Hasty sensorimotor decisions rely on an overlap of broad and selective changes in motor activity.

Humans and other animals are able to adjust their speed-accuracy trade-off (SAT) at will depending on the urge to act, favoring either cautious or hasty decision policies in different contexts. An emerging view is that SAT regulation relies on influences exerting broad changes on the motor system, tuning its activity up globally when hastiness is at premium. The present study aimed to test this hypothesis. A total of 50 participants performed a task involving choices between left and right index fingers, in which incorrect choices led either to a high or to a low penalty in 2 contexts, inciting them to emphasize either cautious or hasty policies. We applied transcranial magnetic stimulation (TMS) on multiple motor representations, eliciting motor-evoked potentials (MEPs) in 9 finger and leg muscles. MEP amplitudes allowed us to probe activity changes in the corresponding finger and leg representations, while participants were deliberating about which index to choose. Our data indicate that hastiness entails a broad amplification of motor activity, although this amplification was limited to the chosen side. On top of this effect, we identified a local suppression of motor activity, surrounding the chosen index representation. Hence, a decision policy favoring speed over accuracy appears to rely on overlapping processes producing a broad (but not global) amplification and a surround suppression of motor activity. The latter effect may help to increase the signal-to-noise ratio of the chosen representation, as supported by single-trial correlation analyses indicating a stronger differentiation of activity changes in finger representations in the hasty context.

Integrated neural dynamics of sensorimotor decisions and actions.

Recent theoretical models suggest that deciding about actions and executing them are not implemented by completely distinct neural mechanisms but are instead two modes of an integrated dynamical system. Here, we investigate this proposal by examining how neural activity unfolds during a dynamic decision-making task within the high-dimensional space defined by the activity of cells in monkey dorsal premotor (PMd), primary motor (M1), and dorsolateral prefrontal cortex (dlPFC) as well as the external and internal segments of the globus pallidus (GPe, GPi). Dimensionality reduction shows that the four strongest components of neural activity are functionally interpretable, reflecting a state transition between deliberation and commitment, the transformation of sensory evidence into a choice, and the baseline and slope of the rising urgency to decide. Analysis of the contribution of each population to these components shows meaningful differences between regions but no distinct clusters within each region, consistent with an integrated dynamical system. During deliberation, cortical activity unfolds on a two-dimensional "decision manifold" defined by sensory evidence and urgency and falls off this manifold at the moment of commitment into a choice-dependent trajectory leading to movement initiation. The structure of the manifold varies between regions: In PMd, it is curved; in M1, it is nearly perfectly flat; and in dlPFC, it is almost entirely confined to the sensory evidence dimension. In contrast, pallidal activity during deliberation is primarily defined by urgency. We suggest that these findings reveal the distinct functional contributions of different brain regions to an integrated dynamical system governing action selection and execution.

Biomechanical costs influence decisions made during ongoing actions.

Accurate interaction with the environment relies on the integration of external information about the spatial layout of potential actions and knowledge of their costs and benefits. Previous studies have shown that when given a choice between voluntary reaching movements, humans tend to prefer actions with lower biomechanical costs. However, these studies primarily focused on decisions made before the onset of movement ("decide-then-act" scenarios), and it is not known to what extent their conclusions generalize to many real-life situations, in which decisions occur during ongoing actions ("decide-while-acting"). For example, one recent study found that biomechanical costs did not influence decisions to switch from a continuous manual tracking movement to a point-to-point movement, suggesting that biomechanical costs may be disregarded in decide-while-acting scenarios. To better understand this surprising result, we designed an experiment in which participants were faced with the decision between continuing to track a target moving along a straight path or changing paths to track a new target that gradually moved along a direction that deviated from the initial one. We manipulated tracking direction, angular deviation rate, and side of deviation, allowing us to compare scenarios where biomechanical costs favored either continuing or changing the path. Crucially, here the choice was always between two continuous tracking actions. Our results show that in this situation, decisions clearly took biomechanical costs into account. Thus, we conclude that biomechanics are not disregarded during decide-while-acting scenarios, but rather, that cost comparisons can only be made between similar types of actions.

Parallel processing of value-related information during multi-attribute decisions.

