Newell & Shanks (N&S) argue that there is little reason to think that unconscious thought plays a significant role in decision making. But they cast their net of evidence too narrowly. In particular, we worry that the research they marshal in support of their claim is either (a) consistent with competing interpretations that are supported by other lines of research or (b) fails to show that all elements that go into reaching a decision are conscious. While the data that N&S cite show that we have more knowledge of the factors that underwrite our decisions than was previously thought, they do not show that this knowledge informs our decisions or is conscious at the time the decision is made.
To begin, many of the studies that N&S cite demonstrate that, at best, people have post facto knowledge of the factors that figure in their decision making, not that those factors are conscious during the decision-making process itself. Consider, for example, N&S's discussion of studies of expert decision making that purport to find a discrepancy between explicit and implicit policies, or between explicit and implicit knowledge of the environmental cues that influence one's decisions. Previous studies had asked people to estimate their reliance on each cue, finding little correlation between those judgments and actual decision-making practice, thus suggesting that their reliance on those cues was unconscious. But if people are presented, instead, with a variety of sets of cue weightings and are asked to select which most closely resembles the strategy they use, then they prove quite accurate. N&S take this to show that people follow a conscious decision strategy. However, that one has a capacity to recognize one's policy does not begin to show that one consciously employs that policy. For the display may trigger an implicit memory of the previously deployed, but an unconscious decision strategy, which then primes the person's selection. One reason to prefer this explanation is that it is a familiar finding in the perception literature that masked stimuli produce implicit memories, which can then modify behavior in future tasks (Leuthold & Kopp Reference Leuthold and Kopp1998; Schacter Reference Schacter1992).
Of course, we grant that this hardly settles the matter and that further experimentation is required. In agreement with N&S's “immediacy” criterion, we suggest that what is needed in this type of case is introspection sampling, where people are cued at irregular intervals to report the contents of their conscious awareness (Hurlburt & Akhter Reference Hurlburt and Akhter2006). Post facto measurements of conscious awareness not only run the risk of participants tapping into implicit memories or accessing knowledge that played no active part in the decision-making process, but they also run the risk of participants forgetting what they had in fact been conscious of.
Similar points hold in connection with N&S's discussion of the Iowa Gambling Task. They point out that when Maia and McClelland (Reference Maia and McClelland2004) use a more explicit and less open-ended set of probe questions than had previously been employed, people show awareness of which decks are the good ones, and of the approximate long-term payoffs of the decks, as soon as they start to choose appropriately. N&S take this to show that it is conscious knowledge of payoffs that drives people's choices. Yet it may be (as Maia & McClelland themselves note) that in their online selections people rely on their affective responses toward the various options, without conscious judgments of relative goodness or approximate payoffs playing any causal role in the process. Indeed, when asked, people may base their judgments of goodness on their concurrent affective reaction while they contemplate making a selection from each deck. Moreover, they may either be able to reconstruct a rough estimate of the payoffs of each deck from memory, or else they may have constructed such a model in an ongoing manner following each trial.
Not only is this alternative construal of the data possible, but it is preferable. For it can smoothly accommodate the findings from patients with damage to ventromedial prefrontal cortex (VMPFC) (which is widely accepted to be the primary site of cortical representations of affect; Rolls Reference Rolls1999). Such patients continue to make bad choices in the Iowa Gambling Task, despite having good knowledge of the expected payoffs from each deck (Bechara et al. Reference Bechara, Damasio and Damasio2000). This finding makes good sense if people base their selections on their affective responses (which are absent or attenuated in VMPFC patients) while at the same time building a conceptual model of the task contingencies. To accommodate these data N&S will need to claim that in the normal case it is conscious judgments of comparative goodness that drive one's affective reactions. We know, however, that affective responses occur quite swiftly. It is implausible to claim that in every case they are preceded by conscious conceptual appraisals of the situation.
Even if we set aside these concerns about N&S's treatment of the data, however, they will need to posit some mechanism that can maintain an approximate running total of the net winnings from each deck. Otherwise the judgments of comparative goodness and likely payoffs from each deck that they appeal to will appear magical. But it is quite implausible that participants are aware of calculating these approximate running totals during the gambling task, in the way that one might be aware of calculating the tip on a restaurant check. Much more likely is that participants rely on an implicit system that operates unconsciously, perhaps along the lines of Dehaene's (Reference Dehaene1997) approximate number system.
In conclusion, N&S demonstrate, at best, that in normal instances of decision making some components of the decision-making process are conscious. This is not surprising, given that attentional resources are likely directed toward the task at hand, and given the connections between attention, working memory, and consciousness (Engle Reference Engle2002; Knudsen Reference Knudsen2007; Prinz Reference Prinz2012). But it is now well established that working memory is a very limited channel (Cowan Reference Cowan2000). It therefore beggars belief that all of the factors that influence decision making could be conscious, if decision making occurs in real time.
