Subverting Cognition: Surrealist Automatism and Brooks’ Intelligence Theory
March 13, 2017 § 3 Comments
In Flesh and Machines: How Robots Will Change Us, Rodney Brooks presents his unique take on the pathway to create meaningful artificial intelligence. To briefly summarize, he suggests that removing clunky algorithms aimed at simulating cognition, while simultaneously creating a direct link between sensation and action, supports more advanced general intelligence (functional intelligence). For me, Brooks’ theory of intelligence found in “the interaction between perception and action” (Brooks Ch.3 “Planetary Ambassadors”) called to mind the techniques of Surrealist painters like Salvador Dalí, Joan Miró, and André Masson. The Surrealists used automatism — painting without conscious thought — to subvert their own cognition and the rational mind, in order to tap into deeper and more raw thoughts and feelings. I argue that given the parallel intention of Brooks’ approach to AI and Surrealist Automatism, an exploration of the latter can help us gain an understanding of Brooks’ method.
Before diving in, let me quickly clarify what I mean when I say ‘meaningful artificial intelligence’ or ‘general intelligence’. In Flesh and Machines, Brooks distinguishes between the tasks and processes traditionally tackled by AI researchers (playing chess at the level of mastery, solving calculus problems, etc) and more practical expressions of intelligence (entering a room, navigating a new environment, avoiding obstacles), and points to the fact that programming a robot to do the latter is a considerable challenge. Thus, I define meaningful or general artificial intelligence as the intelligence that human beings and animals employ in performing ‘basic’ operations, but are far more complicated on a cognitive level than they appear to us. Brooks’ strategy was geared towards cracking the code of programming this kind of intelligence, and it was these simpler actions and motions that the Surrealists sought to simplify their brush strokes and methods too.
Caught in a time of political uncertainty both within the art world and the world at large, the Surrealists reflected on how the conscious mind and higher-level cognition is difficult and beset with ideology from what they saw as a flawed society. They wanted to divorce the art-making process from the constraints of the rational mind indoctrinated by an oppressive society. In order to escape, they adopted a working method called automatism, which allowed them to essentially paint without conscious thinking, thus sourcing the lines and forms that resulted from their subconscious.
Pioneered by André Masson, the Surrealists painted using automatism by first completely clearing their minds. Often, they would even close their eyes or use drugs or natural supplements to achieve a more detached state. Then, they would allow their hand, holding a paintbrush, to flow randomly across the canvas, so that the resulting lines and forms were more a product of chance than conscious manipulation of the paintbrush. In this way, their style was free of rational control. In this way, the Surrealists thought that the compositions they created using automatism came directly from their subconscious — the epicenter of interaction between perception and action. In other words, they tried to simplify their cognitive processes as much as physically possible down to the point where they merely operated on the basis of the interaction between their perception — the way the paintbrush felt across the grain of the canvas — and action — creating a line via the paintbrush on the canvas.
Andre Masson, “Automatic Drawing”, 1924
As you can see, there are strong parallels between the working method of the Surrealists and Brooks’ approach to simulating general intelligence. In explaining the benefits of his theory, Brooks describes his “subsumption architecture” for machines, by which he created a hierarchy of processing levels simulating the process of evolution of traits.
“For Allen [his first physical robot] I targeted three layers. The first was… to make sure that the robot avoided contact with objects… The second layer would give the robot wanderlust, so that it would move about aimlessly. The third layer was to explore purposefully whenever it perceived anything interesting…” Rodney Brooks, Flesh and Machines
Using Brooks’ vocabulary and framework of layers, one can analyze the process of the Surrealists in a similar way. The primary (first, or bottom) layer of automatism was to paint. The fundamental task programmed into the Surrealist’s action was to paint by dipping paintbrush into paint and applying it to the canvas. The second layer was to paint continuously without really stopping. The surrealists were concerned that if they paused longer than an instant, the conscious mind would kick back in. In terms of programming, the continuous painting layer doesn’t have to process how to paint, because the first layer achieved that. Then, the third and tertiary layer was to explore/follow through on a particular line if the sensation of that line or the texture in that region of the canvas captured the artist momentarily. Thus, the automatism employed by some of the Surrealist painters very closely mirrors the “thoughts”of Brooks’ coded AI. In fact, the drawing by Andre Masson above even looks like it could be an aerial view of the paths that Allen may have taken, moving around Brooks’ research lab.
I find the parallels between the techniques of the Surrealists (developed and employed in the early 1900’s) and Brooks’ theory of intelligence (developed and employed in the late 1900’s) to be confirming the validity and ingenuity of Brooks’ theory for machine intelligence. That is to say, if human beings sought to shed the weight and burden of clunky cognitive thought in order to achieve a greater level of functionality in expressing themselves on the canvas, it is incredibly impressive and valid for Brooks to suggest the same for his machines; he argued to remove dedicated “cognition boxes” from his machines, thus eliminating time-consuming and complex cognition algorithms from his AI’s ‘thought’ processes.
