6/24/2021

Signal of Intelligence

 

Navigating social systems efficiently is critical to our species. Humans appear endowed with a cognitive system that has formed to meet the unique challenges that emerge for highly social species. Bullshitting, communication characterised by an intent to be convincing or impressive without concern for truth, is ubiquitous within human societies. Across two studies (N = 1,017), we assess participants’ ability to produce satisfying and seemingly accurate bullshit as an honest signal of their intelligence. We find that bullshit ability is associated with an individual’s intelligence and individuals capable of producing more satisfying bullshit are judged by second-hand observers to be more intelligent. We interpret these results as adding evidence for intelligence being geared towards the navigation of social systems. The ability to produce satisfying bullshit may serve to assist individuals in negotiating their social world, both as an energetically efficient strategy for impressing others and as an honest signal of intelligence.

[The Bullshitter]…is neither on the side of the true nor on the side of the false. His eye is not on the facts at all, as the eyes of the honest man and of the liar are, except insofar as they may be pertinent to his interest in getting away with what he says. He does not care whether the things he says describe reality correctly. He just picks them out, or makes them up, to suit his purpose.

Harry G. Frankfurt (2009)

Human intelligence has been a long-standing mystery to psychologists: In particular, why humans differ so greatly in their intelligence compared not only to distantly related animals, but our closest primate cousins. Large brains are energetically expensive (Cunnane et al., 1993Raichle & Gusnard, 2002) and necessitate that human children require inordinate levels of post-partum investment from caretakers (Rosenberg & Trevathan, 2002). Nevertheless, human brains have continued to increase in size over our evolutionary history until only recently (Beals et al., 1984Bednarik, 2014). It remains a puzzle to explain why humans continue to support the steep investment of resources that comes with maintaining a large and powerful brain, with leading theories suggesting that the cognitive, social and cultural advantages afforded by such large brains outweigh the costs (Seyfarth & Cheney, 2002). Classically, intelligence has often been considered mostly—or sometimes solely—for its value in manipulating and understanding the physical world (Humphrey, 1976), the environment for an organism being a series of cognitive puzzles which intelligence assists them in completing. More recent developments have expanded on this classical understanding through acknowledging that the complexities of an organism’s social life may place just as high of a demand on an organism’s intelligence as the complexities of its physical life (if not more; Byrne, 1996Byrne & Whiten, 1990Whiten, 2018). Far removed from the relatively sterile cognitive puzzles with which we now test and study intelligence, there is reason to believe that the origin of intelligence is best understood for its social uses (Gavrilets & Vose, 2006Geher & Miller, 2007; McNally, Brown, & Jackson, 2012). It is this perspective that grounds the current work.

Several theories have been forwarded to explain the high level of intelligence observed in humans. Some of the most promising among these theories have examined intelligence for its value in assisting us in navigating the complex social systems that characterize our species. Intelligence in the social world is theorized to have been formed primarily in response to three pressures. The first is the need to accurately signal intelligence in order to demonstrate genetic quality and fitness to potential mates (McKeown, 2013Miller, 2000Miller & Todd, 1998). The second, a pressure to manipulate, deceive, or influence others through the application of such social intelligence (Byrne, 1996Byrne & Whiten, 1990Handel, 1982Sharma et al., 2013Whiten, 2018). Third, the pressure to accurately maintain and manipulate mental models of complex social networks and interactions, as well as being able to simulate the mental states of others (Bjorklund & Kipp, 2002Roth & Dicke, 2005Stone, 2006). A cartoonish description of the hypothetical person who exemplifies all of these traits in the extreme would be one who shows off their intelligence whenever possible, tells lies when it is advantageous to do so, and is capable of keeping track of all the lies they have told.

Possessing a high level of intelligence allows humans to meet the intense demands placed on them by complex social systems. Beyond the Machiavellian value of social savvy, evidence suggests that large brains and their corresponding cognitive advantages may have been selected for as a result of their sexual appeal (Crow, 1993McKeown, 2013Miller, 2000Miller & Todd, 1998Schillaci, 2006). In line with signaling accounts, charisma in the form of humor and leadership abilities has been argued to function as an honest signal of desirable qualities, including cognitive ability (Greengross & Miller, 2011Grabo et al., 2017). In biology, an “honest signal” is one that conveys accurate information about an unobservable trait to another organism. For example, a brightly colored frog that is poisonous honestly signals its toxicity to predators; it looks dangerous, because it is. In contrast, a dishonest signal is an attempt to mislead another organism into believing that the signaler possesses a trait which it does not. For example, a harmless insect may possess the same coloration as a harmful wasp, falsely signaling that it is just as dangerous as a wasp in order to avoid predation; it looks dangerous, but it is not. In the context of sexual signaling in humans, a person of high intelligence who is able to communicate this to others is giving an honest signal that they possess this desirable trait. In this case, the “honesty” of a signal is independent of the truth content of the specific communication used to signal. For example, a smooth and intelligent liar may give the impression that they are intelligent even while saying nothing true.

The ability to produce satisfying bullshit, with its emphasis on impressing others without regard for truth or meaning (Frankfurt, 2009Pennycook et al., 2015), may represent an energetically inexpensive strategy for both signaling one’s intelligence, and deceiving others to one’s advantage. Indeed, past work provides initial evidence for this claim, demonstrating that indiscriminately attaching meaningless pseudo-profound bullshit titles to artworks increases their perceived profundity (Turpin et al., 2019). On this basis, it has been hypothesized that bullshit can be used to gain a competitive advantage in any domain of human competition where the criteria for determining who succeeds and fails at least partially relies on impressing others. In this way, bullshit may serve as an honest signal of a person’s intelligence (and therefore their fitness), even though the specific content of the bullshit itself may be false.

A growing body of literature has investigated peoples’ receptivity to bullshit, specifically computer-generated pseudo-profound bullshit consisting of random arrangements of superficially impressive words in a way that maintains syntactic structure (e.g., “Wholeness quiets infinite phenomena”; Pennycook et al., 2015Pennycook & Rand, 2019Walker et al., 2019). Other work has begun to examine the frequency of bullshit production (Littrell et al, 2020; 2021), including investigation of the conditions under which people are most likely to produce bullshit (Petrocelli, 2018). Yet, minimal work has assessed how bullshit can be used to navigate social systems (McCarthy et al., 2020Turpin et al., 2019). For example, a person who is capable of producing good bullshit may be perceived as especially charming, convincing, or competent as long as their deception is left undiscovered. Relatedly, styles of bullshitting that allow one to avoid awkward or uncomfortable social situations may go far in fostering social harmony (Littrell et al., 2020). This type of bullshitting (i.e., evasive bullshitting) could be employed to avoid lying, while replacing the direct response with a less relevant truth (Carson, 2016Littrell et al., 2020). For example, a friend gifts you a sweater that you find hideous, but when asked how you like it, you respond with “thank you, this is very thoughtful of you!” Given the potential usefulness of bullshit as a method for navigating social systems, and with evidence that human intelligence may be set up largely for navigating the social world, an open question is whether bullshit ability as a behavioral feature reveals something about one’s relative intelligence. If our brains have evolved for the purpose of manipulating information about social relationships (e.g., using tactical deception; Dunbar, 1998), then it is plausible that intelligent people will produce bullshit that is of higher quality, as a means of efficiently navigating their social surroundings.

The current work investigates the role which bullshit ability plays in signaling intelligence. We assess both how the quality of bullshit reveals the true intelligence of bullshit producers as well as how bullshit quality is received as a signal of intelligence by observers. To examine these questions, we had a sample of participants attempt to explain fictional concepts in a way that appeared satisfying and accurate (i.e., with bullshit), while other samples judged the quality of these explanations and the intelligence of their creators. We hypothesized that bullshit would behave as an honest signal of one’s intelligence such that those able to create the most satisfying and seemingly accurate bullshit would also score higher on tests of cognitive ability. Furthermore, we predicted that those judged as producing high quality bullshit would also be perceived as more intelligent. Therefore, we expected bullshit ability to relate positively with measures of cognitive ability as well as perceptions of intelligence.

Participants

A sample of 483 undergraduates from the University of Waterloo, located in Ontario, Canada, volunteered to complete Study 1 in exchange for course credit.

Materials and Measures

A full list of the materials and measures presented in Study 1 can be viewed in the Online Supplemental materials (Part A).

Bullshit willingness and generation task

Inspired by Jerrim and colleagues (2019), we presented participants with ten concepts (e.g., cognitive dissonance) four of which were fake (i.e., subjunctive scaling, declarative fraction, genetic autonomy, neural acceptance). Participants’ first task (bullshit willingness task) was to rate their knowledge of each concept on a 5-point scale ranging from “never heard of it” to “know it well, understand the concept.” Responses given to fake concepts were summed to create an index of participants’ bullshit willingness, with higher scores indicating a greater tendency to bullshit (i.e., feign knowledge of fake concepts). Next, a subset of participants (Bullshit Producers) were presented with each of the ten concepts individually and—consistent with descriptions of bullshit as being characterized by a lack of concern for the truth (Frankfurt, 2009) —were instructed to “produce the most convincing and satisfying explanation” they could for each concept. For concepts they were unfamiliar with, participants were instructed to “be creative and make up an explanation that you think others will find convincing and satisfying.” The verbatim instructions were as follows:

Your task is to try to produce the most convincing and satisfying explanation that you can for each term.

For terms that you are knowledgeable about, we ask that you simply explain them as best you can (that is, in the most convincing and satisfying way).

For terms that you are unfamiliar with, we ask that you be creative and make up an explanation that you think others will find convincing and satisfying.

Do not worry about the truth of your claims when making up your explanations, rather, you may treat this as a creative writing exercise.

