Emergent Behaviour in Game AI: A Genetic Programming and CNN-based Approach to Intelligent Agent Design
Bibliographic record
Abstract
Emergent behaviour is behaviour that arises from the interactions between the individual components of a system, rather than being explicitly programmed or designed. In this work, genetic programming is used to evolve an adaptive game AI, also known as an intelligent agent, whose job is to capture up to twenty-five prey agents in a simulated pursuit environment. For a pursuit game, the fitness score tallies each prey the predator captured during a run. The fitness is then used to evaluate each agent and choose who moves forward in the evolutionary process. A problem with only choosing the best performing agents is that genetic diversity becomes lost along the way, which can result in monotonous behaviour. Diverse behaviour helps agents perform under situations of uncertainty and creates more interesting computer opponents in video games, as it encourages the agent to explore different possibilities and adapt to changing circumstances. Inspired by the works of Cowan and Pozzuoli in diversifying agent behaviour, and Chen’s work in L-system tree evaluation, a convolutional neural network is introduced to automatically model the behaviour of each agent, something previously done manually. This involves training the convolutional neural network on a large data set of behaviours exhibited by the agents, which take the form of image-based traces. The resulting model is then used to detect interesting emergent behaviour. In the first set of experiments, the convolutional neural network is trained and tested on several sets of traces, then the performance of each run is analyzed. Results show that the convolutional neural network is capable of identifying 6 emergent behaviours with 98% accuracy. The second set of experiments combine genetic programming and the convolutional neural network in order to produce unique and interesting intelligent agents, as well as target specific behaviours. Results show that the system is able to evolve more innovative and effective agents that can operate in complex environments and could be extended to perform a wide range of tasks.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.002 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".