How do learning progress and novelty drive human curiosity?

What is the primary objective of this study?

The study aims to investigate the cognitive processes underlying curiosity-driven exploration in unknown environments, specifically testing the roles of expected learning progress, perceptual novelty, and random search.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How does the paper define environmental 'noise'?

Noise consists of irregular, unpredictable fluctuations that affect a stimulus but are not a part of the stimulus itself, thereby obscuring it.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How does the paper define environmental 'volatility'?

Volatility refers to how quickly the underlying mean or state of a stimulus or environment changes over time.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What limitation of previous adult studies on learning progress did this study address?

Previous adult studies relied on subjective self-reports of learning progress, which require strong self-monitoring skills and may not match implicit exploration behaviors.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

Why is a purely novelty-based exploration strategy theoretically insufficient in unknown environments?

Relying strictly on novelty can lead individuals to spend excessive time in unlearnable situations that contain only random noise and no useful information.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI

What was the final sample size analyzed in the study?

The final analyzed sample consisted of 55 participants, after discarding 9 due to inclusion criteria and 3 due to technical problems.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What was the behavioral paradigm of the online game played by the participants?

Participants selected one of three characters, watched it hide behind a hedge, and tried to guess where it would reappear by clicking on the hedge.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How were the hiding distributions of the three characters manipulated?

They were independently set with different levels of noise (standard deviation) and volatility (rate of mean change) based on Gaussian distributions.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How did the computational model optimize prediction and detect environment shifts?

It utilized a reduced Bayesian learner designed to infer the mean and standard deviation of distributions while tracking change-point probabilities.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How is change-point probability calculated in the model?

It is computed using Bayes' theorem, comparing the likelihood of a uniform distribution (representing a state change) to a normal distribution.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What happens to past evidence in the model's mean estimation when a change-point occurs?

All evidence observed prior to the change-point is completely disregarded, and the mean estimate is updated using only the latest trial's data.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What role does the learning rate (alpha) play in the model?

The learning rate dictates how strongly the difference between past predictions and new evidence (prediction error) adjusts future expectations.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How is 'current learning progress' (LP_t) defined within the model?

It is defined as the absolute difference in prediction errors between the previous trial and the current trial.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI

How is 'expected learning progress' calculated in the study's model?

It is computed by subtracting the predicted future prediction error from the current prediction error.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI

Describe the hierarchical reinforcement learning structure used in the model.

The model operates on two layers: the first layer estimates the expected mean, and the second layer estimates the expected prediction error using the same learning rate.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How did the researchers quantify perceptual novelty (N)?

Perceptual novelty was modeled non-parametrically as a smooth function of the negative overall exposure (total trials observed) to a given character.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI

How was the model's learning rate (alpha) validated?

By simulating data from 60 synthetic participants and demonstrating a very high correlation (r = 0.978) between the simulated and recovered parameters.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What does a learning-maximization account predict about environment-switching decisions?

It predicts that the probability of leaving an environment depends on current learning progress, and the choice of the next environment depends on the difference in expected progress.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What does a novelty-driven exploration account predict about when a participant will switch?

It predicts that the likelihood of switching to a new stimulus increases as the perceptual novelty of the current stimulus decays over time.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What does a random search account predict about switching behavior?

It predicts that participants disengage from stimuli at random intervals, with cumulative switch probability rising as time since the last switch increases.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What criteria were used to exclude participants from the final analysis?

Participants were excluded if they did not explore more than one character environment, or if they explicitly reported failing to follow task instructions.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

Which factors significantly predicted a participant's decision to leave an environment?

Learning progress and time spent significantly predicted switching decisions, whereas perceptual novelty had no significant effect.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

Which factors significantly predicted which new environment a participant would choose next?

Both the expected learning progress and the perceptual novelty of the options positively and significantly predicted future environmental choices.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How did the influence of perceptual novelty on environment selection change over time?

The impact of perceptual novelty on participants' selection choices significantly decreased as the experiment progressed.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

In which environment conditions did participants fail to minimize prediction error?

Participants failed to minimize prediction error over time in the high-noise environment, although they succeeded in the high-volatility and intermediate environments.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How did participants adapt their learning rate (alpha) to different types of uncertainty?

They increased alpha over time in the high-noise environment, decreased it in the high-volatility environment, and kept it stable in the intermediate environment.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

Which environment did participants subjectively prefer the most?

Participants significantly preferred the intermediate environment and liked the high-noise environment the least.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What is the primary theoretical conclusion regarding human exploration in unknown environments?

The study provides direct evidence that human curiosity-driven exploration is actively guided by the maximization of learning progress.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How does expected learning progress relate to reinforcement learning systems?

Expected learning progress naturally emerges in hierarchical RL systems by subtracting predicted future prediction error from current prediction error.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How do these human findings relate to AI 'learning progress theory'?

They show that humans, like artificial agents, leverage learning progress to systematically structure and optimize their exploration of unknown spaces.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

How do the study's findings support the predictive coding framework?

They offer a mechanistic explanation: by striving to maximize learning progress, organisms systematically and indirectly minimize overall uncertainty.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What does the adaptive tuning of learning rates reveal about cognitive flexibility?

It reveals that human learning is highly adaptive, shifting updating speeds dynamically depending on whether uncertainty stems from noise or volatility.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What future neuroscientific direction do the authors suggest for learning progress research?

They highlight the need to investigate whether learning progress internally activates dopaminergic reward networks similarly to external rewards.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

What developmental difference in exploration-exploitation is noted by the authors?

Young children show a stronger bias toward exploration than adults, meaning they explore more widely but secure fewer immediate rewards.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified

Why was the lack of external rewards essential to this experimental design?

Removing rewards allowed the researchers to isolate pure, curiosity-driven intrinsic motivation without the confounding influence of reward exploitation.

Source: Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration · DOI · quote verified