TL;DR
This paper uses novel single-trial neural decoding to reveal distinct internal neural events during free choice decisions in monkeys, including rapid choices, "changes of mind," and moments of "indecision." These findings provide a window into the dynamic neural processes underlying complex decision-making, moving beyond averaged brain activity.
What Problem Does This Paper Address?
Traditional studies of decision-making often rely on averaging neural activity across many trials. This approach can obscure crucial trial-to-trial variations that are essential for understanding the nuances of decision-making, such as hesitation, wavering, or changing one's mind. The authors aim to overcome this limitation by developing a method to decode neural activity on a moment-by-moment basis within individual trials, allowing them to observe these internal decision-making events as they unfold. Specifically, they investigate how monkeys make choices when presented with multiple viable options, a scenario that mirrors complex real-world decisions.
Methods
The core of this study is a sophisticated single-trial neural decoding approach. The researchers recorded neural activity from the dorsal premotor cortex (PMd) and primary motor cortex (M1) of two monkeys performing a decision-making task. They trained a linear decoder using data from "forced-choice" trials (where only one option was available) to predict the monkey's intended movement. This decoder was then applied to "free-choice" trials (where multiple options were available) to track the neural representation of the decision over time within each trial. To handle the high-dimensional neural data, they employed Gaussian Process Factor Analysis (GPFA) for dimensionality reduction and smoothing before training a Support Vector Machine (SVM) classifier. This allowed them to visualize the neural state's trajectory and identify distinct patterns associated with different decision-making behaviors.
Datasets
The study utilized data from two adult male macaque monkeys (J and N). They collected data over multiple sessions, resulting in seven distinct datasets (J1-T, J1-S, J2-T, J2-S, N1-S, N2-S, N3-S). The task involved a novel decision-maze variant, which included different maze configurations ('T-maze' and 'S-maze') and varying barrier difficulties. This task design allowed for the creation of "forced choices," "free choices," and "encouraged switch" scenarios (where a barrier change mid-trial could prompt a change of mind). No new benchmarks were introduced; rather, the existing task was adapted to elicit more varied decision-making behaviors.
Key Insights and Ideas
- Neural activity mirrors behavioral variability: The study demonstrates that moment-by-moment neural activity in motor cortex can capture the rich behavioral variability observed during free choices, including decisive actions, changes of mind, and indecision.
- "Changes of mind" are not just sensory-driven: The research provides evidence for spontaneous "changes of mind" that occur even without changes in external sensory evidence, suggesting internal re-evaluation processes.
- Indecision has a neural correlate: Moments of behavioral indecision (slowed reaction times) are accompanied by specific neural patterns, suggesting a distinct neural state of "indecision" rather than simply a lack of a motor plan.
- Forced vs. Free choice similarity: Despite the potential for complex internal dynamics in free choices, the initial motor plans formed in PMd and M1 are remarkably similar to those formed during forced choices, suggesting that the core decision-making machinery might operate upstream.
Key Results
- A decoder trained on forced-choice trials generalized well to free-choice trials, achieving high accuracy (e.g., 96% for monkey J, 88% for monkey N at the final time point).
- Three distinct categories of internal neural events were identified during free choices:
- Rapid, unwavering choices: The most common pattern, similar to forced choices.
- "Changes of mind": Neural state initially reflected one choice before switching to reflect the final choice. These occurred in 13% of free-choice trials.
- Momentary "indecision": A delay in forming a clear motor plan, observed in specific conditions and accompanied by behavioral indecision.
- Spontaneous vacillations (changes of mind without external evidence changes) were significantly more common in free-choice trials (13%) than in forced-choice trials (2%).
- Induced changes of mind (e.g., due to barrier changes) were also common, with "likely change" trials showing neural changes of mind 41% of the time.
- Hesitation (slowed RTs when a new option became available but wasn't taken) did not involve re-planning in the decoded neural activity, suggesting a "hesitation to execute" rather than a change of mind.
Limitations and Future Work
- Decoder limitations: The decoder's inability to access data immediately before movement onset means that very late changes of mind or re-planning events might be missed.
- Dataset variability: The authors acknowledge that individual datasets can show differences in behavior and neural strategies, suggesting that future work could explore these variations further.
- Location of decision-making: The study does not pinpoint the exact brain region where the decision is made, only that PMd and M1 reflect the ongoing process. Future work could investigate upstream areas.
- Mechanistic comparison: While "hesitation" is compared to saccadic inhibition, a more detailed mechanistic comparison with other cognitive processes could be explored.
- Future directions: The authors suggest that their approach can be used to investigate other complex decision-making phenomena and that the similarity between forced and free choices has implications for neural prosthetics. They also propose that their findings challenge previous conclusions about the timing of conscious intention and action.
Summary updated