Bentley et al. discuss how technological advances, particularly online connectivity, change human decision making and collective behavior and their empirical study. They usefully caution that although large-scale online data provide new opportunities for the understanding of human behavior, there are also new pitfalls. Bentley et al. present a deliberately simplified conceptual map of different types of human decision making, with two dimensions: the degree to which personal versus social influences shape a decision, and the transparency of the payoffs of a decision. We agree that these are important determinants of decision making, and the two dimensions provide a useful simplification for several applications. However, we would like to draw attention to other determinants of decision making, particularly the influence of past experience, individual differences, and payoff structure, all of which may affect large-scale patterns.

The weighting of personal experience (“individual information”) against information provided from others (“social information”) is a key determinant of human decision making, and numerous factors can determine this weighting, such as the predictability of the environment, the relative costs of social and individual information, or the availability of suitable models to learn from (Boyd & Richerson Reference Boyd and Richerson1985; Laland Reference Laland2004). Frequently, individual and social information will together determine a decision (Kendal et al. Reference Kendal, Coolen, Laland, Dukas and Ratcliffe2009; Salganik et al. Reference Salganik, Dodds and Watts2006). However, adaptive decisions can be made without gathering information (Dall & Johnstone Reference Dall and Johnstone2002; Stephens Reference Stephens1991), and thus the intensity of information use in a decision may predominate over whether individual versus social information is utilized. Important decisions in particular are likely to involve substantial use of both individual and social information. Based on work on animal personalities (e.g., Marchetti & Drent Reference Marchetti and Drent2000), we suggest that the amount of information gathered for decision making may consistently differ between individuals, with certain individuals more likely to utilize both individual and social information. That is, we predict individual and group differences in their sensitivity to information, in the readiness to search for information, as well as in the strategy used to obtain information.

Bentley et al. discuss several aspects of the transparency of decision payoffs, such as whether a right or wrong decision is immediately detectable, whether a payoff can be assessed before a decision is made, the equivalence of payoffs, the degree of understanding of the processes that determine payoffs, and the ease of assessing who to learn from. Since these aspects of transparency may not covary, we urge caution in treating payoff transparency as a unitary entity. This is particularly important because payoff transparency, together with the costs of decision making, are likely to determine whether social or individual information is utilized: When the best decision can easily be determined individually, social information will be less advantageous (Boyd & Richerson Reference Boyd and Richerson1985; Kendal et al. Reference Kendal, Coolen, Laland, Dukas and Ratcliffe2009). Thus, the two dimensions of Bentley et al.’s conceptual map are not independent.

Payoff transparency may depend on numerous additional characteristics, such as the way that competition or frequency dependence affect payoffs (McNamara & Fawcett Reference McNamara, Fawcett, Todd, Hills and Robbins2012), the shape of the “adaptive landscape,” and thus the degree to which similar past decisions inform current decisions, and the kind of social information that models provide (Beppu & Griffiths Reference Beppu, Griffiths, Taatgen and van Rijn2009; Rendell et al. Reference Rendell, Fogarty, Hoppitt, Morgan, Webster and Laland2011). Assessment of payoff transparency is further complicated since a single decision may have multiple, possibly conflicting, payoffs in different domains. For example, depending on circumstances, deviating from group behavior may have economic benefits but social costs, or, alternatively, independent innovation may be socially rewarded but economically costly (Day et al. Reference Day, MacDonald, Brown, Laland and Reader2001; Greve Reference Greve2003). Bentley et al. argue that differences in agent competencies will not be visible at an aggregated scale. However, groups may differ in the way payoffs are assessed, potentially resulting in different processes in different groups. For example, children and adolescents differ in the way they assess losses and long-term rewards, compared to adults (Aïte et al. Reference Aïte, Cassotti, Rossi, Poirel, Lubin, Houdé and Moutier2012). Similarly, experts may outperform the “wisdom-of-the-crowd” depending on the task (Krause et al. Reference Krause, Ruxton and Krause2010), meaning that individual competencies may have significant influence. A striking example of expertise trumping the wisdom-of-the-crowd is the victory of top-ranked chess player Magnus Carlsen over a large online audience (referred to as “The World”) that participated by choosing one of three available moves proposed by a group of chess grandmasters (McClain Reference McClain2010).

We consider whether humans can switch freely between incorporating different kinds of information in their decisions. Each decision could be optimized by determining which information-gathering strategy to apply, which may itself rely on social cues (Toelch et al. Reference Toelch, Bruce, Meeus and Reader2011). However, the costs of assessing which strategy to employ may outweigh any achieved benefits, and prior knowledge and individual characteristics, perhaps in combination with evolved predispositions, could shape the strategy employed. Current, past, or early life experience can be used to determine which information-gathering strategy is most likely to be profitable. For instance, humans playing a computer game apparently used the degree of environmental variability as a heuristic cue for determining whether to copy others (Toelch et al. Reference Toelch, van Delft, Bruce, Donders, Meeus and Reader2009). In nonhuman animals, recent and early life-experience determines the reliance on social information (Chapman et al. Reference Chapman, Ward and Krause2008; Dawson et al. Reference Dawson, Avargues-Weber, Chittka and Leadbeater2013; Katsnelson et al. Reference Katsnelson, Motro, Feldman and Lotem2008; Lindeyer et al. Reference Lindeyer, Meaney and Reader2013). Thus, experience and development may influence future decision making, resulting in carryover effects where experience with one problem type or in one domain influences the decision making employed for other problems. Indeed, individuals may specialize in particular information-gathering strategies.

The opportunities and pitfalls provided by “big data” increasingly apply to other fields that have traditionally utilized relatively small datasets and direct behavioral observations, such as animal behavior and behavioral neuroscience (Sumpter et al. Reference Sumpter, Mann and Perna2012). It is now possible to track groups of free-living animals, their interactions, their physiology, and their neural activity for long periods (Cavagna et al. Reference Cavagna, Cimarelli, Giardina, Parisi, Santagati, Stefanini and Viale2010; Dell'Omo et al. Reference Dell'Omo, Ricceri, Wolfer, Poletaeva and Lipp2000; Krause et al. Reference Krause, Wilson and Croft2011; Vyssotski et al. Reference Vyssotski, Dell'Omo, Dell'Ariccia, Abramchuk, Serkov, Latanov, Loizzo, Wolfer and Lipp2009). While automated recording and testing provide unprecedented opportunities to monitor the behavior of multiple individuals, the use of empirical population-level signatures as indicators of learning processes must be validated (Reader Reference Reader2004), rare behavior patterns should not be overlooked (Krause et al. Reference Krause, Wilson and Croft2011), and, ideally, direct observations should ensure that the relevant behaviors are indeed recorded. For example, electronically tagged mice adopted a group-foraging strategy and thus circumvented an automated system programmed to reward particular individual foraging patterns, a development easily identified by video observations of behavior (Dell'Omo et al. Reference Dell'Omo, Ricceri, Wolfer, Poletaeva and Lipp2000). Large datasets cannot substitute for well-designed studies, but appropriate use of “big-data” methodologies provides novel and valuable tools for the study of human and nonhuman animal social behavior alike.