GATE: An Integrated Assessment Model for AI Automation
By Ege Erdil, Andrei Potlogea, Tamay Besiroglu, Edu Roldan, Anson Ho, Jaime Sevilla, Matthew Barnett, Matej Vrzla, Robert Sandler
DOI https://doi.org/10.48550/arXiv.2503.04941
Abstract
Assessing the economic impacts of artificial intelligence requires integrating insights from both computer science and economics. We present the Growth and AI Transition Endogenous model (GATE), a dynamic integrated assessment model that simulates the economic effects of AI automation. GATE combines three key ingredients that have not been brought together in previous work: (1) a compute-based model of AI development, (2) an AI automation framework, and (3) a semi-endogenous growth model featuring endogenous investment and adjustment costs. The model allows users to simulate the economic effects of the transition to advanced AI across a range of potential scenarios. GATE captures the interactions between economic variables, including investment, automation,innovation, and growth, as well as AI-related inputs such as compute and algorithms. This paper explains the model’s structure and functionality,emphasizing AI development for economists and economic modeling for the AI community. The model is implemented in an interactive sandbox, enabling users to explore the impact of AI under different parameter choices and policy interventions.
The modeling sandbox is available at https://epoch.ai/gate
Overview
The paper introduces the Growth and AI Transition Endogenous (GATE) model, an integrated assessment tool designed to simulate and analyze the economic impacts of advanced artificial intelligence (AI) automation. GATE brings together ideas from computer science and economics to help researchers and policymakers understand how AI developments can influence key economic factors such as output, consumption, investment, and labor markets.
Key Components of the GATE Model
GATE is built on three main modules, each representing a step in the process by which AI influences the economy:
- AI Development Module
- Purpose: Converts money spent (investment) into “effective compute” (a measure of computing power available for AI training and operations).
-
How It Works:
- Hardware and Software R&D: Investment is divided between improving hardware (like chips and data centers) and developing better software (such as more efficient training algorithms).
- Effective Compute: This concept, measured in “effective FLOPs” (floating-point operations), summarizes how much useful computing power is available for AI systems.
- Why It Matters: By modeling the conversion of dollars into compute power, the model explains the engineering process behind AI progress in simple, quantitative terms.
-
AI Automation Module
- Purpose: Translates the available effective compute into the automation of labor tasks.
-
How It Works:
- Training and Inference: The module considers two phases:
- Training: Uses compute to improve AI capabilities (e.g., learning to perform new tasks).
- Inference: Uses compute to deploy AI systems, increasing the “digital worker” capacity.
- Margins of Automation:
- Extensive Margin: Determines which tasks become automated.
- Intensive Margin: Determines how efficiently tasks are automated (i.e., more output per digital worker).
- Labor Impact: The module explores scenarios ranging from smooth reallocation (where displaced human workers move to other tasks) to complete displacement (where workers exit the labor force).
-
Macroeconomic Module
- Purpose: Integrates AI automation with the broader economy.
-
How It Works:
- Standard Growth Framework: Uses a model similar to the Ramsey-Cass-Koopmans framework, where a “social planner” (a theoretical decision-maker) chooses the optimal mix of consumption, saving, and investment.
- Production Function: Combines contributions from automated (AI) labor, human labor, and capital (both accumulable, like machinery, and non-accumulable, like natural resources).
- Feedback Loop: Economic growth increases available resources, which in turn can drive further investment in AI development.
- Outcome: The model shows how automation feeds back into economic performance, potentially accelerating growth under certain conditions.
Optional Add-ons
To enhance its flexibility, GATE includes two optional modules:
-
R&D Externalities Module:
Function: Accounts for “positive externalities” where the benefits of research and development (R&D) spill over to the wider society.
Impact: Demonstrates how individual investment decisions might lead to underinvestment in AI if the broader benefits are not fully captured by the investors.
-
Uncertainty Module:
Function: Introduces uncertainty in the mapping between compute investment and automation capabilities.
Impact: Reflects how investor expectations and real-world outcomes may differ, leading to more cautious investment behavior in AI development.
Main Findings and Implications
Interdisciplinary Approach: GATE bridges technical AI development and economic growth theories. It translates engineering progress (measured in effective compute) into tangible economic outcomes.
Feedback Loops: The model captures how increased AI automation not only boosts current output but also feeds back into further investments in AI, creating a reinforcing cycle that may accelerate economic growth.
Economic Impact Scenarios: By allowing users to change key parameters (such as investment returns, capital adjustment frictions, or the degree of task automation), the model can simulate various scenarios. For example, a higher investment in hardware R&D might lead to more effective compute, which in turn could increase automation and drive economic expansion.
-
Limitations Acknowledged:
-
Simplified Assumptions: The paper notes that the model simplifies some real-world complexities, such as:
- The role of data quality and availability in AI development.
- A basic view of the labor “task space,” which does not fully capture the diversity or complexity of human tasks.
- The exclusion of non-AI related productivity improvements.
-
Future Improvements: The authors emphasize the need for ongoing updates and invite contributions from the research community to refine the model.
Conclusion
In summary, GATE is presented as an initial but comprehensive attempt to quantify the economic effects of AI automation. It combines a detailed, compute-based model of AI development with a framework that translates technological advancements into economic output. The key takeaways are:
- Integrated Perspective: GATE connects investment in AI, the technical progression of AI systems, and macroeconomic outcomes, making it a valuable tool for both economists and AI researchers.
- Policy and Research Utility: The interactive simulation platform (available at www.epoch.ai/GATE) enables users to explore different scenarios, which can inform policy decisions and further academic research.
- Call for Collaboration: The paper concludes by acknowledging its limitations and encouraging further work to refine the assumptions and extend the analysis, especially as new data and insights into AI’s impacts become available.
Example to Illustrate the Process
Imagine an economy where more money is invested in building better computer chips and developing advanced training algorithms:
- Step 1: Increased investment leads to more effective compute (better, faster processing power).
- Step 2: With more effective compute, AI systems can automate a wider range of tasks, such as data analysis or customer service.
- Step 3: As automation increases, overall production grows because tasks are performed more efficiently.
- Feedback: The resulting economic growth then frees up additional resources, encouraging even more investment in AI—thus creating a positive feedback loop.
This example encapsulates how GATE links technical advancements in AI directly to economic growth.