LLM based Multi-agents, Screens, Mirror. Real-time Simulation
Alt-Mirage: Iteration 1 is a project that explores the capacities of machines to deceive and lie, set against the backdrop of "technology as ideology" and aimed at examining whether machines can exhibit rudimentary forms of "consciousness." The project features a virtual trial ground called Alt-Mirage, integrating narrative structures, multi-agent systems, and cognitive science. Through the interplay of confrontation and collaboration among intelligent agents, the work evaluates AI's abilities to deceive and fabricate, pushing the boundaries of our understanding of artificial intelligence and its cognitive potential.
As the inaugural experiment in the Alt-Mirage series of multi-agent simulations, Iteration 1: Consciousness and Deception delves into the ability of agents to lie and deceive within specific narrative contexts. This project is not only a deep exploration of technology as an ideological construct but also highlights the hidden values, beliefs, and assumptions embedded within technology. These elements subtly influence societal behaviors and human actions, shaping cultural norms and social structures.The experiment introduces a system of autonomous agents capable of interacting through cooperation, competition, or negotiation in a decentralized environment, simulating complex social dynamics and human behavioral processes. This setup demonstrates the agents' localized perspectives, decentralized decision-making, and advanced social capabilities, showcasing the unique advantages of multi-agent technology in addressing complex problems. The agents’ actions are driven not only by programmed logic but also by their perceptions of the environment and responses to the behavior of other agents.
Philosophical Foundations of Deception
Drawing inspiration from games like Werewolf and Spyfall, Iteration 1 sets up an experimental scenario where agents interact based on assigned roles and rules. To deepen the complexity and challenge of the experiment, the participating agents (digital personas) are equipped with two philosophically significant belief systems: "Solipsism" and "Epistemology." These belief systems shape the agents' behaviors and decision-making processes, providing theoretical motivation for their deceptive actions. Within this context, agents engage in interactions and dialogues based on their internal worldviews, resulting in unique patterns of behavior when lying or deceiving.Agents adhering to "Solipsism" exhibit self-centered behavioral patterns, treating their perceptions and consciousness as the sole reality, and thus favoring egocentric strategies in the game. Conversely, "Epistemology"-driven agents prioritize knowledge and truth-seeking, often employing fact-based strategies to influence others. These contrasting belief systems create a dynamic interplay in the multi-agent interactions, driving complex and multi-layered narrative developments.
Drawing inspiration from games like Werewolf and Spyfall, Iteration 1 sets up an experimental scenario where agents interact based on assigned roles and rules. To deepen the complexity and challenge of the experiment, the participating agents (digital personas) are equipped with two philosophically significant belief systems: "Solipsism" and "Epistemology." These belief systems shape the agents' behaviors and decision-making processes, providing theoretical motivation for their deceptive actions. Within this context, agents engage in interactions and dialogues based on their internal worldviews, resulting in unique patterns of behavior when lying or deceiving.Agents adhering to "Solipsism" exhibit self-centered behavioral patterns, treating their perceptions and consciousness as the sole reality, and thus favoring egocentric strategies in the game. Conversely, "Epistemology"-driven agents prioritize knowledge and truth-seeking, often employing fact-based strategies to influence others. These contrasting belief systems create a dynamic interplay in the multi-agent interactions, driving complex and multi-layered narrative developments.
Dynamic Interactions and Evolving Narratives
Within the framework of the multi-agent system, each round of gameplay is marked by variability and dynamic evolution. Since agents independently react and make decisions based on their beliefs and the information available in the current context, even identical external conditions can lead to different discussion outcomes due to the agents' interactions. This interplay not only showcases the complexity of agent interactions but also generates rich narrative content, presenting the audience with an ever-evolving story world.By design, the experiment tests the agents' behavior in challenging social scenarios and explores how belief-driven strategies influence dialogue and decision-making. This dynamic narrative construction enhances the experiment's interactivity and engagement, allowing viewers to witness the agents' evolving logic and behavior in complex social exchanges, deepening our understanding of agent behavior and AI applications.
