Transformation vs Tradition: Artificial General Intelligence (AGI) for Arts and Humanities

Zhengliang Liu, Yiwei Li, Qian Cao, Junwen Chen, Tianze Yang, Zihao Wu, John Hale, John Gibbs, Khaled Rasheed, Ninghao Liu, Gengchen Mai, Tianming LiuarXiv 2023 · DOI · OpenReview

Abstract

The rapid advancement of Artificial General Intelligence (AGI), particularly Large Language Models (LLMs) and diffusion-based generators, is fundamentally reshaping the arts and humanities. This paper surveys the transition from narrow algorithmic tasks to sophisticated multimodal systems, exploring the "capability gap" and the "responsibility gap." We advocate for a human-in-the-loop framework to address inherent risks such as bias and hallucination while leveraging AGI to augment human flourishing.

Introduction: The Final Fortress

For decades, the arts and humanities were viewed as the final fortress of human exclusivity—realms defined by subjectivity, emotion, and non-linear creativity. However, the emergence of AGI has blurred the boundary between human intuition and machine computation. This paper addresses how we navigate this transition, moving from "narrow AI" that performs specific tasks to systems capable of synthesizing literature, analyzing history, and generating high-fidelity visual art.

Technical Foundations: Text and Graphics

The evolution of AGI is rooted in several key architectural shifts. In the realm of text, the lineage has moved from bidirectional encoding (BERT) to autoregressive models like the GPT family. The core of this progress is the self-attention mechanism in Transformers, which enables the parallel processing of long sequences and the understanding of complex context.

Figure 6: An LLM synthesizing historical data into a structured summary, demonstrating its utility in the humanities.

In graphics, the field has progressed from Generative Adversarial Networks (GANs) and Variational Autoencoders (VAEs) to state-of-the-art Diffusion Models. Latent Diffusion Models (LDMs), such as Stable Diffusion, operate in a compressed "latent space," allowing for high visual fidelity with significantly reduced computational overhead compared to earlier methods.

The Multimodal Leap and Zero-Shot Reasoning

One of the most significant breakthroughs is the rise of multimodal systems like GPT-4V and LLaVA. These models demonstrate "zero-shot" capabilities—performing tasks without specific fine-tuning. For instance, AGI can now extract place names from 19th-century maps or provide architectural critiques, showcasing a level of spatial reasoning and "common sense" previously reserved for human experts. This cross-domain utility is further evidenced by GPT-4’s ability to analyze complex, multilingual song lyrics to extract abstract themes like "unity."

Key Insights: From Assistant to Creative Partner

AGI is transitioning from a simple "intelligent assistant" used for drafting to a "creative partner" capable of reinterpreting historical narratives. This shift facilitates "evolutionary creativity," where algorithms can generate novel artistic forms inspired by Darwinian principles. The market has already begun to validate this shift, notably with the 2018 sale of a GAN-generated portrait at Christie’s for over $432,000.

Figure 4: A stylized depiction of a robot as a domestic partner, reflecting the conceptual shift from AI as a tool to an integrated partner.

However, the authors posit a "Human Flourishing" framework: the goal is not to replace the artist but to augment human potential. While AI can replicate complex patterns, it currently lacks the lived experience and emotional depth that drive true innovation. The well-known "hand problem"—the struggle of diffusion models to render human fingers correctly—serves as a metaphor for the broader "factuality" problem: high-fidelity output does not guarantee underlying truth.

Challenges and the Responsibility Gap

The democratization of these tools brings significant risks. LLMs are prone to "negligent falsehoods" or hallucinations, which are particularly dangerous in historical or legal contexts. Furthermore, there is a notable "geopolitical favoritism" in training data, where wealthier nations are over-represented, potentially leading to cultural erosion and the concentration of power within a few technology giants.

Conclusion

The frontier of AGI in the arts and humanities is both promising and perplexing. To ensure these technologies align with enduring human values, multi-stakeholder collaboration is essential. By developing robust toxicity filters and bias detectors, we can move toward a future where AGI serves as a powerful catalyst for cultural expression and historical insight rather than a replacement for human agency.

Summary updated

Abstract

Recent advances in artificial general intelligence (AGI), particularly large language models and creative image generation systems have demonstrated impressive capabilities on diverse tasks spanning the arts and humanities. However, the swift evolution of AGI has also raised critical questions about its responsible deployment in these culturally significant domains traditionally seen as profoundly human. This paper provides a comprehensive analysis of the applications and implications of AGI for text, graphics, audio, and video pertaining to arts and the humanities. We survey cutting-edge systems and their usage in areas ranging from poetry to history, marketing to film, and communication to classical art. We outline substantial concerns pertaining to factuality, toxicity, biases, and public safety in AGI systems, and propose mitigation strategies. The paper argues for multi-stakeholder collaboration to ensure AGI promotes creativity, knowledge, and cultural values without undermining truth or human dignity. Our timely contribution summarizes a rapidly developing field, highlighting promising directions while advocating for responsible progress centering on human flourishing. The analysis lays the groundwork for further research on aligning AGI's technological capacities with enduring social goods.
BibTeX
@misc{liu2023transformation,
  author = {Zhengliang Liu and Yiwei Li and Qian Cao and Junwen Chen and Tianze Yang and Zihao Wu and John Hale and John Gibbs and Khaled Rasheed and Ninghao Liu and Gengchen Mai and Tianming Liu},
  title = {{Transformation vs Tradition: Artificial General Intelligence (AGI) for Arts and Humanities}},
  year = {2023},
  eprint = {2310.19626},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2310.19626},
}
Figure