Newell & Shanks (N&S) argue that there is little reason to think that unconscious thought plays a significant role in decision making. But they cast their net of evidence too narrowly. In particular, we worry that the research they marshal in support of their claim is either (a) consistent with competing interpretations that are supported by other lines of research or (b) fails to show that all elements that go into reaching a decision are conscious. While the data that N&S cite show that we have more knowledge of the factors that underwrite our decisions than was previously thought, they do not show that this knowledge informs our decisions or is conscious at the time the decision is made.
To begin, many of the studies that N&S cite demonstrate that, at best, people have post facto knowledge of the factors that figure in their decision making, not that those factors are conscious during the decision-making process itself. Consider, for example, N&S's discussion of studies of expert decision making that purport to find a discrepancy between explicit and implicit policies, or between explicit and implicit knowledge of the environmental cues that influence one's decisions. Previous studies had asked people to estimate their reliance on each cue, finding little correlation between those judgments and actual decision-making practice, thus suggesting that their reliance on those cues was unconscious. But if people are presented, instead, with a variety of sets of cue weightings and are asked to select which most closely resembles the strategy they use, then they prove quite accurate. N&S take this to show that people follow a conscious decision strategy. However, that one has a capacity to recognize one's policy does not begin to show that one consciously employs that policy. For the display may trigger an implicit memory of the previously deployed, but an unconscious decision strategy, which then primes the person's selection. One reason to prefer this explanation is that it is a familiar finding in the perception literature that masked stimuli produce implicit memories, which can then modify behavior in future tasks (Leuthold & Kopp Reference Leuthold and Kopp1998; Schacter Reference Schacter1992).
Of course, we grant that this hardly settles the matter and that further experimentation is required. In agreement with N&S's “immediacy” criterion, we suggest that what is needed in this type of case is introspection sampling, where people are cued at irregular intervals to report the contents of their conscious awareness (Hurlburt & Akhter Reference Hurlburt and Akhter2006). Post facto measurements of conscious awareness not only run the risk of participants tapping into implicit memories or accessing knowledge that played no active part in the decision-making process, but they also run the risk of participants forgetting what they had in fact been conscious of.
Similar points hold in connection with N&S's discussion of the Iowa Gambling Task. They point out that when Maia and McClelland (Reference Maia and McClelland2004) use a more explicit and less open-ended set of probe questions than had previously been employed, people show awareness of which decks are the good ones, and of the approximate long-term payoffs of the decks, as soon as they start to choose appropriately. N&S take this to show that it is conscious knowledge of payoffs that drives people's choices. Yet it may be (as Maia & McClelland themselves note) that in their online selections people rely on their affective responses toward the various options, without conscious judgments of relative goodness or approximate payoffs playing any causal role in the process. Indeed, when asked, people may base their judgments of goodness on their concurrent affective reaction while they contemplate making a selection from each deck. Moreover, they may either be able to reconstruct a rough estimate of the payoffs of each deck from memory, or else they may have constructed such a model in an ongoing manner following each trial.
Not only is this alternative construal of the data possible, but it is preferable. For it can smoothly accommodate the findings from patients with damage to ventromedial prefrontal cortex (VMPFC) (which is widely accepted to be the primary site of cortical representations of affect; Rolls Reference Rolls1999). Such patients continue to make bad choices in the Iowa Gambling Task, despite having good knowledge of the expected payoffs from each deck (Bechara et al. Reference Bechara, Damasio and Damasio2000). This finding makes good sense if people base their selections on their affective responses (which are absent or attenuated in VMPFC patients) while at the same time building a conceptual model of the task contingencies. To accommodate these data N&S will need to claim that in the normal case it is conscious judgments of comparative goodness that drive one's affective reactions. We know, however, that affective responses occur quite swiftly. It is implausible to claim that in every case they are preceded by conscious conceptual appraisals of the situation.
Even if we set aside these concerns about N&S's treatment of the data, however, they will need to posit some mechanism that can maintain an approximate running total of the net winnings from each deck. Otherwise the judgments of comparative goodness and likely payoffs from each deck that they appeal to will appear magical. But it is quite implausible that participants are aware of calculating these approximate running totals during the gambling task, in the way that one might be aware of calculating the tip on a restaurant check. Much more likely is that participants rely on an implicit system that operates unconsciously, perhaps along the lines of Dehaene's (Reference Dehaene1997) approximate number system.
In conclusion, N&S demonstrate, at best, that in normal instances of decision making some components of the decision-making process are conscious. This is not surprising, given that attentional resources are likely directed toward the task at hand, and given the connections between attention, working memory, and consciousness (Engle Reference Engle2002; Knudsen Reference Knudsen2007; Prinz Reference Prinz2012). But it is now well established that working memory is a very limited channel (Cowan Reference Cowan2000). It therefore beggars belief that all of the factors that influence decision making could be conscious, if decision making occurs in real time.