When choosing between options with multiple attributes, do we integrate all attributes into a unified measure for comparison, or does the comparison also occur at the level of each attribute, involving parallel processes that can dynamically influence each other? What happens when independent sensory features all carry information about the same decision factor, such as reward value? To investigate these questions, we asked human participants to perform a two-alternative forced choice reaching task in which the reward value of a target was indicated by two visual attributes-its brightness ("bottom-up," BU feature) and its orientation ("top-down," TD feature). If decisions always occur after the integration of both features, there should be no difference in the reaction time (RT) regardless of the attribute combinations that drove the choice. Counter to that prediction, RT distributions depended on the attribute combinations of given targets and the choices made by participants. RTs were shortest when both attributes were congruent or when the choice was based on the bottom-up feature, and longer when the attributes were in conflict (favoring opposite options). In conflict trials, nearly two-thirds of participants made faster decisions when choosing the option favored by the bottom-up feature than when choosing the top-down-favored option. We also observed mid-reach changes-of-mind in a subset of conflict trials, mostly changing from the bottom-up to the top-down-favored target. These data suggest that multi-attribute value-based decisions are better explained by a distributed process including competition among different features than by a competition based on a single, integrated estimate of value.NEW & NOTEWORTHY We show that during value-based decisions, humans do not always use all reward-related information to make their choice, but instead can "jump the gun" using partial information. In particular, when different sources of information were in conflict, early decisions were mostly based on fast bottom-up information, and sometimes followed by corrective changes-of-mind based on slower top-down information. This supports parallel decision processes among different information sources, as opposed to a single integrated "common currency."

Performance-gated deliberation: A context-adapted strategy in which urgency is opportunity cost.

Finding the right amount of deliberation, between insufficient and excessive, is a hard decision making problem that depends on the value we place on our time. Average-reward, putatively encoded by tonic dopamine, serves in existing reinforcement learning theory as the opportunity cost of time, including deliberation time. Importantly, this cost can itself vary with the environmental context and is not trivial to estimate. Here, we propose how the opportunity cost of deliberation can be estimated adaptively on multiple timescales to account for non-stationary contextual factors. We use it in a simple decision-making heuristic based on average-reward reinforcement learning (AR-RL) that we call Performance-Gated Deliberation (PGD). We propose PGD as a strategy used by animals wherein deliberation cost is implemented directly as urgency, a previously characterized neural signal effectively controlling the speed of the decision-making process. We show PGD outperforms AR-RL solutions in explaining behaviour and urgency of non-human primates in a context-varying random walk prediction task and is consistent with relative performance and urgency in a context-varying random dot motion task. We make readily testable predictions for both neural activity and behaviour.

Evidence and Urgency Related EEG Signals during Dynamic Decision-Making in Humans.

A successful class of models link decision-making to brain signals by assuming that evidence accumulates to a decision threshold. These evidence accumulation models have identified neuronal activity that appears to reflect sensory evidence and decision variables that drive behavior. More recently, an additional evidence-independent and time-variant signal, called urgency, has been hypothesized to accelerate decisions in the face of insufficient evidence. However, most decision-making paradigms tested with fMRI or EEG in humans have not been designed to disentangle evidence accumulation from urgency. Here we use a face-morphing decision-making task in combination with EEG and a hierarchical Bayesian model to identify neural signals related to sensory and decision variables, and to test the urgency-gating model. Forty females and 34 males took part (mean age, 23.4 years). We find that an evoked potential time locked to the decision, the centroparietal positivity, reflects the decision variable from the computational model. We further show that the unfolding of this signal throughout the decision process best reflects the product of sensory evidence and an evidence-independent urgency signal. Urgency varied across subjects, suggesting that it may represent an individual trait. Our results show that it is possible to use EEG to distinguish neural signals related to sensory evidence accumulation, decision variables, and urgency. These mechanisms expose principles of cognitive function in general and may have applications to the study of pathologic decision-making such as in impulse control and addictive disorders.SIGNIFICANCE STATEMENT Perceptual decisions are often described by a class of models that assumes that sensory evidence accumulates gradually over time until a decision threshold is reached. In the present study, we demonstrate that an additional urgency signal impacts how decisions are formed. This endogenous signal encourages one to respond as time elapses. We found that neural decision signals measured by EEG reflect the product of sensory evidence and an evidence-independent urgency signal. A nuanced understanding of human decisions, and the neural mechanisms that support it, can improve decision-making in many situations and potentially ameliorate dysfunction when it has gone awry.

