Can a machine take us by surprise? This is the central argument of Alan Turing’s 1950’s seminal work, “Computing Machinery and Intelligence.” Can a human take another human by surprise? This is one of the central ‘arguments’ at the center of the performance of magic. What can an investigation on the structure of machine learning teach us about the form of magic and about the experience of surprise?
On Machines
Let’s begin at the beginning, of sorts. In the 1840’s, Ada Lovelace came to prominence as a mathematician and is thought of as the first computer programmer after creating a series of operations for Charles Babbage’s “Computing Engine,” which is typically thought of as the first automatic digital computer. Lovelace posited that due to humans building and programming such computing machines, it was not possible for them to take us by surprise.
Nearly a century later, famed mathematician Alan Turing argued the opposite. He states that: “Machines take me by surprise with great frequency. This is largely because I do not do sufficient calculation to decide what to expect them to do, or rather because, although I do a calculation, I do it in a hurried, slipshod fashion, taking risks. Perhaps I say to myself, ‘I suppose the voltage here ought to be the same as there: anyway, let’s assume it is. Naturally I am often wrong, and the result is a surprise for me, for by the time the experiment is done these assumptions have been forgotten. These admissions lay me open to lectures on the subject of my vicious ways, but do not throw any doubt on my credibility when I testify to the surprises I experience.”
If a machine can take a human by surprise, how does this influence our definitions of surprise? More on this later.
Turing created one of the most famous thought experiments of our age as a way to tease out ideas about if machines could possess ‘intelligence’ and, if so, how to distinguish human intelligence from that exhibited by a machine: The Turing Test.
He expanded this to propose that: “Suppose that we have a person, a machine, and an interrogator. The interrogator is in a room separated from the other person and the machine. The object of the game is for the interrogator to determine which of the other two is the person, and which is the machine. [This is done via the interrogator asking a series of questions to each anonymous entity.] The object of the machine is to try to cause the interrogator to mistakenly conclude that the machine is the person; the object of the person is to try to help the interrogator to correctly identify the machine.”
Turing’s test was inspired by a Victorian-era parlor game, the Imitation Game, which possessed a similar framework, but was instead composed of a man, woman, and interrogator. Here, it was the job of the interrogator to distinguish the man from the woman based on asking a series of questions to decipher gender as defined by the stereotypes of the time.
On Magic
As a lifelong magician, I have often wondered as to why certain pieces of magic are more surprising to audiences than others. Why do some effects, such as a cut and restored rope, seem to provide, in lieu of a surprise, a feeling of a rational completion, whereas others provide a borderline level of shock, such as the transformation of a ball into a dove?
I began deeply thinking about the experience of surprise in 2019 after finding Thomas Griffith’s 2014 paper “Revealing ontological commitments by magic.” Here, he poses that “Considering the appeal of different magical transformations exposes some systematic asymmetries. For example, it is more interesting to transform a vase into a rose than a rose into a vase. An experiment in which people judged how interesting they found different magic tricks showed that these asymmetries reflect the direction a transformation moves in an ontological hierarchy: transformations in the direction of animacy and intelligence are favored over the opposite.” This, while on face value seems an obvious observation, actually leads to a nuanced line of thought as to the depth of the ontological commitments that we do hold about each object and both how rigid and flexible they are simultaneously.
Take, for example, one of the items in his list of objects, a blackbird. Blackbirds possess a litany of properties, such as the ability to fly, the sheen of their feathers, the ability to make a distinctive sound, etc. They may also hold a set of shifting personal and cultural connotations for observers, such as summoning thoughts of superstition, ancestral respect, or spirituality based upon regional and historical culture. Hence, a deep investigation of the fixed and fluctuating nature of an object or animal’s properties and the ontological commitments that an audience may hold about them seems worthy of reflection in magic.
