Decoding free decisions and intentions
– by Robin van Bruggen
University of Amsterdam Studentnumber: 10001456 Date: 14-07-2014 Supervisor: Dr. Tobias H. Donner Project: Bachelorthesis Number of words: 5575
Table of contents
Abstract p. 3
Introduction p. 4
The neural substrates of unconscious, free decisions and intentions p. 6
The neural substrates of value based decisions p. 8
The neural substrates of delayed intentions p. 10
Discussion p. 12
Abstract
Libet pointed in his classic EEG-experiment to the SMA as the origin site of decision making, even prior to consciousness. But it's unlikely that an higher mental process as decision making starts in a motoric area. With the use of fMRI in combination with multi voxel pattern analysis, brain areas of importance in the decision making process could be uncovered. While important areas related to decision making were found and even some possible double associations, it remains unclear how these areas exactly work and connect related to the decision making process.
Introduction
Everyday we are constantly making decisions or having intentions to perform an action. Often even without really thinking about it. These are vital human processes and it is therefore important for academic circles to get a grasp of how these processes work in the brain. It can also help us gain more insight in a clinical population with brain damage to areas involved in these processes. When it is possible to localize important areas, they can be spared in brain surgery for instance. There are decisions that have to be made quickly and are therefore called 'forced decisions', such as deciding to drive through an orange light or to hit the brakes in traffic. Others do not have a time limit, and can be made more freely, such as whether to have pasta or rice for dinner. The focus of this paper will be on the latter, so called 'free decisions' (Soon, He, Bode and Haynes, 2013), since they seem to be less instinctive and therefor seem to involve more higher mental processes. Ethical considerations must be kept in mind in this line of research, since knowing what a person is going to decide or intends to do involving sensitive or private matters, can become a violation of privacy.
There is a thin line between decisions and intentions and they are often interlocked. Whereas a decision has to do with the choice of what action to perform, an intention contains the plan of performing that action and keeping that in mind. As such, they can be seen as subsequent subprocesses and will be treated as such in this paper.
Libet showed in his experiment in 1985 that our voluntary actions are initiated by unconscious mental processes long before we consciously decide to act; there is a readiness potential in supplementary motor area (SMA). The readiness potential is a low negative shift in EEG activity (Bode, Soon, Trampel, Turner and Haynes, 2011). The Libet experiment is
controversial for three reasons. First of all because the focus of the this experiment was only on motor-related regions. It is unclear if – and even unlikely that – decision making starts in SMA, since it's a motor planning area (Soon, Brass, Heinzle and Haynes, 2008). It would be more likely that decision making starts in areas involved in high-level planning, such as the prefrontal cortex. Second, it is unsure if the activation is not biased, since it is only a few hundred milliseconds prior to the conscious decision. There is a possibility that the activation seemed to be prior to decision making, while it was actually an artifact in measurement. Finally, it is not sure if the activation is specific, since the Libet experiment used the readiness potential, of which the actual choice cannot be predicted. Since the Libet experiment, researchers have often used the lateralised readiness potential (LRP), where the activity of the contralateral cortex is measured as well and subtracted from the ipsilateral cortex. Choice specific information can be predicted in that way and the LRP is
therefore more specific (Haggard and Eimer, 1999).
Libet conducted his experiment in 1985 with EEG, but since then new techniques have emerged. Nowadays fMRI has largely replaced EEG, since it has a far better spatial resolution and the whole brain can be mapped. It should be noted here that fMRI is an indirect measurement of neural activity, so one should be careful with inferring causal relationships and prediction is far from perfect (Bode, Soon, Trampel, Turner and Haynes, 2011).