Trading accuracy for speed over the course of a decision.

Humans and other animals often need to balance the desire to gather sensory information (to make the best choice) with the urgency to act, facing a speed-accuracy tradeoff (SAT). Given the ubiquity of SAT across species, extensive research has been devoted to understanding the computational mechanisms allowing its regulation at different timescales, including from one context to another, and from one decision to another. However, animals must frequently change their SAT on even shorter timescales-that is, over the course of an ongoing decision-and little is known about the mechanisms that allow such rapid adaptations. The present study aimed at addressing this issue. Human subjects performed a decision task with changing evidence. In this task, subjects received rewards for correct answers but incurred penalties for mistakes. An increase or a decrease in penalty occurring halfway through the trial promoted rapid SAT shifts, favoring speeded decisions either in the early or in the late stage of the trial. Importantly, these shifts were associated with stage-specific adjustments in the accuracy criterion exploited for committing to a choice. Those subjects who decreased the most their accuracy criterion at a given decision stage exhibited the highest gain in speed, but also the highest cost in terms of performance accuracy at that time. Altogether, the current findings offer a unique extension of previous work, by suggesting that dynamic cha*nges in accuracy criterion allow the regulation of the SAT within the timescale of a single decision.NEW & NOTEWORTHY Extensive research has been devoted to understanding the mechanisms allowing the regulation of the speed-accuracy tradeoff (SAT) from one context to another and from one decision to another. Here, we show that humans can voluntarily change their SAT on even shorter timescales-that is, over the course of a decision. These rapid SAT shifts are associated with dynamic adjustments in the accuracy criterion exploited for committing to a choice.

Microstimulation of dorsal premotor and primary motor cortex delays the volitional commitment to an action choice.

Humans and other animals are faced with decisions about actions on a daily basis. These typically include a period of deliberation that ends with the commitment to a choice, which then leads to the overt expression of that choice through action. Previous studies with monkeys have demonstrated that neural activity in sensorimotor areas correlates with the deliberation process and reflects the moment of commitment before movement initiation, but the causal roles of these regions are challenging to establish. Here, we tested whether dorsal premotor (PMd) and primary motor cortex (M1) are causally involved in the volitional commitment to a reaching choice. We found that brief subthreshold microstimulation in PMd or M1 delayed commitment to an action but not the initiation of the action itself. Importantly, microstimulation only had a significant effect when it was delivered close to and before commitment time. These results are consistent with the proposal that PMd and M1 participate in the commitment process, which occurs when a critical firing rate difference is reached between cells voting for the selected option and those voting for the competing one.NEW & NOTEWORTHY The neural substrates of decisions between actions are typically investigated by correlating neural activity and subjects' decision behavior, but this does not establish causality. In a reaching decision task, we demonstrate that subthreshold microstimulation of the monkey dorsal premotor cortex or primary motor cortex delays the deliberation duration if applied shortly before choice commitment. This result suggests a causal role of the sensorimotor cortex in the determination of decisions between actions.

Reaching decisions during ongoing movements.

Neurophysiological studies suggest that when decisions are made between concrete actions, the selection process involves a competition between potential action representations in the same sensorimotor structures involved in executing those actions. However, it is unclear how such models can explain situations, often encountered during natural behavior, in which we make decisions while were are already engaged in performing an action. Does the process of deliberation characterized in classical studies of decision-making proceed the same way when subjects are deciding while already acting? In the present study, human subjects continuously tracked a target moving in the horizontal plane and were occasionally presented with a new target to which they could freely choose to switch at any time, whereupon it became the new tracked target. We found that the probability of choosing to switch increased with decreasing distance to the new target and increasing size of the new target relative to the tracked target, as well as when the direction to the new target was aligned (either toward or opposite) to the current tracking direction. However, contrary to our expectations, subjects did not choose targets that minimized the energetic costs of execution, as calculated by a biomechanical model of the arm. When the constraints of continuous tracking were removed in variants of the task involving point-to-point movements, the expected preference for lower cost choices was seen. These results are discussed in the context of current theories of nested feedback control, internal models of forward dynamics, and high-dimensional neural spaces.NEW & NOTEWORTHY Current theories of decision-making primarily address how subjects make decisions before executing selected actions. However, in our daily lives we often make decisions while already performing some action (e.g., while playing a sport or navigating through a crowd). To gain insight into how current theories can be extended to such "decide-while-acting" scenarios, we examined human decisions during continuous manual tracking and found some intriguing departures from how decisions are made in classical "decide-then-act" paradigms.