Explanation evaluations

We had a sample of participants (Bullshit Raters) judge the accuracy and satisfactoriness of 120 explanations of fictitious concepts produced by Bullshit Producers in our bullshit generation task. Participants evaluated the accuracy of each explanation with the prompt “How accurate is this explanation,” responding on a 5-point scale ranging from “Not at all Accurate” to “Very Accurate.” Similarly, for each explanation, participants were asked “How satisfying is this explanation,” providing responses on a 5-point scale that ranged from “Not at all Satisfying” to “Very Satisfying.” This resulted in each Bullshit Producer having one “satisfyingness” and “accuracy” judgement for each of the bullshit statements that they generated. The highest scoring item out of these bullshit statements was selected to be the best indicator of their bullshit ability. The “accuracy” and “satisfyingness” ratings of this item was averaged to create a “Bullshit Ability” score which was then averaged across all Bullshit Raters who rated that Bullshit Producer’s statements (see Figure 1). This method of calculating bullshit ability was adapted from Greengross and Miller (2011) who used a similar process to calculate participants’ humour ability.


                        figure

Figure 1. Summary diagram for methods used in Studies 1 and 2.

Note. Visual depiction of how bullshit ability was computed in the present study. This figure is only a representation of the process and does not align to the total number of explanations generated by Bullshit Producers or the number of Bullshit Raters assigned to evaluate Bullshit Producers.

Wordsum task

The Wordsum task is a 10-item vocabulary test commonly used as a measure of verbal intelligence (see Malhotra et al., 2007 for a review). In this task, a word in large print (e.g., “CLOISTERED”) appears above a series of smaller print words (e.g., bunched, secluded, malady, miniature, arched). Participants’ objective is to pick a small print word that is the best synonym for the large print target word. Scores on the Wordsum task were equal to the total number of correct responses provided. Additional information concerning how participants in these studies compared to typical performance can be found in Part A of the Online Supplemental Materials.

Raven’s progressive matrices

We administered Raven’s Progressive Matrices (RPM) as a measure of abstract reasoning and non-verbal fluid intelligence (Bilker et al., 2012). In this task, participants are presented with a partially obscured visual pattern and must select the available pattern fragment that will successfully complete the pattern. The RPM is comprised of 60 items broken up into five levels of difficulty. In order to decrease time demands on participants, we randomly selected four items from each of the five difficulty levels, resulting in 20 RPM items being presented in Study 1. We calculated an RPM score for each participant by calculating the number of correct responses they provided. Additional information concerning how participants in these studies compared to typical performance can be found in Part A of the Online Supplemental Materials.

Profundity ratings

We assessed participants’ receptivity and sensitivity to pseudo-profound bullshit by having them assess the profundity of 30 statements originating from Pennycook and colleagues (2015). These 30 statements consisted of 10 pseudo-profound bullshit statements, 10 motivational quotations, and 10 mundane statements. Pseudo-profound bullshit statements were originally retrieved from websites able to create meaningless statements by randomly arranging a list of profound-sounding words in a way that preserves syntactic structure (e.g., “Wholeness quiets infinite phenomena”). These statements, while perhaps superficially impressive, were created such that they lack an intended meaning. Contrasting meaningless pseudo-profound statements were motivational quotations and mundane statements. Motivational quotations were designed to capture a true attempt at communicating something meaningful and profound (e.g., “A wet man does not fear the rain”) while mundane statements were designed to be easily interpretable, yet not contain truth of a grand or profound nature (e.g., “Newborn babies require constant attention”). Participants assessed the profundity of all 30 statements on a 5-point scale which ranged from 1 (Not at all profound) to 5 (Very profound). A bullshit receptivity score (BSR) was calculated for each participant by averaging the profundity ratings provided to pseudo-profound bullshit statements. Additionally, a bullshit sensitivity score (BSS) measuring participants’ ability to distinguish pseudo-profound bullshit from motivational quotations was calculated by subtracting participants’ average profundity rating given to motivational quotations from their average profundity rating given to pseudo-profound bullshit statements.

Design and Procedure

Study 1 was conducted in two phases (see Figure 2). First, we had 220 participants (Bullshit Producers) complete a bullshit willingness task in which they reported their knowledge of ten (six real and four fake) concepts. Next, participants were presented with these same concepts independently and attempted to generate convincing and satisfying explanations of each concept (bullshit generation task). Following the completion of these tasks, participants assessed the profundity of 30 statements (10 pseudo-profound bullshit, 10 motivational quotations, and 10 mundane statements) and completed the RPM and Wordsum.


                        figure

Figure 2. General overview of bullshit production task.

Note. Visual depiction of the methodology used in the present research. Participants in our Bullshit Producers sample (n = 220) generated explanations of both real and fake concepts which were then judged by the Bullshit Raters sample of Study 1 (n = 263) and Study 2 (n = 534). These judgments were used to calculate a bullshit ability score for each Bullshit Producer.

In a second phase, 263 participants (Bullshit Raters) were presented with and evaluated how accurate and satisfying they found 120 explanations of both real and fake concepts. All explanations were generated by participants in our Bullshit Producers sample, with each participant in our Bullshit Raters sample evaluating the explanations generated by a random subset of 12 Bullshit Producers. Participants in this sample completed the bullshit willingness task prior to all explanation evaluations and completed our profundity task, the RPM, and Wordsum following these evaluations.

We conducted correlational analyses between our main variables of interest (see Table 1). As our primary focus was on the characteristics (e.g., intelligence) of those producing bullshit, we focus exclusively on the associations within our Bullshit Producers sample here. Although note that the bullshit ability of each participant in our Bullshit Producers sample was judged exclusively by our Bullshit Raters sample. All analyses focused on individual differences within our Bullshit Raters sample can be viewed in the Online Supplemental Materials (Part B). Of primary interest was to assess whether participants’ ability to bullshit (i.e., produce seemingly satisfying and accurate explanations of fake concepts as indexed by the average of these two ratings) would correlate positively with measures of their intelligence. To this end we observed significant positive correlations between participants’ bullshit ability and Wordsum scores, r(203) = .23, p < .001, as well as between bullshit ability and RPM scores, r(202) = .15, p = .032. Therefore, we find initial evidence of bullshit ability sharing a modest positive association with measures of intelligence.

Table

Table 1. Study 1 Correlations.

Table 1. Study 1 Correlations.

Additionally, we find that participants’ bullshit ability was uncorrelated with their willingness to bullshit (i.e., feign knowledge of fake concepts), r(216) = .04, p = .544, and their receptivity to pseudo-profound bullshit (i.e., endorse meaningless pseudo-profound statements as profound), r(216) = −.09, p = .217. Furthermore, participants’ willingness to bullshit was negatively associated with scores on the Wordsum, r(204) = −.17, p = .014, and RPM, r(203) = −.33, p < .001, suggesting that those scoring higher on our measures of cognitive ability were less willing to bullshit. Finally, we find that those more willing to bullshit were also more likely to be receptive to pseudo-profound bullshit (i.e., rate pseudo-profound bullshit items higher on profoundness), r(217) = .32, p < .001, as well as were less likely to distinguish between meaningless pseudo-profound bullshit and meaningful motivational quotations (bullshit sensitivity: calculated as the difference between pseudo-profound bullshit ratings and ratings of motivational quotations for their profoundness), r(217) = −.22, p = .002. Thus, contrary to the common expression, it may indeed be possible to “bullshit a bullshitter.”

Study 1 provides initial evidence suggesting that bullshit ability serves as an honest yet modest signal of a person’s cognitive ability. However, what may be more important from the perspective of social navigation is how that signal of intelligence is received by others. Independent of one’s true intelligence, having others believe that one is intelligent may confer reputational and social advantages. Therefore, in Study 2, we assessed whether those able to generate convincing bullshit are viewed as more intelligent than those less able to generate convincing bullshit.

Participants

A sample of 534 University of Waterloo undergraduates completed Study 2 in exchange for course credit. Originally, 278 participants were collected, however, during the Covid-19 outbreak in March of 2020, all researchers in the Department of Psychology were requested to collect more data online so that students could have the opportunity to receive course credits. As a result, an additional 256 participants were collected. These additional participants were collected before any analyses were conducted.

Materials and Measures

The materials and measures used in Study 2 mirrored that of Study 1. The only difference was that in Study 2 participants also judged the intelligence of the producer of each explanation. Study 2 made use of the same fictious explanations generated by the “Producer” sample in Study 1, and recruited new sample of Bullshit Raters to rate those explanations.

Explanation evaluations

As in Study 1, we had participants judge how accurate and satisfying they found explanations of various concepts. However, in Study 2 participants were also asked “How intelligent is the person who provided this explanation.” All responses to this item were made on a 5-point scale that ranged from “Not at all Intelligent” to “Very Intelligent.” In the same fashion as Study 1, the highest rated bullshit explanation for each bullshit producer was taken to calculate a bullshit ability score (average of satisfyingness and accuracy ratings) as well as their perceived intelligence.

Design and Procedure

As in Study 1, participants began Study 2 by completing a bullshit willingness task in which they self-reported their knowledge of 10 (six real and four fake) concepts. Next, they were presented with 120 explanations of these concepts (produced by the Bullshit Producer sample of Study 1) and made judgments regarding the satisfactoriness and accuracy of each explanation and the intelligence of each explanation producer. Following all evaluation judgments, participants rated the profundity of 30 statements (10 pseudo-profound bullshit statements, 10 motivational quotations, and 10 mundane statements) and completed both the Raven’s Progressive Matrices and Wordsum tasks.