Recently a new promising technique has emerged: multi voxel pattern analysis
(MVPA; also referred to as multivariate decoding). Conventional mass univariate analysis measures the mere overall activation of brain regions. The fine-grained patterns are often smoothed and important information about spatial patterning and correlation between voxels is lost (Kriegeskorte and Bandettini, 2007). It only tells us what regions are involved in decision-making processes, but not in what specific choice a person is going to make (Haynes, 2011). MVPA is more sensitive and an information based approach, instead of an activation based one. It takes the spatial patterning across the cortex into account and uses multiple classification algorithms to obtain a classification accuracy to tell how well a cognitive state can be predicted (Heinzle et al., 2012). The predictive information found in the spatial patterns can predict the specific content of someone his thoughts. Often the searchlight technique is used, whereby the whole brain can be mapped by using local patterns of brain activity, instead of focusing on regions of interests. (Heinzle et al., 2012). All the experiments that will be discussed in this paper, used a combination of fMRI and MVPA, and most of them used the searchlight technique.
The main question of this this paper what brain areas are involved in holding
intentions and making decisions and how these areas work as a network. Libet pointed to the SMA as the origin site of decision making, but it is unlikely that decision making starts in a motor
planning area. The prefrontal cortex is an important area for higher mental and cognitive functions, such as planning and working memory. Holding intentions and making decisions seem to be higher mental processes as well, so it would be likely that there is a important role for the prefrontal cortex. The focus will be on how different regions relate to different subprocesses of decision making. Since decision making often is not an easy process, the focus will be extended to decisions and intentions that are more complex and have an additional higher mental process that has to be
considered or remembered. In the first case this is the consequences of the decision, in the second it is holding an intention to perform an action at a later time.
In the first section the neural substrates of unconscious, free decisions and intentions will be discussed. Studies done since the Libet experiment using fMRI with MVPA will be
additional higher mental process, will be reviewed. The focus will be on value-based decision making in the second section, and on delayed decisions in the third section. Finally, these findings will be put together and the decision process and its elementary operations and how and where they are localized in the cortex, will be analyzed.
The neural substrates of unconscious, free decisions and intentions
As mentioned in the introduction, it is unlikely that the SMA is brain area where the
process of decision making starts, since the SMA is a motor planning area and not all decisions are motoric. Soon, Brass, Heinze and Haynes (2008) replicated the original experiment, but instead of EEG, they used fMRI with MVPA and the searchlight approach. They used the original task in which subjects had to press a left or right button. A letterstream was used to indicate when subjects had the intention to press the button instead of a clock in the original experiment, as a more accurate measurement. Choice-specific information in Brodmann Area 10 (BA10) was found 7 seconds prior to motor decision, with a prediction accuracy of 60%. Including the sluggishness of the BOLD response, this went up to even 10s. There was choice-specific information in parietal cortex as well. The authors mention a possible double dissociation: ''One interpretation of this finding is that frontopolar cortex was the first cortical stage at which the actual decision was made, whereas precuneus was involved in storage of the decision until it reached awareness''. This study shows that BA10 instead of the SMA seems to be the origin site of decision making.
Bode et al. (2011) replicated the study from Soon et al. (2008) with ultra-high field fMRI at 7 tesla and only used images of anterior frontopolar cortex. With these improvements distortion effects were minimized and the temporal stability could be investigated, which was not addressed in the study from Soon et al. (2008). In addition to the method used the Soon et al. (2008) study, subjects were asked about their behavior and thoughts with a questionnaire to investigate possible biases. The findings of the original study were replicated, activity began approximately 7 seconds before consciously making a decision in BA 10. The prediction accuracy was 57%. They also found that patterns became more stable closer to a conscious decision. Although patterns seemed to remain stable after the decision is made, there was no decodable information anymore. It is hypothesized that once a pattern is stable enough, a threshold is reached, and a conscious decision is made. Note that only images of the frontopolar cortex were made, so the extend of the decision-related region may be underestimated. The prenucleus/posterior cingulate cortex was not covered either, while Soon, Brass, Heinze and Haynes (2008) found that this region is of importance in decision making.
the forming of a decision. But the formerly discussed studies are all concerned with motoric decisions; subjects had to decide to press a button. So it remains unclear whether the frontopolar cortex holds the intention to make a decision in general, or only is concerned with the preparation for a motoric decision. Also the focus of these experiments was mainly on the frontopolar cortex, especially in the study by Bode et al. (2011), where the focus was only on this region.