Motor cortex disruption delays motor processes but not deliberation about action choices.

Decisions about actions typically involve a period of deliberation that ends with the commitment to a choice and the motor processes overtly expressing that choice. Previous studies have shown that neural activity in sensorimotor areas, including the primary motor cortex (M1), correlates with deliberation features during action selection. However, the causal contribution of these areas to the decision process remains unclear. Here, we investigated whether M1 determines choice commitment or whether it simply reflects decision signals coming from upstream structures and instead mainly contributes to the motor processes that follow commitment. To do so, we tested the impact of a disruption of M1 activity, induced by continuous theta burst stimulation (cTBS), on the behavior of human subjects in 1) a simple reaction time (SRT) task allowing us to estimate the duration of the motor processes and 2) a modified version of the tokens task (Cisek P, Puskas GA, El-Murr S. J Neurosci 29: 11560-11571, 2009), which allowed us to estimate subjects' time of commitment as well as accuracy criterion. The efficiency of cTBS was attested by a reduction in motor evoked potential amplitudes following M1 disruption compared with those following a sham stimulation. Furthermore, M1 cTBS lengthened SRTs, indicating that motor processes were perturbed by the intervention. Importantly, all of the behavioral results in the tokens task were similar following M1 disruption and sham stimulation, suggesting that the contribution of M1 to the deliberation process is potentially negligible. Taken together, these findings favor the view that M1 contribution is downstream of the decision process.NEW & NOTEWORTHY Decisions between actions are ubiquitous in the animal realm. Deliberation during action choices entails changes in the activity of the sensorimotor areas controlling those actions, but the causal role of these areas is still often debated. With the use of continuous theta burst stimulation, we show that disrupting the primary motor cortex (M1) delays the motor processes that follow instructed commitment but does not alter volitional deliberation, suggesting that M1 contribution may be downstream of the decision process.

A sensorimotor alternative to coding is possible.

If we abandon the coding metaphor in favor of models of the full behavioral loop, we need a way to dissect that loop into understandable pieces. I suggest that evolutionary data provide a solution. We can subdivide behavior into parallel sensorimotor subsystems by following the phylogenetic history of how those systems differentiated and specialized during our evolution, leading to promising ways of re-interpreting neural activity within the context of its pragmatic role in mediating interaction.

Neural mechanisms for interacting with a world full of action choices.

The neural bases of behavior are often discussed in terms of perceptual, cognitive, and motor stages, defined within an information processing framework that was originally inspired by models of human abstract problem solving. Here, we review a growing body of neurophysiological data that is difficult to reconcile with this influential theoretical perspective. As an alternative foundation for interpreting neural data, we consider frameworks borrowed from ethology, which emphasize the kinds of real-time interactive behaviors that animals have engaged in for millions of years. In particular, we discuss an ethologically-inspired view of interactive behavior as simultaneous processes that specify potential motor actions and select between them. We review how recent neurophysiological data from diverse cortical and subcortical regions appear more compatible with this parallel view than with the classical view of serial information processing stages.

Modulation of Premotor and Primary Motor Cortical Activity during Volitional Adjustments of Speed-Accuracy Trade-Offs.