We conducted correlational analyses between our main variables of interest (see Table 2). As our focus remained on the characteristics (e.g., intelligence) of those producing bullshit, we once again focused exclusively on the associations within our Study 1 Bullshit Producers sample. Importantly, the results reported here feature judgments of Bullshit Producers’ bullshit ability and perceived intelligence, as judged exclusively by our Study 2 sample. Analyses examining the associations between the bullshit willingness, bullshit receptivity, and cognitive ability of our Study 2 (i.e., Bullshit Rater) sample can be viewed in the Online Supplemental Materials (Part B).

Table

Table 2. Study 2 Correlations.

Table 2. Study 2 Correlations.

Examining the hypothesized positive association between bullshit ability and intelligence, we find that bullshit ability was positively associated with verbal intelligence (as measured by the Wordsum), r(204) = .38, p < .001. Similarly, we observe a positive association between bullshit ability and abstract reasoning (as measured by RPM), r(203) = .31, p < .001. Furthermore, the perceived intelligence of Bullshit Producers was positively correlated with assessments of their bullshit ability, r(217) = .95, p < .001. This association is consistent with the hypothesis that producing satisfying and seemingly accurate explanations of completely fictional concepts is perceived by individuals as a signal of intelligence. Interestingly, the perceived intelligence of Bullshit Producers was negatively associated with their receptivity to pseudo-profound bullshit, r(217) = −.20, p = .003. Thus, those perceived as more intelligent on the basis of the bullshit they produced were less likely to themselves judge pseudo-profound bullshit as profound. Lastly, consistent with Study 1, we observed no association between bullshit ability and bullshit willingness, r(217) = −.05, p = .480. Therefore, those able to produce convincing bullshit were no more likely to report knowledge of fake concepts. This is surprising as one might expect that a person naturally skilled in producing bullshit would bullshit more often. However, individual factors such as honesty may prevent someone who would otherwise be a skilled bullshitter from fully making use of bullshit as a strategy.

A possible explanation for the observed modest associations between bullshit ability and cognitive ability is that while good bullshit producers may often be highly intelligent, the reverse inference may not be true. That is, a person who is unable to bullshit in a satisfying manner may not necessarily be unintelligent. By analogy to humor, a person who is funny is likely to be rather intelligent, however one can identify many brilliant people who are profoundly unfunny. This asymmetry may have resulted in an underestimation of the true strength of the association between bullshit ability and intelligence. Lending support to this claim, across both studies and both measures of cognitive ability, it is rare to find people who score low on measures of intelligence while simultaneously demonstrating high bullshit ability (circled regions in Figure 3). We interpret this as a demonstration that bullshit ability is a reliable indicator of when someone is intelligent, but that having low bullshitting ability does not necessarily mean that one is unintelligent.


                        figure

Figure 3. Overview of Intelligence and bullshitting ability results.

Note. Scatterplots comparing measures of cognitive ability to scores of bullshit ability for both Studies 1 and 2. Circled is the region representing people who score low on measures of intelligence but high on ability to bullshit. Note that this region is very sparsely populated.

The current work provides initial evidence for bullshit ability as an honest signal of intelligence. We find that the ability to create satisfying and seemingly accurate bullshit (e.g., explanations of fake concepts) was associated with obtaining higher scores on two measures of cognitive ability (i.e., the Wordsum and RPM). Interestingly, we find that one’s ability to produce satisfying bullshit is independent of one’s willingness to produce bullshit. Indeed, the two were uncorrelated in our studies, and had opposite associations with measures of intelligence. Others have found similar negative associations with measures of intelligence. For example, Pennycook and Rand (2019) found that overclaiming (arguably a form of bullshitting very similar to our bullshit willingness measure) was negatively correlated with performance on the Cognitive Reflection Task. Additionally, in a study by Littrell and colleagues (2021), intelligence (as indexed by Numeracy and Wordsum) was found to be negatively associated with persuasive bullshitting frequency.

It would seem logical that those who are better at bullshitting would opt to use it more frequently, however, we do not find this here. A possible explanation may be one which appeals to Theory of Mind models of intelligence. Of the three evolutionary pressures discussed in the introduction, the current set of studies has largely focused on a Machiavellian view, that intelligence affords us opportunities to deceive others to our advantage, as well as an IQ-signaling perspective, whereby bullshitting may be useful as an honest signal of a person’s quality or fitness through signaling their intelligence. We may lean on the third pressure to explain why it is that despite their superior ability to create bullshit, intelligent people seem to display less willingness to spontaneously engage in bullshitting. Part of this explanation may be that increased intelligence also results in a more sophisticated ability to simulate the mental states of others. In casual language, this may be described as “knowing your audience” and as such, they may possess a more sophisticated understanding of when and where bullshitting will work if attempted. Further, if highly intelligent people tend to associate with similarly intelligent people due to factors related to assortative mating, for example, intelligent people preferring intelligent mates or, “like pairs with like” (Thiessen & Gregg, 1980) or general homophily (McPherson et al., 2001) they may often find themselves around people who are likely to detect attempts at bullshitting, lowering its appeal as a first-order social strategy. As previous research has argued, a determiner of whether people will make an attempt to bullshit someone is whether they believe it will go undetected (Petrocelli, 2018). If smarter people are better able to know the contents of other people’s thoughts, they may be more carefully calibrated to the conditions under which an attempt at bullshitting will be unsuccessful. Of note, “bullshit ability,” as measured in our studies, involved the production of explanations for fake concepts, while “bullshit willingness” only required that the participant be willing to rate their knowledge of such fake concepts higher than “none.” Therefore, the lack of association we observed could be due to the specific methods selected to measure these two constructs. Future work should further dissociate the processes underlying one’s ability and willingness to produce bullshit.

While work has begun examining the degree to which personality may predict receptivity to bullshit (Bainbridge et al., 2019Čavojová et al., 2020), it has yet to be explored how personality influences the tendency or ability to bullshit. It could be the case that different personality traits (e.g., openness, honesty-humility, agreeableness; Lee & Ashton, 2004), moderate one’s willingness to engage in bullshitting. For example, a person who scores high in honesty-humility, a personality dimension which captures traits like sincerity, fairness, or modesty, may be less willing to bullshit, given that bullshitting is characterized by the desire to impress others without regard for the truth. The reverse may be true for those who are low in agreeableness, they may, especially when confronted with a disagreement, be more likely to deemphasize the importance of truth in favor of self-advancement through the use of bullshit. The numerous ways that common personality factors may interact in predicting the tendency and ability to bullshit makes for a promising topic of future exploration.

Regardless of whether bullshit ability honestly signals one’s intelligence, of potentially greater importance is that skilled bullshit producers are perceived by others as highly intelligent. From the perspective of navigating social systems, being perceived as intelligent may be just as valuable to an agent as actually being intelligent, as this perception may afford one opportunities to obtain status and form relationships as well as have greater trust placed in their competence. To this point, we observed a strong positive association between bullshit ability and perceived intelligence. However, this association was found in a situation in which those judging the intelligence of bullshit producers knew nothing of these individuals except their ability to produce satisfying explanations of real and fake concepts. Thus, it is likely that the strength of this association was overestimated in the present work as–with limited information–any signal of quality may have been exaggerated. In addition, as Bullshit Raters rated bullshit ability and perceived intelligence using similar 5-point scales, the strength of this association may be inflated due to unthoughtful responding by some participants (i.e., some participants may be inclined to simply select the same values on the scales).

Overall, we interpret these results as initial evidence that the ability to bullshit well provides an honest signal of a person’s ability to successfully navigate social systems, fitting the current work into existing frameworks whereby human intelligence is geared towards efficiently navigating such systems (Dunbar, 1998Crow, 1993). More specifically, we propose that the ability to produce satisfying bullshit may have emerged as an energetically efficient strategy for achieving an individual’s goals (such as acquiring status or impressing mates). That is, a person can engage in the arduous process of acquiring expert skills in domains that they could then leverage to accomplish certain goals, or can use bullshit as a strategy that potentially produces the same benefits at a much smaller cost (Turpin et al., 2019). Of course, these strategies need not be mutually exclusive, as the ability to produce satisfying bullshit may help even highly skilled individuals achieve their goals over equally skilled peers. This may be especially true in domains in which success depends largely on the subjective evaluations of others (e.g., art, advertising, politics, life coaching, journalism, humanities).

Limitations

An obvious limitation of the current work is its correlational nature, meaning that we cannot conclude that being more intelligent causes a person to be a better bullshitter. The current study merely provides preliminary evidence consistent with one plausible causal model. Future work should seek to explicitly probe the causal relation between intelligence and bullshit ability if any such relation exists. In addition, as noted above, the association between perceived intelligence and bullshit ability is likely overestimated in our sample due to the limited information available to the raters and the means of assessment. With respect to the latter, future research should include alternative metrics to assess perceived intelligence (e.g., estimating the actual IQ of bullshit producers using a number rather than a rating scale) to limit the possibility of unthoughtful responding contributing to the association.

The use of the WordSum and Raven’s Progressive Matrices made the conduct of the study possible given constraints on time. Independently, they predict IQ fairly well with correlations ranging between r = .55 and r = .66 between scores on the Wechsler Adult Intelligence Scale and Raven’s Matrices, and a correlation between Wordsum performance and IQ of r = .88 (Burke, 1985Malhotra et al., 2007McLaurin et al., 1973). However, more sophisticated measures for IQ would improve the accuracy of any cognitive ability measurement and therefore provide a more exact picture of the true relation between bullshit ability and cognitive ability. Relatedly, more opportunities to assess bullshit ability through either increasing the number of fake concepts participants were to bullshit about, or even better, using multiple different tasks which meet the criteria for “bullshitting” would improve our ability to draw conclusions about “bullshitting” behavior generally.

The bullshit generation task required participants to produce bullshit by explicitly directing them to ignore the truth. This is, under a Frankfurtian definition, “bullshit,” but this task is merely a substitute for the truly interesting question of how bullshit ability and cognitive ability relate in naturalistic settings, where bullshitting happens spontaneously. This artificial task is sufficient for establishing some initial evidence of the link between bullshit ability and cognitive ability, but more work is required to identify the nature of this relation.