While Brooks may not have drawn inspiration from the methods of the Surrealists, I find it beautiful that leaders in two remarkably distinct disciplines both arrived at a similar point of seeking a purified relationship between sensation and action to achieve greater expression and intelligence of movement. Though it has often been suggested that links between memories will be the key to creating thinking artificial intelligence, Brooks’ theories have led me for the first time to consider that in the future, progress in AI development will also come from the mutual inspiration between disciplines, especially the humanities in creating more “intelligent” and human-like robots and machines.
❈❈❈
Sources / For more information on Surrealism and Automatism:
https://www.moma.org/learn/moma_learning/andre-masson-automatic-drawing
Patrizio Murdocca
Science/Fiction 
I think your comparison between Surrealist Automatism and Brooks’ approach to artificial intelligence is certainly warranted. After all, both concepts disregard top-down cognition and consciousness in order to perform a function efficiently. Under their respective theories, surrealist painters could eschew the hindrance of intentionality, while robots could react to an environment without expansive and necessarily inefficient algorithms. Yet, I am reluctant to call either practice intelligent.
After all, the purpose of Surrealist Automatism was to discard consciousness, to remove thought from the equation. While Surrealist Automatism produces art that certainly contains emotional value, the only intellectual value of the work is in its countercultural novelty; it is meta-analysis that contains the intellectual merit, not the painter’s process or product.
The similarity between this anti-cognition and Brooks’ ideas is captured in your insight that “the drawing by Andre Masson… could be an aerial view of the paths that Allen may have taken, moving around Brooks’ research lab”. Just as the drawing is purposefully unintellectual, robots constructed under Brooks’ ideology are unintellectual. While they perform reactive processes smoothly, an admittedly monumental accomplishment in robotics, they are more of a breakthrough in function than intelligence; Brooks creates incredibly functional robots, but not intelligent ones. These creations utilize snap judgements to their environment in the same way that surrealist painters let their brush move arbitrarily. Similar to the paintings, Brooks’ robots only are intelligent only to a certain observer. Brooks’ approach to intelligence ignores and makes no progress in the fields of self-regulation, self-reflection, learning, memory, or other primary functions of intelligence. This is not to say that Brooks’ did not make important contributions to robotics. He certainly did. Instead, I propose a distinction between gains in functionality and progress in artificial intelligence.
LikeLiked by 2 people
I think your connection between the Surrealist painters and Brooks piece on artificial intelligence is very creative. I never would have considered such a connection, however, after reading your blog post I can see it very clearly. It makes a lot of sense because even though we like to harshly delineate between STEM and the arts,the human beings that are participating in these disciplines have the same “general intelligence” that allows us to function in ways that we often take for granted. If those involved in the technology world are attempting to isolate this kind of intelligence and implement it into AI, they are likely to conceptualize it the same way as someone in the arts would. What makes it different is how they choose to tackle said problem. How an engineer would choose to program AI is far different than how a Surrealist artist would, but both are equally valid and point to the great diversity of human thought and ability to problem solve. I think this is why interdisciplinary work is so important–it allows for different kinds of thinkers across fields to work together to solve problems. I am not implying that engineers should be taking hallucinogenics in order to solve the puzzle of creating AI, but I am saying that there is a lot to be learned from different kinds of thinkers that I think is often overlooked or undermined.
In class we talk a lot about if human intelligence should be the standard from which we measure AI and whether or not this scale demonstrates an inherent hubris. There are many different kinds of intelligence seen even within human beings, which I think is important when we are thinking about “human intelligence” and what we mean when we talk about AI achieving it. There is a lot to be learned from the varying kinds of expertise and I think a broader conception of intelligence as whole could help lead to creative innovation.
LikeLike
You make a really interesting connection between Surrealism and AI. We’ve discussed art before with poetry and the Turing Test because art is one of the most human things we do: we assign subjective meaning and personal value to abstract concepts. The surrealist movement is often misunderstood, as the artists chose to remove consciousness from their art, creating works that are confusing and often not aesthetically pleasing but more thought provoking. What I find most intriguing about Surrealist art is that the artists chose to abandon consciousness—what makes them so human—because of how inhumane the world had become. Surrealism gained momentum between the two World Wars, when Europe especially was struck with political turmoil and high poverty in city centers. The surrealists sought to escape their worlds by abandoning their consciousness.
I’d be interested to hear what the artists have to say about AI. While they might see it as art and a complete disassociation from the human mind, I’d imagine they’d see it more as an example of human ego, creating an image of himself that is stronger and more intelligent, only to cause further damage. I like the idea of AI as art, but the idea of them being used as weapons or tools seem far more plausible.
LikeLike