Recent work suggests that while animals decide between reaching actions, neurons in dorsal premotor (PMd) and primary motor (M1) cortex reflect a dynamic competition between motor plans and determine when commitment to a choice is made. This competition is biased by at least two sources of information: the changing sensory evidence for one choice versus another, and an urgency signal that grows over time. Here, we test the hypothesis that the urgency signal adjusts the trade-off between speed and accuracy during both decision-making and movement execution. Two monkeys performed a reaching decision task in which sensory evidence continuously evolves over the course of each trial. In different blocks, task timing parameters encouraged monkeys to voluntarily adapt their behavior to be either hasty or conservative. Consistent with our hypothesis, during the deliberation process the baseline and gain of neural activity in decision-related PMd (29%) and M1 cells (45%) was higher when monkeys applied a hasty policy than when they behaved conservatively, but at the time of commitment the population activity was similar across blocks. Other cells (30% in PMd, 30% in M1) showed activity that increased or decreased with elapsing time until the moment of commitment. Movement-related neurons were also more active after longer decisions, as if they were influenced by the same urgency signal controlling the gain of decision-related activity. Together, these results suggest that the arm motor system receives an urgency/vigor signal that adjusts the speed-accuracy trade-off for decision-making and movement execution. Significance statement: This work addresses the neural mechanisms that control the speed-accuracy trade-off in both decisions and movements, in the kinds of dynamic situations that are typical of natural animal behavior. We found that many "decision-related" premotor and motor neurons are modulated in a time-dependent manner compatible with an "urgency" signal that changes between hasty and conservative decision policies. We also found that such modulation influenced cells related to the speed of the reaching movements executed by the animals to report their decisions. These results suggest that a unified mechanism determines speed-accuracy trade-off adjustments during decision-making and movement execution, potentially influencing both the cognitive and motor aspects of reward-oriented behavior.

Trial-to-trial adjustments of speed-accuracy trade-offs in premotor and primary motor cortex.

Recent studies have shown that activity in sensorimotor structures varies depending on the speed-accuracy trade-off (SAT) context in which a decision is made. Here we tested the hypothesis that the same areas also reflect a more local adjustment of SAT established between individual trials, based on the outcome of the previous decision. Two monkeys performed a reaching decision task in which sensory evidence continuously evolves during the time course of a trial. In two SAT contexts, we compared neural activity in trials following a correct choice vs. those following an error. In dorsal premotor cortex (PMd), we found that 23% of cells exhibited significantly weaker baseline activity after error trials, and for ∼30% of these this effect persisted into the deliberation epoch. These cells also contributed to the process of combining sensory evidence with the growing urgency to commit to a choice. We also found that the activity of 22% of PMd cells was increased after error trials. These neurons appeared to carry less information about sensory evidence and time-dependent urgency. For most of these modulated cells, the effect was independent of whether the previous error was expected or unexpected. We found similar phenomena in primary motor cortex (M1), with 25% of cells decreasing and 34% increasing activity after error trials, but unlike PMd, these neurons showed less clear differences in their response properties. These findings suggest that PMd and M1 belong to a network of brain areas involved in SAT adjustments established using the recent history of reinforcement. NEW & NOTEWORTHY: Setting the speed-accuracy trade-off (SAT) is crucial for efficient decision making. Previous studies have reported that subjects adjust their SAT after individual decisions, usually choosing more conservatively after errors, but the neural correlates of this phenomenon are only partially known. Here, we show that neurons in PMd and M1 of monkeys performing a reach decision task support this mechanism by adequately modulating their firing rate as a function of the outcome of the previous decision.

Giant Aneurysm of In Situ Saphenous Vein Graft.

We report a case of a symptomatic and rapidly expanding aneurysm of an in situ saphenous vein graft in a 70-year-old man with extensive prior open and endovascular procedures for aneurysmal disease. He was found to have full-length aneurysmal dilation with rapid progression over the course of 6 months. Successful ligation and exclusion with subtotal excision of the aneurysmal segment was performed, and revision bypass was foregone because of adequate distal perfusion via collateralization. This rare complication of autologous vein bypass graft typically occurs at the anastomoses and merits aneurysmorrhaphy or ligation. Far less common is the scenario of nonanastomotic, true aneurysmal dilation of the entire conduit such as the current patient. Surgical intervention via open, endovascular, or a combination thereof is warranted for rapidly growing or symptomatic defects and those yielding distal malperfusion.

Evidence against perfect integration of sensory information during perceptual decision making.

Perceptual decision making is often modeled as perfect integration of sequential sensory samples until the accumulated total reaches a fixed decision bound. In that view, the buildup of neural activity during perceptual decision making is attributed to temporal integration. However, an alternative explanation is that sensory estimates are computed quickly with a low-pass filter and combined with a growing signal reflecting the urgency to respond and it is the latter that is primarily responsible for neural activity buildup. These models are difficult to distinguish empirically because they make similar predictions for tasks in which sensory information is constant within a trial, as in most previous studies. Here we presented subjects with a variant of the classic constant-coherence motion discrimination (CMD) task in which we inserted brief motion pulses. We examined the effect of these pulses on reaction times (RTs) in two conditions: 1) when the CMD trials were blocked and subjects responded quickly and 2) when the same CMD trials were interleaved among trials of a variable-motion coherence task that motivated slower decisions. In the blocked condition, early pulses had a strong effect on RTs but late pulses did not, consistent with both models. However, when subjects slowed their decision policy in the interleaved condition, later pulses now became effective while early pulses lost their efficacy. This last result contradicts models based on perfect integration of sensory evidence and implies that motion signals are processed with a strong leak, equivalent to a low-pass filter with a short time constant.

