance would be to run E3 and E4 twice such that temporal distance does not change in a linear fashion that correlates with experience. Although the authors suggest that the capacities of ravens parallel those of great apes, it is important to acknowledge that the evidence for animal foresight remains very controversial [3]. Methodological criticisms [1,6] have been leveled at the positive findings they cite [7,8], and other studies have failed to find evidence of even elementary components of planning in great apes, such as preparation for two alternative event outcomes [9] or for two sequential and causally linked events [10]. Thus far animal studies have failed to match the conservative criteria used to demonstrate competence in children [11], namely the use of (i) single trials, to prevent consistent exposure to the same stimulus–reward relationships, and TICS 1719 No. of Pages 2 Trends in Cognitive Sciences, Month Year, Vol. xx, No. yy 1