Haynes et al. (2007) conducted an experiment where subjects made free, non-motoric decisions; they had to add or subtract two numbers. They saw two numbers on the screen and after a variable delay, a screen with four number appeared: the correct answer for addiction, the correct answer for subtraction and two similar but wrong answers, which were rarely chosen. Of
importance is that the answers were randomized, so it was impossible to prepare a motor response. From the choice of answers, it was possible to infer the task that the subject had chosen per trial. Highest decoding accuracy (71%) was in the medial prefrontal cortex, but only during delay. A more superior and posterior region in the medial prefrontal cortex was informative during execution, but not during delay. There was a significant decoding level as well in the lateral prefrontal cortex during delay. By using MVPA, they showed that these activations were specific for the task. The medial and the lateral prefrontal cortex seem to encode intentions. More anterior regions encode intentions prior to execution, while more posterior regions encode during execution. While this experiment shows that the prefrontal cortex is of importance as well in more abstract and non-motoric intentions and decisions as well, it is not informative about unconscious processes.
Lages and Jaworska (2012) replicated the experiments by Soon et al. (2008) and Haynes et al. (2007) but without monitoring fMRI BOLD signals. Their main focus was on finding possible response biases. They were able to replicate the same prediction accuracy as the
experiment by Soon et al. (2008) and Haynes et al. (2007). They found that there are no carryover effects across trials in the experiment by Soon et al. (2008), but warn for response dependency in individuals, since the instruction to make a spontaneous decision is quite contradictory. It is therefor possible that they use a strategy of holding intentions to press a button, and report this when they actually decide to press.
In a recent experiment conducted by Soon, Bode and Haynes (2013) the
adding/subtracting paradigm was used with a letterstream added to investigate whether there is activity prior to making an abstract decision. They found activity in the medial frontopolar cortex and prenucleus/posterior cingulate region, with 59,5 and 59% accuracy. There was significant activity in the angular gyrus as well at 2-4 seconds post decision making. Due to hemodynamic delay, this was probably around the time of the making of the decision. They also found that the SMA is involved in motor responses and does not encode abstract intentions. The mentioned
cortical regions that do encode abstract intentions, are not involved in encoding motor responses. So there seems to be a double dissociation.
From these studies there is more evidence that the brain already starts preparing for an action, prior to the conscious decision to act. It seems that the SMA as mentioned by Libet, is not the origin site for abstract intentions, but is of importance for motor preparation for the execution of this intention. The prefrontal cortex seems to be of importance in the forming of intentions and decision making. The importance of BA10 defined in Soon, Brass, Heinze and Haynes (2008) where a motor decision was made, was found as well in an abstract mental operation by Soon, Bode and Haynes (2013). Since the latter dissociated motor responses from high-level intentions, there is evidence that BA 10 and the prenucleus/posterior cingulate region are involved in intentions and general voluntary decisions. This network is active during a delay period when the decision is consciously formed (Haynes et al., 2007), as well as during the unconscious formation of decisions (Soon et al., 2013). If conscious and unconscious processes have the same neural substrate or are different at finer scale, is not sure. It should be taken into account that the prediction accuracies of these studies although significant, were not very high.
The neural substrates of value-based decisions
Decisions are often times influenced by the reward value we expect to obtain from choosing different options (Kahnt et al., 2010). In the first section the focus was on decisions that were very free, but in daily life decisions often have consequences or rewards; there are reasons to choose on option and not the other. People with damage to the prefrontal cortex are often impulsive and have a bad consideration of possible rewards, as seen for example in the famous Iowa
Gambling Task. It is therefor likely that the prefrontal cortex plays a role in making decisions based on values. The main question of this section is how the rewards and consequences of our decisions are weighed and what areas in the prefrontal cortex are involved in these processes.
