The postulation of a complex decision system (i.e., with multiple subparts and subprocesses) for explaining addictive and other apparently irrational behaviors has some precedents. In most previous models, however, decision systems were almost invariably modeled by two subsystems in conflict with each other. For instance, Thaler and Shefrin (Reference Thaler and Shefrin1981) modeled “the individual as an organization” with two coexisting selves, a “myopic doer” and a “farsighted planner.” (For a recent neural instantiation of this two-self model, see McClure et al. Reference McClure, Laibson, Loewenstein and Cohen2004.) In these dual decisional models, addiction arose when the myopic doer won the competition against the planner for behavioral control, thereby defining only one single pathway to addiction (i.e., from future-oriented, goal-directed actions to blind habits). What is original in Redish et al.'s unified framework for addiction (UFA) is that there is no winner, no loser. Each decision subsystem has its own specific potential failure points or vulnerabilities, each leading to a specific addiction symptom when accessed by a drug or behavior. Multiple simultaneous or sequential interactions among these vulnerabilities would then generate a wide diversity of possible addiction syndromes and/or transitions. Thus, according to UFA, addiction should be a spectrum disorder whose phenotypic diversity reflects the complexity of the mammalian brain decision system. This new link between brain decision system complexity and addiction diversity has several interesting evolutionary and developmental ramifications. Most of them, however, are almost completely ignored in the target article. In this commentary, I highlight those with special relevance for future research in the field.

First, the divergent evolution of the complexity of mammalian brain decision systems implies that the diversity of possible addictions should vary between different species, in general, and between nonhuman mammals and humans, in particular. Interestingly, the component of the brain decision system with the highest number of potential failure points – the planning subsystem – is also the component that has diverged the most in mammalian evolution. Planning for the far future – which depends on an intact and fully developed prefrontal cortex – has long been, and still is, thought to set humans apart from other animals, including mammals (e.g., Baumeister & Heatherton Reference Baumeister and Heatherton1996; Searle Reference Searle2001; Suddendorf & Corballis Reference Suddendorf and Corballis2007). Although more research is needed to better define the planning abilities of nonhuman mammals, current evidence nevertheless clearly indicates that they are “stuck in time” (Roberts Reference Roberts2002). For instance, rats or mice – the most frequently used animal models for human addictions – do plan for the immediate future in a goal-directed manner (Dickinson & Balleine Reference Dickinson and Balleine1994), but they are incapable of generating and maintaining long-term goals. The profound divergence in planning between nonhuman mammals and humans should somehow constrain the “addiction space” that the former could explore compared to the latter. To take an extreme example, imagine an agent endowed with a situation-recognition subsystem and a habit subsystem but no planning subsystem (and thus no higher-order subsystem that coordinates the activity of the whole). According to UFA, this myopic agent should be exposed to only three decisional vulnerabilities (i.e., Vulnerabilities 6, 7, and 10) and, therefore, should be affected by a restricted subset of addictions (i.e., habit-like addiction). Therefore, a critical goal for future research using UFA will be to identify among the actual spectrum of addiction syndromes affecting modern humans the syndromes that are unique to humans and those that are common, ceteris paribus, with other animals with less complex decision systems. This research has obvious implications for the current status and future development of animal models of human addictions and, in fine, for the neurobiology of addictions, which is still largely based on animal research.

Second, as the brain decision system evolves between species, it also develops within a single individual. In view of its complexity, however, it is highly unlikely that all of the different subparts of the brain decision system will depend on the same developmental programs and thus will develop at the same pace. For instance, the planning subsystem of humans develops more slowly than the habit subsystem (e.g., Blakemore Reference Blakemore2008; Ernst et al. Reference Ernst, Pine and Hardin2006). Humans until late in adolescence are shortsighted planners who depend heavily on parental and/or societal foresight. The differential development of each subpart of the brain decision system implies that not all potential decisional failure points will be accessible during the same developmental stage. As a result, the age of onset of drug use is expected to affect the diversity of addiction syndromes and/or transitions that can affect a developing individual. An important challenge for future research will be to determine how the addiction spectrum varies with the age of onset of drug use and how this variation is correlated with the differential development of each subpart of the brain decision system. Alternatively, drug use may also interfere with the proper development of the brain decision system. Thus, drugs of abuse could not only distort or subvert the normal function of a fully developed decision substructure, as conceptualized by UFA, but also retard or even accelerate its development per se. This would change the number of decisional vulnerabilities accessible at each developmental stage and, therefore, the spectrum of addiction pathways. Future research is clearly needed to add developmental dynamics into UFA.

Finally, I would like to stress that although UFA is no doubt an important step toward a unified neurobiological theory of addiction with interesting evolutionary and developmental ramifications, it still falls short of explaining choices between different kinds of rewards that are inherent to drug use and addiction. A recent series of controlled choice experiments from our laboratory provides a particularly vivid illustration of this important limitation (Lenoir et al. Reference Lenoir, Serre, Cantin and Ahmed2007). Rats with a long history of intravenous cocaine self-administration were allowed to choose between two different actions: one action was rewarded by intravenous cocaine, the other by a brief access to water sweetened with saccharine or sucrose. Previous research has established that prolonged cocaine self-administration accesses most of the decisional vulnerabilities identified in UFA, including vulnerabilities 2 (Ahmed et al. Reference Ahmed, Kenny, Koob and Markou2002; Reference Ahmed, Lutjens, van der Stap, Lekic, Romano-Spica, Morales, Koob, Repunte-Canonigo and Sanna2005), 4 (Paterson & Markou Reference Paterson and Markou2003), 5 (Ahmed & Cador Reference Ahmed and Cador2006; Mantsch et al. Reference Mantsch, Yuferov, Mathieu-Kia, Ho and Kreek2004), and 7 (Ahmed et al. Reference Ahmed, Lin, Koob and Parsons2003). Thus, according to UFA, with repeated choice, most chronically cocaine-exposed rats should rapidly develop a strong preference for the cocaine-associated action. Contrary to this prediction, however, we consistently found that virtually all rats preferred the sweet sensation over the artificial sensation of cocaine, a preference that was not surmountable by increasing cocaine doses (Lenoir et al. Reference Lenoir, Serre, Cantin and Ahmed2007). The discovery that taste sweetness surpasses cocaine reward, even in cocaine-sensitized rats with a long history of cocaine self-administration, represents a serious anomaly, not only for UFA but also for each of the separate neurobiological theories of addiction that it has unified. This anomaly does not necessarily invalidate UFA, however; it may indicate that the decision-making models on which UFA is based are largely incomplete. Specifically, these models were essentially built from data obtained in choice studies involving different dimensions of the same kind of reward (e.g., delay, amount, and probability). They are therefore not well adapted for predicting the outcome of choice between rewards as different in kind as a drug reward and a taste reward. Alternatively, this anomaly could also indicate that, as suggested earlier in the context of evolution, rats cannot constitutively develop all the addiction syndromes that may affect humans. More research on choices between drug and non-drug rewards is required here to resolve this anomaly.