Symbolic representations stand at the crossroads of human cognition and artificial intelligence (AI). The process by which humans naturally abstract complex systems into simpler, symbolic interpretations has long been a topic of intrigue. It’s the cognitive mechanism that lets us see patterns, make predictions, and form a coherent understanding of the world around us. For AI, achieving a similar level of abstraction has been a significant challenge. This project aims to bridge this gap. The goal of this endeavor is to implement and expand upon a novel symbol learning method, referred to as DeepSym. The essence of DeepSym is its unique approach to learning: it interacts with an environment and gathers state-action-effect tuples. This data becomes the training bedrock for a sophisticated deep encoder-decoder network equipped with binary units, representing our symbolic abstractions. This method not only has the potential to revolutionize how AI systems perceive and interact with their environments but also aligns with the broader ambition of neurosymbolic computing. Neurosymbolic approaches aspire to blend the raw computational power of neural networks with the logical precision of symbolic systems. It’s an integration of low-level sensory input processing with high-level logical reasoning, aspiring to mirror human cognition. The allure of this project is further magnified by its applicability in diverse domains. One such domain, as mentioned, is the vast and intricate world of Minecraft. Its open-ended nature makes it an ideal testing ground for DeepSym, potentially revealing both the strengths and areas of improvement for the method. Furthermore, the project promises substantial academic contributions, with prospects of evolving into a conference paper, signaling its significance in the academic and AI research communities.

In the ambitious realm of “Learning Symbolic Representations from Unsupervised Interactions with the Environment,” this project is strategically poised with several clear-cut objectives. Initially, the task is to not only implement the renowned DeepSym methodology but also to enhance its capabilities, ensuring it remains at the forefront of symbol learning. Recognizing the boundless potential of real-world simulations, the project endeavors to immerse DeepSym within the vast, open-ended world of Minecraft, utilizing established platforms such as Malmo [2] and MineRL [3] to foster this experimental integration. Diving deeper, there’s a commitment to iterate and tinker with diverse network architectures, a step pivotal for unveiling the myriad symbols that may arise from such interactions. Beyond the technological milestones, there’s a parallel academic aspiration: to distill the insights, breakthroughs, and knowledge from this research into a comprehensive form, aiming for the esteemed recognition of a conference paper. In essence, these objectives serve as the project’s compass, guiding it through both uncharted territories and established landscapes. In short, the project sets out to:

• Implement and enhance the DeepSym method.

• Conduct experiments in the Minecraft environment using platforms like Malmo [2] or MineRL [3].

• Modify and experiment with various network architectures to identify emerging symbols.

• Achieve results worthy of being transformed into a conference paper.