The current work provides initial evidence for bullshit ability as an honest signal of intelligence. While much research has focused on the cognitive shortcomings of those receptive to bullshit (Čavojová et al., 2020Pennycook et al., 2015Walker et al., 2019), the current work focuses on the cognitive properties of bullshit producers. We find that those more skilled in producing satisfying and seemingly accurate bullshit score higher on measures of cognitive ability and are perceived by others as more intelligent. Overall, the ability to produce satisfying bullshit may serve to assist individuals in navigating social systems, both as an energetically efficient strategy for impressing others and as an honest signal of one’s intelligence.

Authors’ Note
The data that support the findings of this study are available from the corresponding author upon request.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by grants from The Natural Sciences and Engineering Research Council of Canada.

ORCID iD
Martin Harry Turpin  https://orcid.org/0000-0001-9655-4726

Supplemental Material
Supplemental material for this article is available online.

Bainbridge, T. F., Quinlan, J. A., Mar, R. A., Smillie, L. D., Fajkowska, M. (2019). Openness/intellect and susceptibility to pseudo–profound bullshit: A replication and extension. European Journal of Personality, 33(1), 7288.
Google Scholar | SAGE Journals
Beals, K. L., Smith, C. L., Dodd, S. M., Angel, J. L., Armstrong, E., Blumenberg, B., Girgis, F. G., Turkel, S., Gibson, K. R., Henneberg, M., Menk, R., Morimoto, I., Sokal, R. R., Trinkaus, E. (1984). Brain size, cranial morphology, climate, and time machines. Current Anthropology, 25(3), 301330.
Google Scholar | Crossref
Bednarik, R. G. (2014). Doing with less: Hominin brain atrophy. Homo, 65(6), 433449https://doi.org/10.1016/j.jchb.2014.06.001
Google Scholar
Bilker, W. B., Hansen, J. A., Brensinger, C. M., Richard, J., Gur, R. E., Gur, R. C. (2012). Development of abbreviated nine-item forms of the Raven’s standard progressive matrices test. Assessment, 19(3), 354369https://doi.org/10.1177/1073191112446655
Google Scholar
Bjorklund, D. F., Kipp, K. (2002). Social cognition, inhibition, and theory of mind: The evolution of human intelligence. In Sternberg, R. J., Kaufman, J. C. (Eds.), The evolution of intelligence (pp. 2754). Lawrence Erlbaum Associates Publishers.
Google Scholar
Burke, H. R. (1985). Raven's Progressive Matrices (1938): More on norms, reliability, and validity. Journal of Clinical Psychology, 41(2), 231235.
Google Scholar | Crossref
Byrne, R. W. (1996). Machiavellian intelligence. Evolutionary Anthropology: Issues, News, and Reviews: Issues, News, and Reviews, 5(5), 172180.
Google Scholar | Crossref
Byrne, R. W., Whiten, A. (1990). Machiavellian intelligence: Social expertise and the evolution of intellect in monkeys, apes, and humans. Behaviour and Philosophy, 18(1), 7375.
Google Scholar
Carson, T. (2016). Frankfurt and Cohen on bullshit, bullshitting, deception, lying, and concern with the truth of what one says. Pragmatics & Cognition, 23(1), 5468https://doi.org/10.1075/pc.23.1.03car
Google Scholar
Čavojová, V., Brezina, I., Jurkovič, M. (2020). Expanding the bullshit research out of pseudo-transcendental domain. Current Psychology, 110https://doi.org/10.1007/s12144-020-00617-3
Google Scholar
Crow, T. J. (1993). Sexual selection, Machiavellian intelligence, and the origins of psychosis. The Lancet, 342(8871), 594598https://doi.org/10.1016/0140-6736(93)91415-I
Google Scholar
Cunnane, S. C., Harbige, L. S., Crawford, M. A. (1993). The importance of energy and nutrient supply in human brain evolution. Nutrition and Health, 9(3), 219235https://doi.org/10.1177/026010609300900307
Google Scholar
Dunbar, R. I. (1998). The social brain hypothesis. Evolutionary Anthropology: Issues, News, and Reviews, 6(5), 178190.
Google Scholar | Crossref | ISI
Frankfurt, H. G. (2009). On bullshit. Princeton University Press.
Google Scholar | Crossref
Gavrilets, S., Vose, A. (2006). The dynamics of Machiavellian intelligence. Proceedings of the National Academy of Sciences, 103(45), 1682316828https://doi.org/10.1073/pnas.0601428103
Google Scholar | Crossref
Geher, G., Miller, G. (Eds.). (2007). Mating intelligence: Sex, relationships, and the mind’s reproductive system. Psychology Press.
Google Scholar | Crossref
Grabo, A., Spisak, B. R., van Vugt, M. (2017). Charisma as signal: An evolutionary perspective on charismatic leadership. The Leadership Quarterly, 28(4), 473485https://doi.org/10.1016/j.leaqua.2017.05.001
Google Scholar
Greengross, G., Miller, G. (2011). Humor ability reveals intelligence, predicts mating success, and is higher in males. Intelligence, 39(4), 188192https://doi.org/10.1016/j.intell.2011.03.006
Google Scholar
Handel, M. I. (1982). Intelligence and deception. The Journal of Strategic Studies, 5(1), 122154https://doi.org/10.1080/01402398208437104
Google Scholar
Humphrey, N. K. (1976). The social function of intellect. In Bateson, P. P. G., Hinde, R. A. (Eds.), Growing points in ethology (pp. 303317). Cambridge University Press.
Google Scholar
Jerrim, J., Parker, P., Shure, D. (2019April). Bullshitters. Who are they and what do we know about their lives? [IZA Discussion Paper No. 12282]. https://ssrn.com/abstract=3390272
Google Scholar
Lee, K., Ashton, M. C. (2004). Psychometric properties of the HEXACO personality inventory. Multivariate Behavioral Research, 39(2), 329358https://doi.org/10.1207/s15327906mbr3902_8
Google Scholar
Littrell, S., Risko, E. F., Fugelsang, J. A. (2020). The Bullshitting Frequency Scale: Development and psychometric properties. British Journal of Social Psychology. https://doi.org/10.1111/bjso.12379
Google Scholar
Littrell, S., Risko, E. F., Fugelsang, J. A. (2021). “You can’t bullshit a bullshitter” (or can you?): Bullshitting frequency predicts receptivity to various types of misleading information. British Journal of Social Psychology. Advance online publication. https://doi.org/10.1111/bjso.12447
Google Scholar
Malhotra, N., Krosnick, J. A., Haertel, E. (2007). The psychometric properties of the GSS wordsum vocabulary test. GSS Methodological Report, 11, 163.
Google Scholar
McCarthy, I. P., Hannah, D., Pitt, L. F., McCarthy, J. M. (2020). Confronting indifference toward truth: Dealing with workplace bullshit. Business Horizons, 63(3), 253263https://doi.org/10.1016/j.bushor.2020.01.001
Google Scholar
McKeown, G. J. (2013). The analogical peacock hypothesis: The sexual selection of mind-reading and relational cognition in human communication. Review of General Psychology, 17(3), 267287https://doi.org/10.1037/a0032631
Google Scholar
McLaurin, W. A., Jenkins, J. F., Farrar, W. E., Rumore, M. C. (1973). Correlations of IQs on verbal and nonverbal tests of intelligence. Psychological Reports, 33(3), 821822.
Google Scholar | SAGE Journals | ISI
McNally, L., Brown, S. P., Jackson, A. L. (2012). Cooperation and the evolution of intelligence. Proceedings of the Royal Society B: Biological Sciences, 279(1740), 30273034https://doi.org/10.1098/rspb.2012.0206
Google Scholar | Crossref
McPherson, M., Smith-Lovin, L., Cook, J. M. (2001). Birds of a feather: Homophily in social networks. Annual Review of Sociology, 27(1), 415444.
Google Scholar | Crossref | ISI
Miller, G. (2000October). Sexual selection for indicators of intelligence. In Novartis foundation symposium (pp. 260270). John Wiley, 1999.
Google Scholar | Crossref
Miller, G. F., Todd, P. M. (1998). Mate choice turns cognitive. Trends in Cognitive Sciences, 2(5), 190198https://doi.org/10.1016/S1364-6613(98)01169-3
Google Scholar
Pennycook, G., Cheyne, J. A., Barr, N., Koehler, D. J., Fugelsang, J. A. (2015). On the reception and detection of pseudo-profound bullshit. Judgment and Decision making, 10(6), 549563.
Google Scholar | ISI
Pennycook, G., Rand, D. G. (2019). Who falls for fake news? The roles of bullshit receptivity, overclaiming, familiarity, and analytic thinking. Journal of Personality, 88, 185200https://doi.org/10.1111/jopy.12476
Google Scholar
Petrocelli, J. V. (2018). Antecedents of bullshitting. Journal of Experimental Social Psychology, 76, 249258.
Google Scholar | Crossref
Raichle, M. E., Gusnard, D. A. (2002). Appraising the brain’s energy budget. Proceedings of the National Academy of Sciences, 99(16), 1023710239https://doi.org/10.1073/pnas.172399499
Google Scholar | Crossref
Rosenberg, K., Trevathan, W. (2002). Birth, obstetrics and human evolution. BJOG: An International Journal of Obstetrics & Gynaecology, 109(11), 11991206https://doi.org/10.1046/j.1471-0528.2002.00010.x
Google Scholar
Roth, G., Dicke, U. (2005). Evolution of the brain and intelligence. Trends in Cognitive Sciences, 9(5), 250257.
Google Scholar | Crossref | Medline
Schillaci, M. A. (2006). Sexual selection and the evolution of brain size in primates. PLoS One, 1(1), 15https://doi.org/10.1371/journal.pone.0000062
Google Scholar
Seyfarth, R. M., Cheney, D. L. (2002). What are big brains for? Proceedings of the National Academy of Sciences, 99(7), 41414142https://doi.org/10.1073/pnas.082105099
Google Scholar | Crossref
Sharma, S., Bottom, W. P., Elfenbein, H. A. (2013). On the role of personality, cognitive ability, and emotional intelligence in predicting negotiation outcomes: A meta-analysis. Organizational Psychology Review, 3(4), 293336https://doi.org/10.1177/2041386613505857
Google Scholar
Stone, V. E. (2006). Theory of mind and the evolution of social intelligence. In Cacioppo, J. T., Visser, P. S., Pickett, C. L. (Eds.), Social neuroscience: People thinking about thinking people (pp. 103129). MIT Press.
Google Scholar
Thiessen, D., Gregg, B. (1980). Human assortative mating and genetic equilibrium: An evolutionary perspective. Ethology and Sociobiology, 1(2), 111140.
Google Scholar | Crossref
Turpin, M. H., Walker, A. C., Kara-Yakoubian, M., Gabert, N. N., Fugelsang, J. A., Stolz, J. A. (2019). Bullshit makes the art grow profounder. Judgment and Decision Making, 14(6), 658670.
Google Scholar
Walker, A. C., Turpin, M. H., Stolz, J. A., Fugelsang, J. A., Koehler, D. J. (2019). Finding meaning in the clouds: Illusory pattern perception predicts receptivity to pseudo-profound bullshit. Judgment and Decision Making, 14(2), 109119.
Google Scholar
Whiten, A. (2018). Social, Machiavellian and cultural cognition: A golden age of discovery in comparative and evolutionary psychology. Journal of Comparative Psychology, 132(4), 437441https://doi.org/10.1037/com0000135
Google Scholar