Reading Griffiths’ study seemed to address much of my line of inquiry, though the paper mainly addresses ‘interestingness’ while not necessarily focused on the experience of surprise itself. (I highly recommend that anyone with an interest in this topic read the full paper.)
On Surprise
After having spent the past year researching surprise through the lenses of science, philosophy, sociology, and innumerable conversations, I feel little closer to a grasp of this slippery experience. To summarize a few quick thoughts:
From a scientific standpoint, in the study “Intuitions about magic track the development of intuitive physics” by the team of Casey Lewry, Kaley Curtis, Nadya Vasilyeva, Fei Xu, and Thomas L. Griffiths, they note that: “However, there is some debate surrounding the extent to which looking time is an accurate measure of surprise. Wang et al. (2004) note that when they refer to violation-of-expectation paradigms as measuring infants’ surprise, ‘surprise’ is shorthand for a state of attention or interest. While there is wide consensus that a difference in looking time indicates detection of a difference between the two events, some have argued that this attention or interest could be caused by familiarity with the event or prediction of an event, rather than by surprise at a violation of an expectation, thus providing no evidence for an understanding of the physical principle in question (e.g., Bogartz, Shinskey, & Speaker, 1997; Jackson & Sirois, 2009). However, as Hamlin (2014) explains, the evidence for infants’ surprise at an event is distinct from evidence for infants’ prediction of an event, and well-designed research can distinguish between these two interpretations.”
Research on the phenomenology of surprise holds ideas based on the richness of the lived, embodied experience. In “Phenomenology of Error and Surprise: Peirce, Davidson, and McDowell” by Elizabeth F. Cooke, she notes: “Our experience of surprise from a first-person point of view tells us that we often do make incorrect judgments about the world and feel compelled to rescind them. And for error recognition to occur, some things must be in place: an inquirer with concepts and a set of beliefs, as well as a nonego acting on the inquirer. The subject’s experience is partly of the self as an object in relation to the world. […] Surprise is felt as causal and conceptual within experience and provides conceptual friction with the world. It serves as an empirical self-corrective insofar as it is forced by a nonego, yet still conceptual.
English cultural theorist Mark Fisher explores the haziness of surprise’s adjacent experiences such as the weird and the eerie, and how weirdness notes a feeling ‘this does not belong,’ and the intertwining of the historical ideas of the weird and of fate itself.”
Finally, in many conversations with lay people across a broad spectrum of industries, ages, etc., there seems to be no shortage of interest and opinions on this topic. I have encountered sentiments along a full spectrum that surprise is nothing more than a biological being startled, to the feeling that all experience is emotion driven and thus surprise is a personal and cultural fleeting experience.
Magic, Machine Learning, and Surprise
My thoughts on these topics converged in spring of 2021. I was fortunate enough to be named as an Affiliate of meta LAB (at) Harvard, the greatest honor of my life to date. One of the original Principals of the lab, the ever-inspiring Matthew Battles, posed the question to me whether the original Victorian parlor “Imitation Game” was similar to the parlor magic of the era and further: A) if this style of magic was in my wheelhouse and B) if using parlor magic might be an interesting way to explore these ideas of surprise and computational intelligence.
I was taken aback by this beautiful line of thought and began an immediate deep dive into the relationships between magic, surprise, and machine learning. Could I design an algorithm to generate descriptions of novel magic effects with varying levels of surprise factors? This is what I set out to do. After reaching out to some brilliant computer scientists and coders, my initial thought was to join with them to create algorithms: inspired by some of Griffiths’ thoughts on interestingness, and to also weave in my own ideas on the properties of objects noted in his studies that seemed to contribute to my ongoing ideas of what may constitute a surprising transformation.
I anticipated being blown away by truly surprising results that were more novel and exciting than our own historical effects. At the time of writing (spring 2022) the algorithm is still in progress, yet at every turn, I find myself diving deeper and deeper into the historic magic texts that I utilized in initial design research to train the algorithms. (I wanted to only use historic magic catalogs, as I planned to have public conversation around this process and thus wanted to have text that did not include methods.)
