How do learning progress and novelty drive human curiosity?
People decide to stop exploring a task when they stop learning from it, and they choose their next task based on both how much they expect to learn and how new it is.
Source
Contributions of expected learning progress and perceptual novelty to curiosity-driven exploration
What they did
Researchers tested 55 participants in an online game where they selected and tracked 3 different cartoon characters hiding behind a hedge. Each character's hiding spot followed a unique probability distribution with differing levels of noise and volatility. The researchers tracked when participants chose to switch characters and which character they selected next, using computational modeling to analyze how learning progress and novelty affected these choices.
What they found
Participants were more likely to keep playing with a character when their learning progress was high, with the probability of switching increasing as learning progress decreased, showing a beta of -0.13. When choosing a new character, participants were significantly drawn to both expected learning progress (beta of 0.24) and perceptual novelty (beta of 0.87). Furthermore, the influence of novelty on their choices decreased over time, while the influence of expected learning progress remained stable.
The limits
What it doesn't show
First, the study was conducted entirely online with a sample of undergraduate students, which may not generalize to broader or older populations. Second, because the study did not include external rewards, it does not show how curiosity-driven exploration interacts with or trade-offs against extrinsic motivations like money or points. Finally, while the computational model infers internal states of learning progress, it cannot definitively prove these calculations are consciously occurring in the participants' minds.
Key terms
- Volatility
- The rate at which the underlying rules or patterns of an environment change over time.
- Noise
- Random fluctuations or irregular variations in a stimulus that obscure the underlying pattern.
- Expected learning progress
- An agent's prediction of how much their accuracy or performance will improve if they continue to sample a particular source of information.
- Perceptual novelty
- The degree of unfamiliarity or newness of a physical stimulus based on how little an observer has interacted with it.
- Intrinsic motivation
- A drive to engage in an activity or behavior for its own sake or for internal rewards, such as curiosity or learning, rather than for external incentives.
- Computational modeling
- The use of mathematical and computer-based algorithms to simulate, quantify, and understand cognitive processes or behavioral strategies.
Flashcards
Want these cards to stick?
Save the deck to NoteFren and study it with spaced repetition.
Quiz yourself
What key limitation of previous adult studies on learning progress did this research address?
Common questions
Did participants receive any rewards or money for finding the characters?
No, the participants did not receive any external rewards, points, or financial incentives for correctly guessing where the characters were hiding. Their behavior was driven entirely by intrinsic curiosity.
How did the researchers measure learning progress without asking participants directly?
Instead of relying on subjective self-reports, the researchers used a computational model to track how much participants reduced their prediction errors trial-by-trial, using these math-based changes as an objective measure of learning progress.
Which type of hiding environment did participants prefer the most?
Participants showed a significant preference for the intermediate environment, which had moderate levels of both noise and volatility, over the highly noisy or highly volatile ones.
How does novelty affect our choices differently over time compared to learning progress?
Novelty is highly influential early on when everything is new, but its effect decreases significantly as time passes. In contrast, the desire to make expected learning progress remains a steady, long-term influence on exploration choices.
More on Learning