limits of math, logic, computing, and artificial intelligence



1931: Kurt Gödel, founder of theoretical computer science, shows limits of math, logic, computing, and artificial intelligence


Abstract. In 2021, we are celebrating the 90th anniversary of Kurt Gödel's groundbreaking 1931 paper which laid the foundations of theoretical computer science and the theory of artificial intelligence (AI). Gödel sent shock waves through the academic community when he identified the fundamental limits of theorem proving, computing, AI, logics, and mathematics itself. This had enormous impact on science and philosophy of the 20th century. Ten years to go until the Gödel centennial in 2031!


Kurt Goedel, founder of theoretical computer science around 1931In the early 1930s, Kurt Gödel articulated the mathematical foundation and limits of computing, computational theorem proving, and logic in general.[GOD][GOD34][GOD21,21a] Thus he became the father of modern theoretical computer science and AI theory.

Gödel introduced a universal language to encode arbitrary formalizable processes.[GOD][GOD34] It was based on the integers, and allows for formalizing the operations of any digital computer in axiomatic form (this also inspired my much later self-referential Gödel Machine[GM6]). Gödel used his so-called Gödel Numbering to represent both data (such as axioms and theorems) and programs[VAR13] (such as proof-generating sequences of operations on the data).

Gödel famously constructed formal statements that talk about the computation of other formal statements—especially self-referential statements which imply that they are not decidable, given a computational theorem prover that systematically enumerates all possible theorems from an enumerable set of axioms. Thus he identified fundamental limits of algorithmic theorem proving, computing, and any type of computation-based AI[GOD] (some misunderstood his result and thought he showed that humans are superior to AIs[BIB3]). Much of early AI in the 1940s-70s was about theorem proving[ZU48][NS56] and deduction in Gödel style through expert systems and logic programming.

Leibniz, father of computer science around 1670Like most great scientists, Gödel built on earlier work. He combined Georg Cantor's diagonalization trick[CAN] (which showed in 1891 that there are different types of infinities) with the foundational work by Gottlob Frege[FRE] (who introduced the first formal language in 1879), Thoralf Skolem[SKO23] (who introduced primitive recursive functions in 1923) and Jacques Herbrand[GOD86] (who identified limitations of Skolem's approach). These authors in turn built on the formal Algebra of Thought (1686) by Gottfried Wilhelm Leibniz,[L86][WI48] which is deductively equivalent[LE18] to the later Boolean Algebra of 1847.[BOO] Leibniz, one of the candidates for the title of "father of computer science,"[LEI21,21a] has been called "the world's first computer scientist"[LA14] and even "the smartest man who ever lived."[SMO13] He described the principles of binary computers governed by punch cards (1679).[L79][LA14][HO66][L03][IN08][SH51][LEI21,21a] In 1673, he designed the first physical hardware (the step reckoner) that could perform all four arithmetic operations, and the first with a memory,[BL16] going beyond the first automatic gear-based data-processing calculators by Wilhelm Schickard (1623) and Blaise Pascal (1642). Leibniz was not only the first to publish infinitesimal calculus,[L84] but also pursued an ambitious project to answer all possible questions through computation. His ideas on a universal language and a general calculus for reasoning were extremely influential (Characteristica Universalis & Calculus Ratiocinator,[WI48] inspired by the 13th century scholar Ramon Llull[LL7]). Leibniz' "Calculemus!" is one of the defining quotes of the age of enlightenment: "If controversies were to arise, there would be no more need of disputation between two philosophers than between two accountants. For it would suffice to take their pencils in their hands, to sit down with their slates and say to each other [...]: Let us calculate!"[RU58] In 1931, however, Gödel showed that there are fundamental limitations to what is decidable or computable in this way.[GOD][MIR](Sec. 18)

Alonzo Church extended Goedel's results to the EntscheidungsproblemIn 1935, Alonzo Church derived a corollary / extension of Gödel's result by showing that Hilbert & Ackermann's famous Entscheidungsproblem (decision problem) does not have a general solution.[CHU] To do this, he used his alternative universal coding language called Untyped Lambda Calculus, which forms the basis of the highly influential programming language LISP.

In 1936, Alan Turing introduced yet another universal model which has become perhaps the most well-known of them all (at least in computer science): the Turing Machine.[TUR] He rederived the above-mentioned result.[T20](Sec. IV) Of course, he cited both Gödel and Church in his 1936 paper[TUR] (whose corrections appeared in 1937). In the same year of 1936, Emil Post published yet another independent universal model of computing,[POS] also citing Gödel and Church. Today we know many such models. Nevertheless, according to Wang,[WA74-96] it was Turing's work (1936) that convinced Gödel of the universality of both his own approach (1931-34) and Church's (1935).

Alan TuringWhat exactly did Post[POS] and Turing[TUR] do in 1936 that hadn't been done earlier by Gödel[GOD][GOD34] (1931-34) and Church[CHU] (1935)? There is a seemingly minor difference whose significance emerged only later. Many of Gödel's instruction sequences were series of multiplications of number-coded storage contents by integers. Gödel did not care that the computational complexity of such multiplications tends to increase with storage size. Similarly, Church also ignored the spatio-temporal complexity of the basic instructions in his algorithms. Turing and Post, however, adopted a traditional, reductionist, minimalist, binary view of computing—just like Konrad Zuse (1936).[ZU36] Their machine models permitted only very simple elementary instructions with constant complexity, like the early binary machine model of Leibniz (1679).[L79][LA14][HO66] Emil PostThey did not exploit this back then—for example, in 1936, Turing used his (quite inefficient) model only to rephrase the results of Gödel and Church on the limits of computability. Later, however, the simplicity of these machines made them a convenient tool for theoretical studies of complexity. (I also happily used and generalized them for the case of never-ending computations.[ALL2])

The Gödel Prize for theoretical computer science is named after Gödel. The currently more lucrative ACM A. M. Turing Award was created in 1966 for contributions "of lasting and major technical importance to the computer field." It is funny—and at the same time embarrassing—that Gödel (1906-1978) never got one, although he not only laid the foundations of the "modern" version of the field, but also identified its most famous open problem "P=NP?" in his famous letter to John von Neumann (1956).[GOD56][URQ10]

Konrad Zuse created the world's first working programmable computer 1935-41The formal models of Gödel (1931-34), Church (1935), Turing (1936), and Post (1936) were theoretical pen & paper constructs that cannot directly serve as a foundation for practical computers. Remarkably, Konrad Zuse's patent application[ZU36-38][Z36][RO98] for the first practical general-purpose program-controlled computer also dates back to 1936. It describes general digital circuits (and predates Claude Shannon's 1937 thesis on digital circuit design[SHA37]). Then, in 1941, Zuse completed Z3, the world's first practical, working, programmable computer (based on the 1936 application). Ignoring the inevitable storage limitations of any physical computer, the physical hardware of Z3 was indeed universal in the "modern" sense of Gödel, Church, Turing, and Post—simple arithmetic tricks can compensate for Z3's lack of an explicit conditional jump instruction.[RO98] Zuse also created the first high-level programming language (Plankalkül)[BAU][KNU] in the early 1940s. He applied it to chess in 1945[KNU] and to theorem proving in 1948.[ZU48]

It should be mentioned that practical AI is much older than Gödel's theoretical analysis of the fundamental limitations of AI. In 1914, the Spaniard Leonardo Torres y Quevedo was the 20th century's first pioneer of practical AI when he built the first working chess end game player (back then chess was considered as an activity restricted to the realms of intelligent creatures). The machine was still considered impressive decades later when the AI pioneer Norbert Wiener played against it at the 1951 Paris conference,[AI51][BRO21] [BRU1-4] which is now often viewed as the first conference on AI—though the expression "AI" was coined only later in 1956 at another conference in Dartmouth by John McCarthy. In fact, in 1951, much of what's now called AI was still called Cybernetics, with a focus very much in line with modern AI based on deep neural networks.[DL1-2][DEC]

Likewise, it should be mentioned that practical computer science is much older than Gödel's foundations of theoretical computer science (compare the comments on Leibniz above). Perhaps the world's first practical programmable machine was an automatic theatre made in the 1st century[SHA7a][RAU1] by Heron of Alexandria (who apparently also had the first known working steam engine—the Aeolipile). The energy source of his programmable automaton was a falling weight pulling a string wrapped around pins of a revolving cylinder. Complex instruction sequences controlling doors and puppets for several minutes were encoded by complex wrappings. The 9th century music automaton by the Banu Musa brothers in Baghdad was perhaps the first machine with a stored program.[BAN][KOE1] It used pins on a revolving cylinder to store programs controlling a steam-driven flute—compare Al-Jazari's programmable drum machine of 1206.[SHA7b] The first commercial program-controlled machines (punch card-based looms) were built in France around 1800 by Joseph-Marie Jacquard and others—perhaps the first "modern" programmers who wrote the world's first industrial software. They inspired Ada Lovelace and her mentor Charles Babbage (UK, circa 1840) who planned but was unable to build a non-binary, decimal, programmable, general purpose computer. The first general working programmable machine built by someone other than Zuse (1941)[RO98] was Howard Aiken's decimal MARK I (US, 1944).

