Forecasting AGI: A Granular, CHC-Based Approach
Habeeb Abdulfatah · Team Habyb
Submitted to The AI Forecasting Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.
The paper introduces a data-driven framework for forecasting Artificial General Intelligence (AGI) based on the Cattell-Horn-Carroll (CHC) theory of cognition. It breaks AGI into ten measurable cognitive domains and maps each to existing AI benchmarks. Using GPT-4 (2023) and projected GPT-5 (2025) data, the study applies exponential trend extrapolation to predict human-level proficiency across domains. Results show rapid progress in reading, writing, and math by 2028, but major bottlenecks in memory and reasoning until the 2030s. The approach provides a granular, reproducible, and governance-relevant method for tracking AI progress and informing strategic planning.
Reviews
Building upon recent work on “A Definition of AGI” is sensible and timely. However, the project would benefit from more qualitative justifications of some of the predictions, and also from using some other forecasting methodology in addition to trend extrapolation. For example, your model predicts that Long-Term Memory Storage will reach 100% by 2038, despite both GPT-4 and GPT-5 scoring zero on that factor. Why?
Some of the graphs are a bit odd, race to 100% proficiency” - it isn’t aligned with the forecasts listed in the table
The work tries to operationalize CHC domains by mapping them into measurable benchmarks. They then try to forecast when AI would hit the 100% proficiency level for each domain.
I think building it off CHC is a good idea, and it is somewhat fair to weigh each category equally as a starting point. However, I'm not sure what you are doing exactly to do the forecasting: How are you doing exponential trend extrapolation given two data points? The forecasting aspect of this work seems to be making big jumps in logic, and more details (and justification) around implementation details would be nice.
There is also an assumption that the currently existing set of benchmarks is sufficient, and if there exists a model that can solve all reading-related benchmarks that we would deem the model to be human-level at reading. It is often the case that new benchmarks pop up over time measuring a different aspect of a skill, especially if current models are completely hopeless at that aspect, so you may also want to define a % that all benchmarks cover instead of 100%.
Read full reviewShow less
Cite this project
@misc{abdulfatah2025forecasting,
title = {{Forecasting AGI: A Granular, CHC-Based Approach}},
author = {Habeeb Abdulfatah},
year = {2025},
month = nov,
note = {Submitted to The AI Forecasting Hackathon, an Apart Research Sprint},
howpublished = {\url{https://apartresearch.com/sprints/projects/forecasting-agi-a-granular-chcbased-approach-9n7z}},
url = {https://apartresearch.com/sprints/projects/forecasting-agi-a-granular-chcbased-approach-9n7z}
}More from The AI Forecasting Hackathon
- View project: System Dynamics Game-Theoretic Model of the AI Development Race
System Dynamics Game-Theoretic Model of the AI Development Race
System Dynamics BCN
A Game theoretic / System Dynamics model of the race dynamics of the US, China, and EU, as a follow up to the Armstrong et al. (2016) paper “Racing to the Precipice”. We find preliminary results where knowledge of …
- View project: ExogenousAI
ExogenousAI
Fibonacci
Current AI capability forecasting methodologies, including EpochAI's Direct Approach and Biological Anchors framework, primarily rely on internal metrics such as training compute and scaling laws while assuming stable …
- View project: AI Incidents Forecasting
AI Incidents Forecasting
KLACE
This research develops a framework for forecasting AI incidents to help predict future risks. We have developed two models that forecasts incidents which include calibrated 90% prediction intervals with backtests. These …