Hampton and O'Doherty (2006) showed that it is possible to apply the multivariate decoding paradigm with high accuracy on value-based decisions. In their experiment, subjects had to choose between to stimuli and received either monetary gain or loss. After a while the “correct” response became the “incorrect” one and vice versa, and the stimuli were located at random on left or right side of the screen. They decoded global and local signals from areas of interest defined in previous research and interactions between those areas. Global signals have a spatial scale larger than 8 mm, where local signals have a smaller spatial scale than 8 mm. Of the nine regions of interest the anterior cingulate cortex (ACC), medial prefrontal cortex and ventral striatum had a higher combined accuracy (67%) than each region alone. The dorsal anterior cingulate cortex seems
to contribute the most, and the insula and dorsolateral prefrontal cortex (dlPFC) had a high decoding accuracy, but not when combined. This study shows that MVPA is applicable to value-based decision making, but it is not clear how the areas with a significant decoding accuracy contribute to value-based decision making. A flaw of this study is that they a priori decided what regions of interest were and did not scan the whole brain. So possible important areas may be uncovered in this study.
Kahnt, Heinzle, Park and Haynes (2010) point to a different region, namely the orbitofrontal cortex (OFC) as a main region for value-based decision making. They used a more sophisticated task and the searchlight technique on the fMRI data, which makes it possible to map the whole brain. Subjects performed a task in which they saw rotating dots. Rotation ranged from 100% clockwise to 100% counter clockwise and color ranged from 100% red to 100% green. Specific combinations were predictive of reward value. Subjects had to report either the rotation direction or color and the value of the cue was given as a reward. Several brain regions had a significant decoding accuracy independent of sensory properties; medial OFC, ventral striatum, the dorsomedial and dorsolateral prefrontal cortex, the frontopolar cortex, the precuneus, the lateral posterior parietal cortex, the amygdala and hippocampus. But only the OFC did not follow the sensory properties of the cues. Furthermore, patterns in the medial OFC are similiar during anticipation and outcome. This is an indication that anticipated reward values are used to guide behavior.
The studies discussed so far, focused on quite simple decision making, where participants could either win or lose. But often rewards than can be obtained are different and sometimes even conflicting. More attributes have to be considered, such as health versus taste. Each attribute of an object contains information about the reward, and whereas one object can have the same combined value as another, they can have a different variability (for example, high/low versus intermediate). Kahnt, Heinzle, Park and Haynes (2011) used objects that vary in even three visual attributes (shape, color and coherence of moving dots). They found that the combined values of the multi-attribute objects is represented in the ventromedial prefrontal cortex (vmPFC) and the
variability of the reward predictions of the individual attributes are represented in the dlPFC. These results suggest that the dlPFC signals ambiguity and may be involved in integrating value
predictions into a combined value. The vmPFC represents the combined value and can directly guide choices.
An important area left uncovered so far in reward-based decision making research is that associations between decisions and their outcomes are not formed spontaneously: they have to be learned. The subsequent question becomes how these associations are learned and what areas are
involved. Kahnt, Heinzle, Park and Haynes (2011b) had subjects learn six cue-outcome associations. The predicted reward presentations were probed using a free-choice task.
Representations appeared earlier in the striatum than in the dlPFC and OFC, so there are different time courses of learning and possibly multiple reward signals in the brain. They showed that value-coding fMRI patterns during the reward-predicting cue and rewarding outcome become more similar during learning. Kahnt, Heinzle, Park and Haynes (2010) already suggested that anticipated reward values guide behavior. Here this finding is extended by showing that the pattern similarity found there as well, is only present after learning. Only the reward representations dlPFC were related to performance in the subsequent decisions, suggesting that the dlPFC has a role in reward based action selection.