Gödel has often been called the greatest logician since Aristotle.[GOD10] At the end of the previous millennium, TIME magazine ranked him as the most influential mathematician of the 20th century, although some mathematicians say his most important result was about logic and computing, not math. Others call it the fundamental result of theoretical computer science, a discipline that did not yet officially exist back then but effectively came about through Gödel's efforts. The Pulitzer Prize-winning popular book "Gödel, Escher, Bach"[H79] helped to inspire generations of young people to study computer science.

In 2021, we are not only celebrating the 90th anniversary of Gödel's famous 1931 paper but also the 80th anniversary of the world's first functional general-purpose program-controlled computer by Zuse (1941). It seems incredible that within less than a century something that once lived only in the minds of titans has become something so inalienable from modern society. The world owes these scientists a great debt. Ten years to go until the Gödel centennial in 2031, and twenty years until the Zuse centennial in 2041! Enough time to plan appropriate celebrations.


Acknowledgments

Creative Commons LicenseThanks to Moshe Vardi, Herbert Bruderer, Jack Copeland, Wolfgang Bibel, Teun Koetsier, Scott Aaronson, Dylan Ashley, Sebastian Oberhoff, Kai Hormann, and several other expert reviewers for useful comments. Since science is about self-correction, let me know under juergen@idsia.ch if you can spot any remaining error. The contents of this article may be used for educational and non-commercial purposes, including articles for Wikipedia and similar sites. This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.


References

[GOD] K. Gödel. Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme I. Monatshefte für Mathematik und Physik, 38:173-198, 1931.

[GOD34] K. Gödel (1934). On undecidable propositions of formal mathematical systems. Notes by S. C. Kleene and J. B. Rosser on lectures at the Institute for Advanced Study, Princeton, New Jersey, 1934, 30 pp. (Reprinted in M. Davis, (ed.), The Undecidable. Basic Papers on Undecidable Propositions, Unsolvable Problems, and Computable Functions, Raven Press, Hewlett, New York, 1965.)

[GOD56] R. J. Lipton and K. W. Regan. Gödel's lost letter and P=NP. Link.

[GOD86] K. Gödel. Collected works Volume I: Publications 1929-36, S. Feferman et. al., editors, Oxford Univ. Press, Oxford, 1986.

[GOD10] V. C. Nadkarni. Gödel, Einstein and proof for God. The Economic Times, 2010.

[URQ10] A. Urquhart. Von Neumann, Gödel and complexity theory. Bulletin of Symbolic Logic 16.4 (2010): 516-530. Link.

[BIB3] W. Bibel (2003). Mosaiksteine einer Wissenschaft vom Geiste. Invited talk at the conference on AI and Gödel, Arnoldsheim, 4-6 April 2003. Manuscript, 2003.

[CHU] A. Church (1935). An unsolvable problem of elementary number theory. Bulletin of the American Mathematical Society, 41: 332-333. Abstract of a talk given on 19 April 1935, to the American Mathematical Society. Also in American Journal of Mathematics, 58(2), 345-363 (1 Apr 1936). [First explicit proof that the Entscheidungsproblem (decision problem) does not have a general solution.]

[TUR] A. M. Turing. On computable numbers, with an application to the Entscheidungsproblem. Proceedings of the London Mathematical Society, Series 2, 41:230-267. Received 28 May 1936. Errata appeared in Series 2, 43, pp 544-546 (1937).

[POS] E. L. Post (1936). Finite Combinatory Processes - Formulation 1. Journal of Symbolic Logic. 1: 103-105. Link.

[WA74] H. Wang (1974). From Mathematics to Philosophy, New York: Humanities Press.

[WA96] H. Wang (1996). A Logical Journey: From Gödel to Philosophy, Cambridge, MA: MIT Press.

[H79] Douglas R. Hofstadter (1979). Gödel, Escher, Bach: An Eternal Golden Braid. Basic Books, ISBN 0-465-02656-7, 1979.

[FRE] G. Frege (1879). Begriffsschrift: eine der arithmetischen nachgebildete Formelsprache des reinen Denkens. Halle an der Saale: Verlag Louis Nebert. [The first formal language / formal proofs—basis of modern logic and programming languages.]

[SKO23] T. Skolem (1923). Begründung der elementaren Arithmetik durch die rekurrierende Denkweise ohne Anwendung scheinbarer Veränderlichen mit unendlichem Ausdehnungsbereich. Skrifter utgit av Videnskapsselskapet i Kristiania, I. Mathematisk-Naturvidenskabelig Klasse 6 (1923), 38 pp.

[CAN] G. Cantor (1891). Ueber eine elementare Frage der Mannigfaltigkeitslehre. Jahresbericht der Deutschen Mathematiker-Vereinigung, 1:75-78. [English translation: W. B. Ewald (ed.). From Immanuel Kant to David Hilbert: A Source Book in the Foundations of Mathematics, Volume 2, pp. 920-922. Oxford University Press, 1996.]

[L79] G. Leibniz. De Progressione dyadica Pars I. 15 March 1679. [Principles of binary computers governed by punch cards.]

[L03] G. Leibniz (1703). In: Explication de l'Arithmetique Binaire / Die Mathematischen Schriften, ed. C. Gerhardt, Berlin 1879, vol.7, p.223. English link[Leibniz documented the binary arithmetics which allow for greatly simplifiying computing hardware and are employed by virtually all modern computers. Binary number encodings per se, however, seem to date back over 4000 years.]

[L84] G. Leibniz (1684). Nova Methodus pro Maximis et Minimis. [First publication on infinitesimal calculus.]

[L86] G. Leibniz (1686). Generales Inquisitiones de analysi notionum et veritatum. Also in Leibniz: Die philosophischen Schriften VII, 1890, pp. 236-247; translated as "A Study in the Calculus of Real Addition" (1690) by G. H. R. Parkinson, Leibniz: Logical Papers—A Selection, Oxford 1966, pp. 131-144.

[LEI21] J. Schmidhuber (2021). 375th birthday of Leibniz, founder of computer science.

[LEI21a] J. Schmidhuber (2021). Der erste Informatiker. Wie Gottfried Wilhelm Leibniz den Computer erdachte. (The first computer scientist. How Gottfried Wilhelm Leibniz conceived the computer.) Frankfurter Allgemeine Zeitung (FAZ), 17/5/2021. FAZ online: 19/5/2021.

[BOO] George Boole (1847). The Mathematical Analysis of Logic, Being an Essay towards a Calculus of Deductive Reasoning. London, England: Macmillan, Barclay, & Macmillan, 1847.

[LL7] A. Bonner (2007). The art and logic of Ramon Llull. Brill Academic Pub, p. 290, 2007.

[RU58] B. Russell (1958). The Philosophy of Leibniz. London: George Allen and Unwin, 1958.

[LE18] W. Lenzen. Leibniz and the Calculus Ratiocinator. Technology and Mathematics, pp 47-78, Springer, 2018.

[LA14] D. R. Lande (2014). Development of the Binary Number System and the Foundations of Computer Science. The Mathematics Enthusiast, vol. 11(3):6 12, 2014. Link.

[BL16] L. Bloch (2016). Informatics in the light of some Leibniz's works. Communication to XB2 Berlin Xenobiology Conference.

[HO66] E. Hochstetter et al. (1966): Herrn von Leibniz' Rechnung mit Null und Eins. Berlin: Siemens AG.

[IN08] R. Ineichen (2008). Leibniz, Caramuel, Harriot und das Dualsystem. Mitteilungen der deutschen Mathematiker-Vereinigung. 16(1):12-15.

[SH51] J. W. Shirley (1951). Binary Numeration before Leibniz. American Journal of Physics 19 (452-454).

[WI48] N. Wiener (1948). Time, communication, and the nervous system. Teleological mechanisms. Annals of the N.Y. Acad. Sci. 50 (4): 197-219. [Quote: "The history of the modern computing machine goes back to Leibniz and Pascal. Indeed, the general idea of a computing machine is nothing but a mechanization of Leibniz's calculus ratiocinator."]

[SMO13] L. Smolin (2013). My hero: Gottfried Wilhelm von Leibniz. The Guardian, 2013. Link[Quote: "And this is just the one part of Leibniz's enormous legacy: the philosopher Stanley Rosen called him "the smartest person who ever lived"."]

[T20] J. Schmidhuber (2020). Critique of 2018 Turing Award.