So far, although some outputs have been intriguing, funny, and perhaps surprising, they all seem to lack the deep psychological nuance that the original magic effects in the catalogs reference. I have been continually inspired by the delicate psychology that was built and honed over hundreds of years by those in the field of magic.
Perhaps, as the algorithm I am working with my collaborators on gets trained further, it will prove me wrong. Time will tell.
Modern machine learning is a dance between the human and the non, the trained and the untrained, and the level of human intervention that allows for the rendering of new and certainly unexpected forms. Perhaps an unpopular opinion, but maybe machine learning is only a tool to help us get to the root of understanding our own humanity. By seeing where machine learning does and does not meet or confound our expectations, we can explore why we hold the beliefs that we do and if they are worth rethinking or leaning further into.
When I am met with a new algorithmically generated output of a magic effect, it prompts me to think deeply about why I find it less surprising, and if I were to perform it, what would make it more surprising. So perhaps, for now, machine learning is just that, a tool to help us investigate the beautiful history of our field and the subtle psychological underpinnings that we are lucky enough to play within.
A Note
I will be further investigating these ideas in my new performance and installation work, “Taken by Artificial Surprise,” which I will be debuting with CultureLAB in New York City in July of 2022 as part of an ongoing series of work exploring this topic. I also must note that this project was inspired by my time as an Affiliate at metaLAB at Harvard. During this time, I encountered a diversity of ideas and research, and discourse with metaLAB members also greatly assisted the ideation process.
Sources
Ada Lovelace.” Ada Lovelace | Babbage Engine | Computer History Museum, www.computerhistory.org/babbage/adalovelace/.
“Ada Lovelace.” Encyclopædia Britannica, Encyclopædia Britannica, Inc., www.britannica.com/biography/Ada-Lovelace.
A. M. TURING, I.—COMPUTING MACHINERY AND INTELLIGENCE, Mind, Volume LIX, Issue 236, October 1950, Pages 433–460, https://doi.org/10.1093/mind/LIX.236.433
Cooke, Elizabeth F. “Phenomenology of Error and Surprise: Peirce, Davidson, and McDowell.” Transactions of the Charles S. Peirce Society, vol. 47, no. 1, 2011, pp. 62–86, https://doi.org/10.2979/trancharpeirsoc.47.1.62. Accessed 19 Apr. 2022.
Fisher, Mark. The Weird and the Eerie, 2016. Print.
Griffiths, Thomas. Revealing Ontological Commitments by Magic – Cocosci.princeton.edu. cocosci.princeton.edu/tom/papers/magic.pdf.
Lewry C, Curtis K, Vasilyeva N, Xu F, Griffiths TL. Intuitions about magic track the development of intuitive physics. Cognition. 2021 Sep; 214:104762. doi: 10.1016/j.cognition.2021.104762. Epub 2021 May 26. PMID: 34051423.
Oppy, Graham, and David Dowe. “The Turing Test.” Stanford Encyclopedia of Philosophy, Stanford University, 4 Oct. 2021, plato.stanford.edu/entries/turing-test/.
Jeanette Andrews is a magician, artist, and independent researcher based in New York City. Andrews presented her first magic performance at age four, was paid to do her first magic show at age six, began running her business that day, and has never had another job since. She has staged sold-out and standing-room-only performances for Fortune 500 companies, theaters, and universities and has presented commissioned and site-specific works for The Smithsonian’s Cooper Hewitt, the International Museum of Surgical Science, and the Museum of Contemporary Art Chicago. Andrews is an Affiliate of Harvard’s metaLAB and current artist in residence for Culture Lab LIC in New York city and prior artist-in-residence for High Concept Labs and The Institute for Art and Olfaction. Illusion is Ms. Andrews’ life’s work and her performances have been praised by the Chicago Tribune, PBS, and the New York Times.
Recent Comments