From these studies it seems that the prefrontal cortex plays a great role in value-based decision making. Whereas Hampton and O'Doherty (2006) mention the ACC, mPFC and striatum as a possible important network for reward-based decision making based on several regions of interest, Kahnt, Heinzle, Park and Haynes (2010), with a more sophisticated technique found that the OFC seems to play an important role in anticipating rewards to guide behavior. When choices have different or possible conflicting rewards, the dlPFC seems to be of importance in signaling this ambiguity and may have a role in reward based action selection, while the vmPFC represents the combined values of these rewards and can guide choices. There is evidence that the association between decisions and their rewards when learned, is used to guide behavior. The brain might use the neural substrates of an actual outcome of a reward to represent the predictions. It also seems that there are different reward signals in the striatum than in the OFC and dlPFC, because the former had earlier activation that the latter. If these areas encode different subprocesses of value-based decision making is a question that can be investigated in future research.
The neural substrates of delayed intentions
We often have to remember to do something at a later time, meaning we have to hold the intention to perform an action. Often we have to hold this intention while engaging in other activities. Most of the time the decision is already made, but the decision that is made has to be remembered to perform the action at a later time. These intentions are therefor called “delayed intentions” (Mommennejad and Haynes, 2012). An example of this mentioned by Mommennejad and Haynes (2012) is having the intention to read at the moment, but also maintaining the intention to drain the pasta that is boiling at a later time. Since this is a higher mental process, it is
hypothesized that the prefrontal cortex plays a role in delayed intentions. The ability to hold an intention in mind to retrieve it at a later time is sometimes referred to as “prospective memory”(PM;
Brandimonte et al., 1996, mentioned in Gibert, 2011). In this section the focus will be on will be on how the PM works and what brain areas are involved.
Gilbert (2011) shows that the rostrolateral prefrontal cortex (rlPFC) plays a role in these delayed intentions. Participants performed a n-back task with encode-store-retrieve cycles embedded (e.g. “press button x when you encounter the stimulus again”). There was increased rlPFC activation while storing delayed intentions. The content of these intentions were represented in distributed posterior regions, but not the in the rlPFC itself. This supports a possible retrieval function for the rlPFC.
In the experiment done by Gilbert (2011) subjects got a cue when to perform the action, but delayed intentions do not always have a specifically set time. This is in line with the model proposed by Brass and Haggard (2008): the 'what, when, whether model' of intentional action. They state that there are at least three kinds of information generation: what to do, when to do it and whether to do it. Mommennejad and Haynes (2012) used a time-based experiment instead of an event-based experiment as Gilbert (2011) did. They investigated the representation of future intentions across a period in which subjects were busy performing another task. By using long durations (15-25s) they could distinguish between the encoding of intentions while being
maintained and while being retrieved. Subjects needed to internally track time. They found that the anterior prefrontal cortex (aPFC) is of importance in what subjects intend to do next and when they intend to do it, when occupied with another task. What subjects intended to do next could be decoded from the dorsomedial aPFC during the maintaining of an intention and in the ventrolateral aPFC during retrieval. When subjects intend to do this, could be decoded from the bilateral and medial aPFC during delay and from the medial aPFC and prenucleus during retrieval. This suggest a role for the aPFC in PM tasks with different components.
An important question that remains unanswered so far is whether it matters what kind of other activities are done while maintaining an intention, or put in other words: whether the encoding of intentions depends on the availability of resources during the delay period.
Mommenejad and Haynes (2013) addressed this question by varying the task load. In one condition the delay period was task-free and in another condition the delay was occupied. Delayed intentions were decoded from the rlPFC during both delays. Decoding was from the rostromedial PFC during occupied delays only, where the demand was high and from the ACC, SMA and prenucleus during task free delays, where the demand was low. There seems to be a double dissociation.