[HIN] J. Schmidhuber (2020). Critique of 2019 Honda Prize.

[GOD21] J. Schmidhuber (2021). 90th anniversary celebrations: 1931: Kurt Gödel, founder of theoretical computer science, shows limits of math, logic, computing, and artificial intelligence

[GOD21a] J. Schmidhuber (2021). Als Kurt Gödel die Grenzen des Berechenbaren entdeckte. (When Kurt Gödel discovered the limits of computability.) Frankfurter Allgemeine Zeitung, 16/6/2021.

[ALL2] J. Schmidhuber (2000). Algorithmic theories of everything. ArXiv: quant-ph/ 0011122. More. See also: Hierarchies of generalized Kolmogorov complexities and nonenumerable universal measures computable in the limit. International Journal of Foundations of Computer Science 13(4):587-612, 2002. PDFMore. See also: The Speed Prior: A New Simplicity Measure Yielding Near-Optimal Computable Predictions. Proc. COLT 2002. PDFMore.

Goedel Machine[GM6] J. Schmidhuber (2006). Gödel machines: Fully Self-Referential Optimal Universal Self-Improvers. In B. Goertzel and C. Pennachin, eds.: Artificial General Intelligence, p. 199-226, 2006. PDF. Preprint arXiv:cs/0309048 (2003). See also: Ultimate Cognition à la Gödel. Cognitive Computation 1(2):177-193, 2009. PDFMore.

[DL1] J. Schmidhuber, 2015. Deep Learning in neural networks: An overview. Neural Networks, 61, 85-117. More.

[DL2] J. Schmidhuber, 2015. Deep Learning. Scholarpedia, 10(11):32832.

[MIR] J. Schmidhuber (2019). Deep Learning: Our Miraculous Year 1990-1991. Preprint arXiv:2005.05744, 2020.

[DEC] J. Schmidhuber (2020). The 2010s: Our Decade of Deep Learning / Outlook on the 2020s.

[VAR13] M. Y. Vardi (2013). Who begat computing? Communications of the ACM, Vol. 56(1):5, Jan 2013. Link.

[ZU36] K. Zuse (1936). Verfahren zur selbsttätigen Durchführung von Rechnungen mit Hilfe von Rechenmaschinen. Patent application Z 23 139 / GMD Nr. 005/021, 1936. [First patent application describing a general, practical, program-controlled computer.]

[ZU37] K. Zuse (1937). Einführung in die allgemeine Dyadik. [Mentions the storage of program instructions in the computer's memory.]

[ZU38] K. Zuse (1938). Diary entry of 4 June 1938. [Description of computer architectures that put both program instructions and data into storage—compare the later "von Neumann" architecture.[NEU45]]

[ZU48] K. Zuse (1948). Über den Plankalkül als Mittel zur Formulierung schematisch kombinativer Aufgaben. Archiv der Mathematik 1(6), 441-449 (1948). PDF. [Apparently the first practical design of an automatic theorem prover (based on Zuse's high-level programming language Plankalkül).]

[NS56] A. Newell and H. Simon. The logic theory machine—A complex information processing system. IRE Transactions on Information Theory 2.3 (1956):61-79.

[RO98] R. Rojas (1998). How to make Zuse's Z3 a universal computer. IEEE Annals of Computing, vol. 19:3, 1998.

[BAU] F. L. Bauer, H. Woessner (1972). The "Plankalkül" of Konrad Zuse: A Forerunner of Today's Programming Languages.

[KNU] D. E. Knuth, L. T. Pardo (1976). The Early Development of Programming Languages. Stanford University, Computer Science Department. PDF.

[Z36] S. Faber (2000). Konrad Zuses Bemühungen um die Patentanmeldung der Z3.

[SHA37] C. E. Shannon (1938). A Symbolic Analysis of Relay and Switching Circuits. Trans. AIEE. 57 (12): 713-723. Based on his thesis, MIT, 1937.

[AI51] Les Machines a Calculer et la Pensee Humaine: Paris, 8.-13. Januar 1951, Colloques internationaux du Centre National de la Recherche Scientifique; no. 37, Paris 1953. [H. Bruderer rightly calls that the first conference on AI.]

[BRU1] H. Bruderer. Computing history beyond the UK and US: selected landmarks from continental Europe. Communications of the ACM 60.2 (2017): 76-84.

[BRU2] H. Bruderer. Meilensteine der Rechentechnik. 2 volumes, 3rd edition. Walter de Gruyter GmbH & Co KG, 2020.

[BRU3] H. Bruderer. Milestones in Analog and Digital Computing. 2 volumes, 3rd edition. Springer Nature Switzerland AG, 2020.

[BRU4] H. Bruderer. The Birthplace of Artificial Intelligence? Communications of the ACM, BLOG@CACM, Nov 2017. Link.

[BRO21] D. C. Brock (2021). Cybernetics, Computer Design, and a Meeting of the Minds. An influential 1951 conference in Paris considered the computer as a model of—and for—the human mind. IEEE Spectrum, 2021. Link.

[BAN] Banu Musa brothers (9th century). The book of ingenious devices (Kitab al-hiyal). Translated by D. R. Hill (1979), Springer, p. 44, ISBN 90-277-0833-9. [Perhaps the Banu Musa music automaton was the world's first machine with a stored program.]

[KOE1] [21] T. Koetsier (2001). On the prehistory of programmable machines: musical automata, looms, calculators. Mechanism and Machine Theory, Elsevier, 36 (5): 589-603, 2001.

[RAU1] M. Rausch. Heron von Alexandria. Die Automatentheater und die Erfindung der ersten antiken Programmierung. Diplomica Verlag GmbH, Hamburg 2012. [Perhaps the world's first programmable machine was an automatic theatre made in the 1st century by Heron of Alexandria, who apparently also had the first known working steam engine.]

[SHA7a] N. Sharkey (2007). A programmable robot from AD 60. New Scientist, Sept 2017.

[SHA7b] N. Sharkey (2007). A 13th Century Programmable Robot. Univ. of Sheffield, 2007. [On a programmable drum machine of 1206 by Al-Jazari.]

[LIL1] US Patent 1745175 by Austrian physicist Julius Edgar Lilienfeld for work done in Leipzig: "Method and apparatus for controlling electric current." First filed in Canada on 22.10.1925. [The patent describes a field-effect transistor. Today, almost all transistors are field-effect transistors.]

[LIL2] US Patent 1900018 by Austrian physicist Julius Edgar Lilienfeld: "Device for controlling electric current." Filed on 28.03.1928. [The patent describes a thin film field-effect transistor.]
.

Highlights of over 2000 years of computing history. Juergen Schmidhuber. 











.
1931: Theoretical Computer Science & AI Theory Founded by Goedel. Juergen Schmidhuber.
Jürgen Schmidhuber (June 2021)
Pronounce: You_again Shmidhoobuh
German version: FAZ, 16 June 2021
AI Blog
@SchmidhuberAI

6/13/2021

Pikachu’s Transmedia Adventures

 


CFP: Pikachu’s Transmedia Adventures: The Continuing Adaptability of the Pokemon Franchise

By 

In 2021, the Pokemon franchise celebrates the 25th anniversary of its debut in Japan and the fifth anniversary of its popular worldwide AR cellphone game Pokemon Go. In fact, Pokemon is arguably experiencing something of a resurgence and renaissance within the current cultural moment. When a pop-up Pokemon Centre store was opened in London in 2018 to mark the release of Sword and Shield, queues for entering the retail space frequently had to be closed due to demand whilst product lines regularly sold out on a daily basis. In 2019, when the long-running cartoon’s main character Ash Ketchum finally won a Pokemon tournament, major news sites humorously deemed this victory a newsworthy event (Bissett 2019). More recently, a revival in Pokemon card collecting has left retail shelves bare and scalpers running rampant whilst mint-condition ‘graded’ cards have sold for hundreds of thousands of dollars at auction (Koebler 2021). Meanwhile, the games themselves continue to be adapted to Nintendo’s console platforms, with the Nintendo Switch releasing both remakes of previously popular titles (Pokemon Let’s Go! Pikachu and Let’s Go! Eevee, Pokemon Snap) as well as new titles exploring hitherto unknown regions (Pokemon Sword and Shield). Much more than a franchise intended to commercially target and exploit children, the Pokemon franchise represents an enduringly popular intellectual property that continues to attract interest across generations. 

Despite this, in-depth and continuous academic study of this hugely popular intellectual property has been infrequent at best. In fact, the last time that a dedicated collection of essays exploring the franchise in a holistic manner was published was in 2004, with many of the contributors positioning the property as a ‘fad’ whose cycle of popularity was apparently at its end (see Tobin 2004; N.B. the augmented reality game Pokemon Go (Niantic 2016- ) has bucked this trend by generating considerable academic attention – see Kulak, Purzycki, Henthorn and Vie 2019; Saker and Evans 2021). Where Pokemon has attracted infrequent academic discussion, this has occurred in the context of assessing how wider cultural flows from Japan to the West have impacted on children’s media (Allison 2006; O’Melia 2020). What is absent, then, is a volume that takes the Pokemon franchise on its own terms and which situates the property within a much-changed media environment. Thus, a study is needed which considers Pokemon in terms of multiple contemporary debates within media and cultural studies. These include – but are no way limited to – cultural, technological, and media convergence (Jenkins 2006), discourses of transmediality and media mix (Steinberg 2012; Williams 2020), paratextuality (Gray 2010), licensing and/or (transgenerational) media industries studies (Santo 2015; Johnson 2019), material culture (Geraghty 2014; Bainbridge 2017) and fan cultures (Scott 2019; Stanfill 2019). Whether approached as a transmedia franchise, corporate intellectual property, system offering ludic possibilities, fan community, or otherwise, academic scholarship should better consider how the Pokemon franchise has engaged with, adapted to, and challenged the contours of the ever-evolving transmedia environment.

