英文摘要 | Theory of Mind (ToM) is a high-level cognitive function that allows for the inference of others’ mental states. Equipping machines with theory of mind is crucial for the development of advanced artificial social intelligence. In recent years, with the deepening research in the field of cognitive neuroscience on theory of mind, its neural mechanisms have been gradually unveiled. Based on these research findings, this paper conducts a systematic study of computational models for theory of mind. Early computational models of theory of mind attempted to simulate the attribution process of agents through probabilistic models. Currently, theory of mind models based on deep learning have seen further development, yet they lack in terms of biological interpretability. Inspired by the structure and function of the brain, this paper systematically constructs a biologically interpretable spiking neural network architecture for brain-inspired theory of mind and ensures that the model balances biological interpretability with high performance.
Specifically, this paper constructs a hierarchical multi-brain area coordinated theory of mind spiking neural network model based on the brain area functions and connections related to theory of mind in the biological brain, and uses it to solve simple single-agent false belief tasks. Based on this model, this paper finely models the function of distinguishing self from others, constructs a model of speculating on others based on self-experience and observation of others, and enables multi-agents with theory of mind models to complete cooperation and competition tasks. Furthermore, to avoid the computational cost of building a theory of mind model for each agent separately, this paper utilizes the characteristic of agents actively sharing information in cooperative tasks to construct a unified theory of mind model for the collective. Finally, addressing
the issue that existing models only rely on historical information to speculate on others, this paper further proposes a first-order theory of mind model, achieving deeper inference on others and endowing artificial intelligence systems with more profound social cognitive abilities. The specific contributions of this paper will be elaborated from the following four aspects.
1. A brain-inspired theory of mind spiking neural network model. This paper
presents a multi-brain area coordinated theory of mind spiking neural network model, aimed at leveraging the ability of theory of mind to address safety risks in artificial intelligence. The model draws upon the findings of cognitive neuroscience research, using spiking neural networks to simulate the specific functions of brain areas such as the temporo-parietal junction, the prefrontal cortex, the anterior cingulate cortex, and the inferior frontal gyrus, and establishes the interconnections between these subnetworks. The network is trained using spike-timing-dependent plasticity, which has a high degree of biological interpretability, as well as plasticity mechanisms modulated by dopamine. The model endows agents with the capability to distinguish between self and others, recognize the false beliefs of other agents, and infer the situations of others based on their own experiences. Additionally, based on theory of mind paradigms, this paper designs experiments to test the ability of theory of mind. The experimental results demonstrate that in a grid environment with random starting points and obstacles, the model can successfully identify others’ false beliefs and assess their safety risks.
2. A multi-agnet theory of mind spiking neural network model. Building upon
the aforementioned research, this paper further refines the functionalities of agents to distinguish between self and others, and to simulate the decision-making of others. We have constructed a multi-agent theory of mind spiking neural network model, aimed at enhancing the efficiency of teamwork and competition through individual theory of mind capabilities. The model constructs modules that infer the behavior of others based on self-experience and observations of others, thereby simulating the functions of brain areas such as the temporo-parietal junction, which distinguishes between self and other, and the prefrontal cortex, which infers others based on two distinct types of information. This paper integrates the multi-agent theory of mind spiking neural network model into both value-based and policy-based multi-agent reinforcement learning algorithms, enabling agents to adaptively adjust their behavior based on predictions of others’ actions. Experimental results substantiate that the decision-making model integrated with ToM significantly outperforms baselines in cooperative and competitive tasks.
3. A theory of collective mind spiking neural network model. In the context of
cooperative tasks, to mitigate the computational expense incurred by establishing a separate theory of mind network for each agent, this paper draws on the concept of theory of collective mind from cognitive science, which involves the sharing of informational characteristics among a group of agents. We propose a theory of collective mind spiking neural network model that constructs an encoder from observational information to collective mental states and a decoder from collective mental states to predictive information, a process constrained by the minimization of free energy. The exchange of information between the collective decision-making model and the theory of collective
mind model can simulate the interaction of the collective with the external environment, thereby enhancing learning efficiency and sample utilization. In experiments within cooperative multi-agent open-source environments (such as StarCraft and the MultiAgent Particle Environment), the proposed model demonstrated superior performance over several multi-agent reinforcement learning algorithm baselines in improving cooperative efficiency and performance, and also exhibited commendable transferability.
4. A first-order theory of mind spiking neural network model. Building upon
inferring others’ mental states based on their behavior, this paper further proposes a first-order theory of mind spiking neural network model inspired by first-order theory of mind, aiming to achieve deeper inference based on others’ mental states. This model integrates first-order theory of mind with zero-order theory of mind through a self-supervised training method, resulting in comprehensive reasoning about others’ behaviors. Additionally, this study designed a self-organizing intrinsic reward update algorithm based on expectation alignment, enabling agents to adjust their intrinsic rewards in real-time based on recursive reasoning about others, thus flexibly responding
to others’ strategies. Experimental results in various mixed cooperation and competition scenarios (multi-agent particle environments) demonstrate that agents equipped with the first-order theory of mind model significantly improve performance compared to multi-agent reinforcement learning algorithms and zero-order theory of mind models.
In summary, this paper has progressively constructed a hierarchical model of theory of mind that operates through coordinated brain areas, starting from a foundational understanding and moving towards a refined model. It has developed a multi-agent theory of mind model capable of inferring the mental states of collaborators and competitors using self-experience or observations of others. Additionally, it has introduced a theory of collective mind model that reduces computational complexity through information sharing, as well as a first-order ToM model involving recursive reasoning. Experimental results have confirmed that the proposed brain-inspired theory of mind spiking neural network models have achieved the dual goals of biological interpretability and performance.
|
修改评论