Context-dependent urgency influences speed-accuracy trade-offs in decision-making and movement execution.

Speed-accuracy tradeoffs (SATs) exist in both decision-making and movement control, and are generally studied separately. However, in natural behavior animals are free to adjust the time invested in deciding and moving so as to maximize their reward rate. Here, we investigate whether shared mechanisms exist for SAT adjustment in both decisions and actions. Two monkeys performed a reach decision task in which they watched 15 tokens jump, one every 200 ms, from a central circle to one of two peripheral targets, and had to guess which target would ultimately receive the majority of tokens. The monkeys could decide at any time, and once a target was reached, the remaining token movements accelerated to either 50 ms ("fast" block) or 150 ms ("slow" block). Decisions were generally earlier and less accurate in fast than slow blocks, and in both blocks, the criterion of accuracy decreased over time within each trial. This could be explained by a simple model in which sensory information is combined with a linearly growing urgency signal. Remarkably, the duration of the reaching movements produced after the decision decreased over time in a similar block-dependent manner as the criterion of accuracy estimated by the model. This suggests that SATs for deciding and acting are influenced by a shared urgency/vigor signal. Consistent with this, we observed that the vigor of saccades performed during the decision process was higher in fast than in slow blocks, suggesting the influence of a context-dependent global arousal.

Rapid prediction of biomechanical costs during action decisions.

When given a choice between actions that yield the same reward, we tend to prefer the one that requires the least effort. Recent studies have shown that humans are remarkably accurate at evaluating the effort of potential reaching actions and can predict the subtle energetic demand caused by the nonisotropic biomechanical properties of the arm. In the present study, we investigated the time course over which such information is computed and comes to influence decisions. Two independent approaches were used. First, subjects performed a reach decision task in which the time interval for deciding between two candidate reaching actions was varied from 200 to 800 ms. Second, we measured motor-evoked potential (MEPs) to single-pulse transcranial magnetic stimulation (TMS) over the primary motor cortex (M1) to probe the evolving decision at different times after stimulus presentation. Both studies yielded a consistent conclusion: that a prediction of the effort associated with candidate movements is computed very quickly and influences decisions within 200 ms after presentation of the candidate actions. Furthermore, whereas the MEPs measured 150 ms after stimulus presentation were well correlated with the choices that subjects ultimately made, later in the trial the MEP amplitudes were primarily related to the muscular requirements of the chosen movement. This suggests that corticospinal excitability (CSE) initially reflects a competition between candidate actions and later changes to reflect the processes of preparing to implement the winning action choice.

Toward a neuroscience of natural behavior.

One of the most exciting new developments in systems neuroscience is the progress being made toward neurophysiological experiments that move beyond simplified laboratory settings and address the richness of natural behavior. This is enabled by technological advances such as wireless recording in freely moving animals, automated quantification of behavior, and new methods for analyzing large data sets. Beyond new empirical methods and data, however, there is also a need for new theories and concepts to interpret that data. Such theories need to address the particular challenges of natural behavior, which often differ significantly from the scenarios studied in traditional laboratory settings. Here, we discuss some strategies for developing such novel theories and concepts and some example hypotheses being proposed.

Generating meaning: active inference and the scope and limits of passive AI.

Prominent accounts of sentient behavior depict brains as generative models of organismic interaction with the world, evincing intriguing similarities with current advances in generative artificial intelligence (AI). However, because they contend with the control of purposive, life-sustaining sensorimotor interactions, the generative models of living organisms are inextricably anchored to the body and world. Unlike the passive models learned by generative AI systems, they must capture and control the sensory consequences of action. This allows embodied agents to intervene upon their worlds in ways that constantly put their best models to the test, thus providing a solid bedrock that is - we argue - essential to the development of genuine understanding. We review the resulting implications and consider future directions for generative AI.