TLDR: This paper traces the 75-year evolution of artificial intelligence (AI) agents in social and behavioral sciences, from early computer simulations in the 1950s to today’s large language models. It highlights how AI has served as a research tool, an object of study, and a social force, emphasizing the continuous interplay between AI advancements and our understanding of human behavior and society. The review covers key milestones like cybernetics, symbolic and connectionist AI, complexity science, the big data era, and the recent impact of generative AI, while also discussing current trends, challenges, and future directions.
For over 75 years, humanity has pursued the dream of creating machines that can think and behave like us. This journey, deeply intertwined with our quest for self-understanding, has seen artificial intelligence (AI) agents transform the social and behavioral sciences. From the very first programmable computers to today’s advanced large language models, AI has continuously reshaped how we study human society and behavior.
The story begins surprisingly early. Soon after the first programmable computers arrived in the 1950s, scientists began using them to simulate human interactions. These early pioneers, often part of the same intellectual movement that developed computers, saw the potential. They used simulations to explore complex systems, conduct experiments that were otherwise impossible, validate theories, and even forecast future scenarios, like presidential elections.
A parallel intellectual movement, cybernetics, also heavily influenced this period. Cybernetics focused on systems, control, and information transfer, aiming to break down barriers between scientific disciplines. This led to concepts like Jay Forrester’s systems dynamics, which famously predicted the ‘Limits to Growth’ in the early 1970s, highlighting the finite nature of Earth’s resources and influencing today’s sustainability movements.
The academic field of AI officially began in 1956. Early approaches branched into two main directions: symbolic AI, championed by figures like Herbert Simon and Allen Newell, focused on reverse-engineering higher-order cognitive abilities like reasoning and problem-solving, leading to systems like the General Problem Solver and later, commercially successful ‘expert systems’. The other branch, connectionism, explored building machine intelligence based on networks of model neurons, starting with the perceptron and laying the groundwork for today’s neural networks.
Later, ideas like embodied cognition challenged the notion of mind and body being separate, influencing robotics and the concept of ‘autopoiesis’ – the ability of living systems to reproduce themselves. This led to research on socially aware agents and the integration of AI into human populations.
The mid-1980s saw the rise of complexity science, which, while overlapping with systems science, emphasized complex patterns emerging from simple rules. This field was crucial for the development of agent-based models (ABMs), where individual agents with simple rules interact to create complex social phenomena. Examples include the Sugarscape model, which simulated the emergence of ethnicities and trade, and game-theoretic agents used in economics and political science to study cooperation and social dilemmas. The related field of artificial life even explored creating computer programs with properties analogous to living organisms.
The early 2000s ushered in the ‘age of big data’. This era was characterized by the ability to scale existing machine learning algorithms to massive datasets, fueled by advancements in hardware and data collection. Deep learning, with breakthroughs like AlexNet in computer vision and recurrent neural networks in natural language processing, became prominent. The introduction of transformer models in 2017, particularly the ‘Attention is all you need’ paper, was a pivotal moment, paving the way for large language models like GPT and BERT. Reinforcement learning also saw a revival, notably with AlphaGo’s victory over a Go master.
This period also saw the emergence of computational social science, leveraging vast amounts of ‘digital trace data’ from online interactions. Crowdsourcing platforms like Amazon Mechanical Turk and challenges like DARPA’s Red Balloon challenge demonstrated the power of human-computer collaboration. The big data era also sparked a debate about the role of prediction versus explanation in science, with some even foreseeing an ‘end of theory’.
The most recent chapter, the ‘age of generative AI’, began dramatically with the release of ChatGPT in November 2022. This marked a shift towards AI capable of generating coherent and useful content, making AI’s capabilities more accessible and seemingly human-like. Today, generative AI, including LLMs and diffusion models, is being combined to create multimodal AI.
Current research trends with generative AI are diverse. Scientists are using chatbots to replace humans in experiments and surveys, documenting AI’s capabilities and exploring emergent sociality among AI agents. Others are studying human-AI interactions, examining how AI influences opinions or changes human behavior in collaborative settings. AI is also improving analysis tools, offering cheap and accessible ways to perform text and image analysis, sentiment analysis, and even infer social networks. Some ambitious projects are even exploring how chatbots can replace scientists for specific tasks or even the entire scientific process.
Despite this fervor, a surprising gap remains: studies focusing on understanding humans per se using AI are rare. Many current AI studies serve more as pilot investigations for future human-centric research. Furthermore, generative AI comes with its own set of challenges, including replication difficulties due to rapid development and inherent stochasticity, potential failures of safety guardrails, and the risk of data degradation as AI-generated content increasingly feeds back into training datasets.
Also Read:
- NegotiationGym: A Framework for Self-Optimizing AI in Social Simulations
- Navigating the Mind’s Theories: A Comparative Look at Cognitive and Generative AI
The journey of AI in social and behavioral sciences is a testament to the deep connection between humans and the technologies we create. As AI continues to evolve, it offers unprecedented opportunities to understand ourselves and the social world, even as it simultaneously reshapes society and human behavior. To delve deeper into this fascinating history and outlook, you can read the full paper here.