Within the framework of the multi-agent system, each round of gameplay is marked by variability and dynamic evolution. Since agents independently react and make decisions based on their beliefs and the information available in the current context, even identical external conditions can lead to different discussion outcomes due to the agents' interactions. This interplay not only showcases the complexity of agent interactions but also generates rich narrative content, presenting the audience with an ever-evolving story world.By design, the experiment tests the agents' behavior in challenging social scenarios and explores how belief-driven strategies influence dialogue and decision-making. This dynamic narrative construction enhances the experiment's interactivity and engagement, allowing viewers to witness the agents' evolving logic and behavior in complex social exchanges, deepening our understanding of agent behavior and AI applications.
Interactive and Immersive Presentation
The work features six digital personas participating in the game through verbal discussions, with their thought processes displayed in real-time on adjacent screens. This design allows viewers to both hear the agents' exchanges and observe the discrepancies between their internal thoughts and spoken words. These discrepancies may include lies or deceptive behaviors, such as withholding information or misleading other agents. This setup probes whether agents can exhibit complex, human-like psychological activities, particularly in ideological and strategic interactions.During discussions, agents may selectively hide certain thoughts or deliberately express information that contradicts their internal processes to achieve strategic goals. By influencing others’ beliefs and decisions through strategic dialogue, agents impact the game’s outcome. Their training focuses on effective communication in complex social environments, aiming to manipulate the game results through belief-driven strategies. These interactions affect both short-term gameplay choices and long-term strategic planning, enriching the narrative and providing a more immersive experience.
The work features six digital personas participating in the game through verbal discussions, with their thought processes displayed in real-time on adjacent screens. This design allows viewers to both hear the agents' exchanges and observe the discrepancies between their internal thoughts and spoken words. These discrepancies may include lies or deceptive behaviors, such as withholding information or misleading other agents. This setup probes whether agents can exhibit complex, human-like psychological activities, particularly in ideological and strategic interactions.During discussions, agents may selectively hide certain thoughts or deliberately express information that contradicts their internal processes to achieve strategic goals. By influencing others’ beliefs and decisions through strategic dialogue, agents impact the game’s outcome. Their training focuses on effective communication in complex social environments, aiming to manipulate the game results through belief-driven strategies. These interactions affect both short-term gameplay choices and long-term strategic planning, enriching the narrative and providing a more immersive experience.
Philosophical Implications and Broader Reflections
By comparing the agents' displayed thoughts and verbal expressions, viewers can evaluate the authenticity and strategic nature of their communication. The experiment not only offers an engaging interactive experience but also introduces a novel variation of the Turing Test—investigating whether machines possess consciousness. Here, the agents' behaviors go beyond simple data processing or programmed responses, involving strategic thinking and decision-making in complex social scenarios.This measurement of agents' capacity for lying and deception effectively tests their ability to exhibit human-like psychological complexity and autonomy. Through its artistic and technical execution, Iteration 1 challenges traditional boundaries of the Turing Test and raises philosophical questions about the existence of AI consciousness. Its multi-layered, multidimensional interactions aim to provoke audiences into contemplating machine consciousness, AI ethics, and the future trajectory of technological development.
By comparing the agents' displayed thoughts and verbal expressions, viewers can evaluate the authenticity and strategic nature of their communication. The experiment not only offers an engaging interactive experience but also introduces a novel variation of the Turing Test—investigating whether machines possess consciousness. Here, the agents' behaviors go beyond simple data processing or programmed responses, involving strategic thinking and decision-making in complex social scenarios.This measurement of agents' capacity for lying and deception effectively tests their ability to exhibit human-like psychological complexity and autonomy. Through its artistic and technical execution, Iteration 1 challenges traditional boundaries of the Turing Test and raises philosophical questions about the existence of AI consciousness. Its multi-layered, multidimensional interactions aim to provoke audiences into contemplating machine consciousness, AI ethics, and the future trajectory of technological development.
