Another Door of Perception
A theory called the “drunken monkey hypothesis[1]” mentions that humans are keen drinkers, on the one hand, probably due to the human ancestor apes, in the transition from living arboreal to terrestrial, evolved to save their lives in order to tolerate the alcohol contained in the fallen ripe fruits; on the other hand, generally speaking, ripe fruits with the smell of ethanol also indeed prove their high sugar content and could strongly provide the apes with the energy they needed to survive. Then in the process of natural selection, apes gradually became accustomed to ingesting a certain amount of alcohol due to the novel sensory experience they got with alcohol. Chimps in West Africa also have been seen stealing sap fermented in barrels by humans[2], and the reason for this has been speculated by some scholars to be due to apes’ special preference for alcohol. Today we can almost say that the ape species taking in substances doing something towards their neuro was far no longer a survival mechanism, but a mania of being spiritually fucked.
At the same time, another deadly temptation that is even crazier keeps appearing in every slot in the history of mankind: drugs. It is not difficult to conclude that drugs have a far more bizarre effect on human nerves than alcohol. Maybe alcohol just dulls your mind or reveals your nature, but drugs are literally destroying everyone who gets close to them. But even so, why do humans remain tirelessly faithful to these poisons? It is exactly because the sensory experience of taking drugs is so intense that it often even leads the user to another “wonderland”. The famous English writer Aldous Huxley, author of “Brave New World,” also wrote a book called “The Doors of Perception”. The book recorded Huxley’s sensory experience after taking the psychedelic drug “Mescaline” extracted from certain cactus species.
Here are some extractions from the book:
Visual impressions are greatly intensified and the eye recovers some of the perceptual innocence of childhood, when the sensum was not immediately and automatically subordinated to the concept. Interest in space is diminished and interest in time falls almost to zero.
Though the intellect remains unimpaired and though perception is enormously improved, the will suffers a profound change for the worse[3].
These explicitly conform with what we expected. The point is that these addictive, nerve-damaging psychedelic drugs enhance the human senses. This kind of augmentation can be observed in many ways, as mentioned in the book: the seen things become simple. Also, you may “wrongly” extract the characteristics of what is in front of your eyes, and this may be the reason for some of the so-called “hallucinations”. Some studies say that hallucinogens slow down the transmission of neurotransmitters and significantly increase the secretion of dopamine and 5-hydroxytryptamine[4], leading to the block of the correct connection between your sensory distortion caused by the drugs and previous experiences, but the truth is that no one has been able to explain what is going on in your brain when you hallucinate.
Ok. I assume you are already quite confused about what I laboriously talked about using several paragraphs. Now you are allowed to think back to the AI, or specifically, the neural network. Exactly, the neural network is just like our neural system, a black box. According to this, Google’s project initially released in 2015 called DeepDream[5] offered some heuristics. DeepDream is a computer vision program that uses a convolutional neural network to find and enhance patterns in images via algorithmic pareidolia, thus creating a dream-like psychedelic appearance in the deliberately over-processed images (extracted from Wikipedia). This project is undoubtedly fun, but also it was not implemented only for fun.
Even though deep neural networks have evolved promisingly over the decades, even surpassing human accuracy in image recognition a few years ago, what exactly happens during the whole process and what role each layer of the neural network plays is still as mysterious as ever. DeepDream was implemented to get a glimpse of something that has taken place inside the black box.
So far, we may be able to connect the dots: human irresistibility to hallucinogens; the bizarre dream world; and the abstraction of features by neural networks. By simulating discrete neural electrical signals into continuous information that can be backpropagated, and then by artificially tuning the neural network, we may be able to understand the logic and mechanisms implicit when humans sense. Isn’t it a heartening trend that from the 1950s, connectionists started taking lessons from the great human brain, and now brain science or cognitive science may even, in turn, take possibilities from neural networks?
Not only for these rational aspects that are technically and scientifically inclined, AI also provides inspiration for a more emotional area of humanity — art — that definitely cannot be ignored. Let’s first discuss the relationship between AI and art from a more intuitive perspective: can AI-generated mediums such as images, music, or videos be considered art? This is a question that has been so much debated. But yeah, it is inevitable. After all, at this stage, the rise of AI-generated art is still treated as an extremely new concept, and the public, accustomed to experiencing the traditional art forms, will always think that AI-generated art does not require the artist to have the technical skills actually required by the art form being created. There is a common perception that AI-generated art is largely based on cold mechanical-like algorithms and theories, and that even when it comes to the technology involved, it is going to be computer technology rather than art-related techniques. Back in 2019, Microsoft Xiaoice (similar to Tay of the US and Rinna of Japan) graduated from the Central Academy of Fine Arts in China and sold works she created at a graduation auction. Although the auction was more of a commercial event than an artistic one, we can still assume to a certain extent that Microsoft Ice has passed the Turing test in terms of artistic realism. However, even so, a large amount of the public still doubts the artistry of AI art.