Especially, the prefrontal cortex seems to play an important role in delayed intentions. But where Gilbert (2011) found a role for the rlPFC in retrieval, Haynes et al. (2007) and
with the sort of task. Gilbert (2011) had a task where mostly visual stimuli were used, while Haynes et al. (2007) and Mommenejad and Haynes (2013) used a more abstract approach, where subjects had do to simple math with numbers or answer questions about them (e.g. odd or even). Another difference is that Gilbert (2011) used fixed button presses, while Haynes et al. (2007) and
Mommenejad and Haynes (2013) used more randomized responses. A possible interpretation is that when a task gets more abstract, more rostral regions recruited. Next to the rlPFC, the aPFC seems to play an important role in maintaining and retrieving delayed intentions as well, when not only has to be maintained what has to be done, but also internally has to be tracked when it has to be done. A vital question that is yet unanswered is what areas are involved when has to be decided if the
intention will be executed; the whether-component of the model proposed by Brass and Haggard (2008).
Discussion
The aim of this paper was to answer what the neural substrates are of decision making and holding and retrieving intentions. With the MVPA paradigm was found that especially the prefrontal cortex seems to be an important area. Different regions of the (pre)frontal cortex and some in the parietal cortex seem to play a role in different aspects of decision making. The initiation of making a decision seems to start in the frontopolar cortex and prenucleus/posterior cingulate region, even before someone is consciously deciding to act. Often times the decisions we make have certain consequences or rewards, when this is the case, the orbitofrontal cortex seems to be of crucial importance. There seem to be other areas involved as well, such as the striatum, but it is yet unclear how these areas exactly contribute. If rewards are in conflict, the dorsolateral prefrontal cortex seems to signal this ambiguity and the ventromedial prefrontal cortex guides decisions by combining the different values of these rewards. When the intention to perform a certain action has to be kept in mind, the (rostro)lateral prefrontal cortex seems to maintain and retrieve this intention. Another area that maintains and retrieves intentions is the anterior prefrontal cortex, which also seems to be important when also has to be kept in mind when to execute this action. But there are some mixed results in this line of research.
We can conclude that mainly the prefrontal cortex and part of the parietal cortex play a vital role in the decision making process, and that the frontopolar cortex seems to be the origin site of this process, instead of the SMA pointed out by Libet in 1985. There seem to be different areas contributing to different subprocesses of decision making, and they seem to be working as a network. While important areas related to decision making were found and even some possible double associations, it remains unclear how these areas exactly work and connect related to the
decision making process. The brain is an extremely complex organ and we are only beginning to understand how it works. In that line, recently developed techniques were discussed, so on one hand, this is not surprising. To get a better understanding of how the brain areas related to decision making work as a network, the use of high resolution fMRI or intercranial measurements could be useful.
It should be noted as well that although the combination of fMRI and the MVPA technique used in these studies is a good one, it is not a perfect one. With neuroimaging techniques, it is not possible to see the activity as it really is, but only through indirect measurements. In the case of fMRI, these are the changes hemoglobin levels of blood in the brain; the so called BOLD responses. A relevant issue is that although the decoding accuracies found in the studies discussed in this paper are often significant, they are not extremely high (often around 60%). This means that although there is a connection between signals and decisions or intentions, it is not a very tight one. As Haynes (2011) mentions, there should be a 100% percent decoding accuracy to possibly speak of causality. That the decoding accuracies here are not as high, can be caused by the spatial and temporal limitations of the fMRI technique or because the decoding technique needs a great amount of training samples to get the exact predictive pattern. It is possible that better methods allow for an accuracy that is closer to perfect, but there is a realistic possibility as well that a perfect, or maybe even, a significantly better prediction accuracy is not obtainable. To determine which of these options is the case, better techniques need to be developed. Still, even if a 100% decoding accuracy can be found, there is the possibility that the early activity is not decision related, but only
background activity (Haynes, 2011).
