A team from South Ural State University and the Russian Academy of Sciences' Institute of Psychology has built a system to automatically evaluate human creativity. Their method replaces manual scoring of projective psychological tests with a trained neural network.

The engineering effort started with a dataset of 3,000 Torrance Tests of Creative Thinking results from Russian participants. Using this data, the team developed an AI model to process outcomes from multiple creativity diagnostics. The core technical decision was to train the network on this specific, curated dataset rather than a general image or text corpus. This approach tailors the model to the nuanced scoring rubrics of psychological assessment.

In production, the system delivers a Torrance test result in minutes, a task that typically requires significant expert analysis. The team also expanded the model's output beyond the six standard Torrance metrics, incorporating additional indicators for cultural code and personal traits. A public web portal now allows users to take tests based on the Torrance and Urban methodologies, receiving automated scores without a psychologist.

The architecture presents a clear trade-off: it excels at rapid, consistent scoring of known test patterns but does not replicate the interpretive depth of a human expert. It functions as a force multiplier for psychologists, handling initial quantitative analysis. This project mirrors a separate initiative where Russian emergency services use a neural network to analyze fire scene photos, reducing a 40-60 minute documentation process to seconds. Both cases demonstrate a practical design pattern: applying constrained models to well-defined, labor-intensive analytical tasks.

Source: Lenta.RU