Thus, a new perspective is urgently needed to be paid keen attention to. Since it is difficult to directly equate AI with art, why not focus on the inspiration that AI-generated art can bring to people? As I mentioned earlier in this article, deep neural networks mimic the perceptual structure of the human brain in a relatively tractable way, to the point that researchers are now able to visualize the emphasis on different granularity features in each layer of the neural network. This is often difficult to achieve in pure art, and even very talented artists sometimes seek to enter the doors of perception with the hallucinogens mentioned earlier when they run out of inspiration or have bad lives. In terms of painting, we know the story of Van Gogh and absinthe; in terms of music, the Beatles received their first marijuana from Bob Dylan; when it comes to literature, we can easily think of Henry Miller of the Beat Generation, who was controversial but inexorably charming. But that doesn’t mean that hallucinogens were a way out or an escape for these artsy giants. Studies have shown that hallucinogens don’t really open the doors for everyone, but only benefit those who are originally creative. Hallucinogens allow already great artists to enhance their perception of the universe, and now we have artificial intelligence, with such gentle, harmless, and controllable tools as deep neural networks to help people form artists to ordinary people to concretize the feast of the augmented perception.
I used RunwayML to train an AI generative model in the Interaction Design class. I fed the model with Dunhuang wall paintings from northwest China, using a pre-trained model of illustration provided inside the application. Even though the output images look far from the real Dunhuang wall paintings, they are successful in some ways. For even though they do not capture the features of the Dunhuang wall paintings very well, they clearly fit the relatively underlying wall painting features, as seen in these generated images. The evidence is that these generated images fit well with the features of the wall paintings from the Stone Age, such as Chauvet Cave Art. These AI-generated images better learn the texture of the stone walls and the overall color atmosphere, but fail to continue to fit the content and lines of the wall paintings at higher levels. There are many factors contributing to this problem, from the quality of the dataset to the pre-trained model chosen and also the tuning of parameters within the model, so I don’t think this means that the generation is a complete failure, at least I appreciate that the AI is trying to learn different levels of features. As well, we can see that many parts of the original mural have been overstated and enlarged in the generated images. This is precisely because the model’s perception of these parts appears to have a similar effect to hallucinogenic art. It is as if when you stare at the starry sky, it somehow becomes a sticky vortex.
The example of the wall paintings is perhaps too simple to support what I mentioned. Even the human brain can quickly process its features at different levels. But what is certain is that AI art similar to this can definitely offer more possibilities for art. We can discuss what AI art can bring to us for ordinary people and for artists separately.
First, on the one hand, for those who are already recognized as artists, AI art certainly can or will provide more inspiration for them in the future. As I have mentioned many times in this article, deep neural networks can help people visualize parts of their thought process that are difficult to capture and understand, thus playing a role similar to traditionally what hallucinogens have done to artists — opening the doors of perception. AI helps them see more and perceive more, then the artist is able to use his or her own creativity to react to the beauty they perceive through the AI. Even better work will be possible. On the other hand, for the general public, some of them may be concerned that well-packaged AI will be abused by the less qualified due to its ease of use, resulting in a proliferation of lower-quality artists and artworks. Strictly speaking, this is indeed a hidden danger, but conversely, just because of its high accessibility, we will be able to let more ordinary people get to know art, experience art, and practice a sense of beauty by participating in the creation of AI art. It is very likely that in the near future, AI art will contribute to a leap forward in human aesthetics as a whole.
The point I want to emphasize is that today, and for a long time to come, we will be in the era of so-called weak artificial intelligence, and robots that fully pass the Turing test will not appear soon. So, artificial intelligence will remain as a tool for humans for a long time. Since as a tool, we can call it a medium in the process of creating art, we need not worry too much about the orthodoxy of the art it generates. After all, it is the person who uses the tool and creates the art that determines the value of the works. Even if the kind of proliferation we fear finally occurs, I believe that it will eventually come to an equilibrium, where art ultimately belongs to creativity, and the artistic experience of the public will be enhanced as a result. So, never underestimate anything that is on the rise even if you can’t accept it for the moment. You only need to think back to photography in the nineteenth century when at the beginning of how many traditionalists despised and attacked, and then look back at the laudable status of the art of photography today, then the whole thing is not that hard to be understood. How can the art of artificial intelligence not follow the same path?
Reference
[1]. Dudley, Robert. “The Drunken Monkey.” 2014, doi:10.1525/9780520958173.
[2]. “Chimpanzees in West Africa Observed Indulging in Habitual Drinking.” The Guardian, Guardian News and Media, 9 June 2015, www.theguardian.com/science/2015/jun/10/chimpanzees-bossou-south-eastern-guinea-habitual-drinking#:~:text=Chimpanzees in west Africa observed indulging in habitual drinking,-Inhabitants of forests&text=They are the chimpanzees of,indulge in regular, habitual drinking.
[3]. Huxley, Aldous. The Doors of Perception. Harper Perennial, 2009.
[4]. Ly, Calvin, et al. “Psychedelics promote structural and functional neural plasticity.” Cell reports 23.11 (2018): 3170–3182.
[5]. “Inceptionism: Going Deeper into Neural Networks.” Google AI Blog, 17 June 2015, ai.googleblog.com/2015/06/inceptionism-going-deeper-into-neural.html.