While two sections focused on decision making with an additional higher mental process were discussed, these experiments still are not remotely equal to decision making in daily life. This will of course always be an issue with experiments in laboratory settings, since they are artificial. But it seems a possibility for future research to keep decreasing this gap. This can be done by making the choices that participants have to make more realistic, for example by outlining a situation and to ask participants how they would react and what choice they would make, while measuring their brain activity. Another interesting question that is not yet answered, is if there is a difference in the quality of the decision making process between people. This is especially relevant to the field of value-based decision making, since every person has different rewards they value. It would be interesting to know how big these differences typically are and how they affect brain activity and areas involved. In the original studies the focus was on the unconscious preparation prior to the conscious decision, but this has not yet been done in studies focusing on delayed
This paper found that many different areas in the brain are vital for decision making. Additional research is needed to eliminate ambiguities and find if more regions are involved and how the decision making process exactly works as a network. Given that decision making is such a vital human process, understanding how these processes work in the brain, is of importance to academic circles, as well as to the clinical practice.
References
Bode, S., He, A. H., Soon, C. S., Trampel, R., Turner, R. & Haynes, J. D. (2011). Tracking the unconscious generation of free decisions using uItra-high field fMRI. PloS ONE, doi:
10.1371/journal.pone.0021612
Brass, M. & Haggerd, P. (2008). The what, when, whether model of intentional action. Neuroscientist, 14(4), 319-325.
Gilbert, S. J. (2011). Decoding the content of delayed intentions. The Journal of Neuroscience, 31(8), 2888-2894.
Haggerd, P. & Eimer, M. (1999). On the relation between brain potentials and the awareness of voluntary movements. Experimental Brain Research, 126, 128-133.
Hampton, A.N. and O'Doherty, J. P. (2006). Decoding the neural substrates of reward-related decision making with functional MRI. PNAS, 104(4), 1377-1382.
Haynes, J. D. (2011). Decoding and predicting intentions. Annals of the New York Acadamy of Sciences, 1224, 9-21.
Haynes, J. D., Sakai, K., Rees, G., Gilbert, S., Frith, C. & Passingham, R. E. (2007). Reading hidden intentions in the human brain. Current Biology, 17(4), 323-328.
Heinzle, T. et al. (2012). Multivariate decoding of fMRI data. E-Neuroform, 3(1), 1-16.
Kahnt, T., Heinzle, J., Park, S. Q. & Haynes, J. D. (2010). The neural code of reward anticipation in human orbitofrontal cortex. PNAS, 107(13), 6010-6015.
Kahnt, T., Heinzle, J., Park, S. Q. & Haynes, J. D. (2011). Decoding different roles for vmPFC and dlPFC in multi-attribute decision making. Neuroimage, 56(2), 709-715.
predictions across learning. The Journal of Neuroscience, 31(41), 14624-14630.
Kriegeskorte, N., Bandettini, P. (2007). Analyzing for information, not activation, to exploit high-resolution fMRI. Neuroimage, 38(4), 649-662.
Lages, M. & Jaworska, K. (2012). How predictable are “spontaneous decisions” and “hidden intentions”? Comparing classification results based on previous responses with multivariate pattern analysis of fMRI BOLD signals. Frontiers in Psychology, 56(3), 1-8.
Mommennejad, I. & Haynes, J. D. (2012). Human anterior prefrontal cortex encodes the „what‟ and „when‟ of future intentions. Neuroimage, 61(1), 139-148.
Mommennejad, I. & Haynes, J. D. (2013). Encoding of prospective tasks in the human prefrontal cortex under varying task loads. Journal of Neuroscience, 33(44), 17342-17349.
Soon, C. S., Brass, M., Heinze, H. L. & Haynes, J. D. (2008). Unconscious determinants of decisions in the human brain. Nature neuroscience, 11, 543-545.
Soon, C. S., He, A. H., Bode, S. & Haynes, J. D. (2013). Predicting free choices for abstract intentions. PNAS, 110(15), 6217-6222.
