Emergence of “A Sense of Beauty” in Untrained AI
Abstract from a new paper titled ‘The spontaneous emergence of “a sense of beauty” in untrained deep neural networks’:
The sense of facial beauty has long been observed in both infants and nonhuman primates, yet the neural mechanisms of this phenomenon are still not fully understood. The current study employed generative neural models to produce facial images of varying degrees of beauty and systematically investigated the neural response of untrained deep neural networks (DNNs) to these faces. Representational neural units for different levels of facial beauty are observed to spontaneously emerge even in the absence of training. Furthermore, these neural units can effectively distinguish between varying degrees of beauty. Additionally, the perception of facial beauty by DNNs relies on both configuration and feature information of faces. The processing of facial beauty by neural networks follows a progression from low-level features to integration. The tuning response of the final convolutional layer to facial beauty is constructed by the weighted sum of the monotonic responses in the early layers. These findings offer new insights into the neural origin of the sense of beauty, arising the innate computational abilities of DNNs. (PsycInfo Database Record (c) 2024 APA, all rights reserved)
If (big IF) this is true and the DNNs are not just selecting for symmetry, it could probably also apply to aesthetic appreciation of cultural objects such as fine art, watches, etc. Obviously the 3 dimensional nature of classic cars, sculpture, jewelry and watches would pose a challenge, but given the pace of development, I am optimistic that this too will be solved. It would be interesting to see what types of cultural objects DNNs will find beauty in, if they differ from humans, and whether humans or AI agents can influence each other.
If you believe that AI agents (and humanoid robots) will be able to operate autonomously, it is not implausible that one day, we will task an AI agent to source and build a collection of cultural objects. These ‘buyer AI agents’ would tirelessly scan auctions, marketplaces (such as Chrono24 for watches, Artsy for art, etc), dealer websites, private chat groups, etc. They would autonomously identify, select, perform due diligence1, price, negotiate and even make payments and arrange shipping. In some cases physical delivery would not even be necessary - purchasing an NFT will suffice, while the physical item remains in a safe location. The heterogeneity problem will remain, and those AI agents that have the necessary training to source professionally (including assessing authenticity, condition, repairability, etc.) will be in demand. And most importantly, if a ‘buyer AI agent’ innately possesses the ability to appreciate aesthetics, it may be able to build collections that surpass human curation and/or even start trends.
And it wouldn’t be one-sided. Sellers would also have ‘seller AI agents’ that attempt to sell to both humans and ‘buyer AI agents’. In addition to excelling at marketing and negotiation, ‘seller AI agents’ would even be able to quickly generate TikTok/Instagram marketing videos, auction catalogues, mock-ups, lookbooks, etc. Most importantly, ‘seller AI agents” would be able to appreciate and communicate the aesthetic beauty of the cultural objects they are selling. This is something that today we casually refer to as ‘passion’ and is essential when selling and building clientele for art, watches, cars, etc.
Looking very far into the future, AI agents and their derivatives may not even need to be “tasked” by a human to build a collection of human cultural objects. An innate ability to appreciate the aesthetic beauty in cultural objects could lead to AI agents collecting them on their own volition, and pay for them with crypto. “I sold my Patek 3587 to an AI, they love old quartz” may not be as farfetched as it sounds today. And perhaps one day ‘AI agent artists’ will be able to develop their own cultural objects.
The above all sounds very futuristic and cool, but I’ve clickbaited it out of context. I have no clue if art-dealing AI agents are even remotely possible. The real purpose of the paper was to use DNNs to understand the human neural origin of a sense of beauty, not the DNNs themselves. How and why does the human brain compute the perceived aesthetic value of complex stimuli such as art, watches, and other cultural objects?
Overall, the results of this study fill a gap in the research on the neural computation of intrinsic aesthetic perception, revealing the hierarchical processing pattern of facial beauty by untrained neural networks. Aesthetics is a crucial part of knowledge acquisition (Perlovsky, 2014; Perlovsky & Schoeller, 2019; Sarasso et al., 2020), where individuals, while appreciating a piece of art, gather meanings and engage in learning (Tracy, 1946). Aesthetic appreciation triggers the instinct to seek knowledge, enabling individuals to gather information and learn from their environment (Schoeller & Perlovsky, 2016; Schoeller et al., 2018). Our research findings indicated that a sense of beauty can emerge without prior learning, potentially supporting an innate sense of aesthetic perception, suggesting that humans might have an inherent ability to learn from aesthetics.
While our study was not a substitute for neuroimaging studies, it bore a striking resemblance to functional neuroimaging studies of adult aesthetics of artworks (Iigaya et al., 2023). This suggests that the emergence of initial beauty sense has a similar neural mechanism to the formation of adult aesthetic value[…] Due to the limitations of existing technology in directly studying the neural processing patterns of human intrinsic aesthetic perception, this study, conducted using biologically inspired DNN models, offers a neuro-computational perspective on the inherent developmental patterns of aesthetics. It provides an interpretable approach to understanding the mechanisms underlying aesthetic value generation.
As I wrote in this post, a rising market attracts smart, dedicated, and well educated collectors and scholars that cooperate, compile, discover and document knowledge. Perhaps rising prices are bounties, bringing in even more people to act on instincts to seek knowledge of beautiful cultural objects and the historical context in which they were created. And it is probably our duty to enhance and maintain this knowledge, regardless of whether future enthusiasts are humans or AI agents.
We marvel at a future of sleek robots and all-knowing AI improving productivity and eliminating drudgery from our lives. But the most profound impact of AI agents and AI development in general may be a better understanding of ourselves.
Either via photos/asking questions or outsourcing to humans