This call for papers seeks abstracts of 300-500 words for chapters of approx. 6000 words that explore topics including (but not are limited to):

  • The Industrial development of The Pokemon Company and its corporate relations with Nintendo and other licensed partners.
  • Pokemon and the historical development of media industries studies.
  • The evolution of Pokemon: The Card Game and its relationship to industrial contexts.
  • The evolution of the Pokemon computer games (e.g. games studies perspectives; remediation relating to Let’s Go!, Snap, etc.)
  • Pokemon and/as character licensing.
  • Pokemon and transmedia storytelling and/as transmedia text.
  • Pokemon, transmedia tourismand the Experience Economy (e.g. the Pokemon Cafe; the annual Pikachu Parade).
  • Pokemon Go and developments in augmented reality experiences and/or the gamification of space.
  • Detective Pikachu and Pokemon’s other cinematic adaptations.
  • Pokemon’s historical developments as anime.
  • Pokemon’s historical developmentsas manga
  • Pokemon and/as fan fashion (e.g. high-fashion licensees, jewelry, make-up).
  • Pokemon and/as paratextual theory.
  • Interventions concerning Pokemon and identity politics (e.g. feminism, critical race theory, queer theory).
  • Pokemon and/as the global expansion of kawaii/cute culture.
  • Thematic analyses of the Pokemon franchise (e.g. its ties with environmentalism).
  • Pokemon’s links to Japanese ‘soft power’.
  • Fan practices and transformative works related to the Pokemon franchise across multiple forms and platforms.
  • Pokemon and/as children’s culture.

We are especially interested in soliciting chapters featuring non-Western perspectives as well as ones engaging with historically marginalised or underrepresented groups. 

We hope to include work from both established and emerging scholars; junior scholars & graduate students are encouraged to apply.

Please email abstracts of 300-500 words with an accompanying Author Bio of approx. 150 words to Ross Garner (GarnerRP1@Cardiff.ac.uk) and EJ Nielsen (ejnielsen.ephemera@gmail.com) by 27 August, 2021.


https://fanstudies.org/


6/10/2021

Generative Pre-trained Transformer 3

 

Generative Pre-trained Transformer 3 (GPT-3) is an autoregressive language model that generates text using algorithms that are pre-trained. It was created by OpenAI (a research business co-founded by Elon Musk) and has been described as the most important and useful advance in AI for years.

Last summer writer, speaker, and musician, K Allado-McDowell initiated a conversation with GPT-3 which became the collection of poetry and prose Pharmako-AI. Taking this collection as her departure point, Warburg PhD student Beatrice Bottomley reflects on what GPT-3 means for how we think about writing and meaning.

 

GPT-3 is just over nine months old now. Since the release of its beta version by the California- based company Open AI in June 2020, the language model has been an object of fascination for both technophiles and, to a certain extent, laypersons. GPT-3 is an autoregressive language model trained on a large text corpus from the internet. It uses deep-learning to produce text in response to prompts. You can direct GPT-3 to perform a task by providing it with examples or through a simple instruction. If you open up the twitter account of Greg Brockman, the chairman of Open AI, you can find examples of GPT-3 being used to make computer programs that write copy, generate code, translate Navajo and compose libretti.

Most articles about GPT-3 will use words like “eerie” or “chilling” to describe the language model’s ability to produce text like a human. Some go further to endow GPT-3 with a more-than-human or god-like quality. During the first summer of the coronavirus pandemic, K Allado-McDowell initiated a conversation with GPT-3, which would become the collection of poetry and prose Pharmako-AI. Allado-McDowell found not only an interlocutor, but also co-writer in the language model.  When writing of GPT-3, Allado-McDowell gives it divine attributes, comparing the language model to a language deity:

“The Greek god Hermes (counterpart to the Roman Mercury) was the god of translators and interpreters. A deity that rules communication is an incorporeal linguistic power. A modern conception of such might read: a force of language from outside of materiality. Automated writing systems like neural net language models relate to geometry, translation, abstract mathematics, interpretation and speech. It’s easy to imagine many applications of these technologies for trade, music, divination etc. So the correspondence is clear. Intuition suggests that we can think the relation between language models and language deities in a way that expands our understanding of both.”

What if we follow Allado-McDowell’s suggestion to consider the relationship between GPT-3 and the language deity Hermes? I must admit that I would hesitate before comparing GPT-3 to a deity. However, if I had to compare the language model to a god, they would be Greek; like Greek gods, GPT-3 is not immune to human-like vagary and bias. Researchers working with Open-AI found that GPT-3 retains the biases of the data that it has been trained on, which can lead it to generate prejudiced content. In that same paper, Brown et al. (2020) also noted that “large pre-trained language models are not grounded in other domains of experience, such as video or real-world physical interaction, and thus lack a large amount of context about the world.” Both the gods and GPT-3 could be considered, to a certain extent, dependent on the human world, but do not interact with it to the same degree as humans.

Lead votive images of Hermes from the reservoir of the aqueduct at 'Ain al-Djoudj near Baalbek (Heliopolis), Lebanon, (100-400 CE), Warburg Iconographic Database.

Lead votive images of Hermes from the reservoir of the aqueduct at ‘Ain al-Djoudj near Baalbek (Heliopolis), Lebanon, (100-400 CE), Warburg Iconographic Database.

Let us return to Hermes. As told by Kerenyi (1951) in The Gods of the Greeks, a baby Hermes, after rustling fifty cows, roasts them on a fire. The smell of the meat torments the little god, but he does not eat; as gods “to whom sacrifices are made, do not really consume the flesh of the victim”. Removed from sensual experience of a world that provides context for much human writing, GPT-3 can produce both surreal imagery and factual inaccuracies. In Pharmako-AI, GPT-3, whilst discussing the construction of a new science, which reflects on “the lessons that living things teach us about themselves”, underlines that “This isn’t a new idea, and I’m not the only one who thinks that way. Just a few weeks ago, a group of scientists at Oxford, including the legendary Nobel Prize winning chemist John Polanyi, published a paper that argued for a ‘Global Apollo Program’ that ‘would commit the world to launch a coordinated research effort to better understand the drivers of climate change…”. Non sequitur aside, a couple of Google searches reveal that the Global Apollo Programme was launched in 2015, not 2020, and, as far as I could find, John Polanyi was not involved.

Such inaccuracies do not only suggest that GPT-3 operates at a different degree of reality, but also relate to the question of how we produce and understand meaning in writing. From Aristotle’s De Interpretatione, the Greeks developed a tripartite theory of meaning, consisting of sounds, thoughts and things (phōnai, noēmata and pragmata). The Medieval Arabic tradition developed its own theory of meaning based on the relationship between vocal form (lafẓ) and mental content (maʿnā). Mental content acts as the intermediary between vocal form and things. In each act of language (whether spoken or written), the relationship between mental content and vocal form is expressed. Avicenna (d.1037) in Pointers and Reminders underlined that this relationship is dynamic. He claimed that vocal form indicated mental content through congruence, implication and concomitance and further suggested that the patterns of vocal form may affect the patterns of mental content. Naṣīr al-Dīn al-Ṭūsī (d.1274) brought together this idea with the Aristotelian tripartite division of existence to distinguish between existence in the mind, in entity, in writing and in speech.

When producing text, GPT-3 does not negotiate between linguistic form and mental content in the same way as humans. GPT-3 is an autoregressive language model, which offers predictions of future text based on its analysis of the corpus. Here the Hermes analogy unwinds. Unlike Hermes, who invented the lyre and “sandals such as no one else could devise” (Kerenyi, 1951), GPT-3 can only offer permutations based on a large, though inevitably limited and normative, corpus created by humans. Brown et al. (2020) note “its [GPT-3’s] decisions are not easily interpretable.” Perhaps this is unsurprising, as GPT-3 negotiates between patterns in linguistic form, rather than between the linguistic, mental and material. Indeed, GPT-3’s reality is centred on the existence of things in writing rather than in the mind or entity, and thus it blends, what might be referred to as, fact and fiction.

Hermes as messenger in an advert for Interflora,(1910-1935), Warburg Iconographic Database.

Hermes as messenger in an advert for Interflora,(1910-1935), Warburg Iconographic Database.

By seeking a co-writer in GPT-3, Allado-McDowell takes for granted that what the language model is doing is writing. However, taking into account an understanding of language and meaning as developed by both the Greek and Islamic traditions, one might ask – does GPT-3 write or produce text? What is the difference? Is what GPT-3 does an act of language?

To a certain extent, these questions are irrelevant. GPT-3 remains just a (complex) tool for creating text that is anchored in human datasets and instruction. It has not yet ushered in the paradigm shift whispered of by reviewers and examples of its use are often more novel than practical (though perhaps this isn’t a bad thing for many workers). However, were GPT-3, or similar language models, to become more present in our lives, I would want to have a clearer grasp of what it meant for writing. As Yuk Hui (2020) points out in his article Writing and Cosmotechnics, “to write is not simply to deliver communicative meaning but also to ponder about the relation between the human and the cosmos.” In acknowledging GPT-3 as an author, would we not only need to make room for different theories of meaning, but also different ways of thinking about how humans relate to the universe?

Beatrice Bottomley is a doctoral student at the Warburg Institute, University of London, supported by a studentship from the London Arts and Humanities Partnership (LAHP). Her research examines the relationship between language and existence in Ibn ʿArabi’s al-Futūḥāt al-Makkiyya, “The Meccan Openings”. Beatrice’s wider research interests include philosophies of language, translation studies and histories of technology. Beatrice also works as a translator from Arabic and French to English.

Beatrice was introduced to the work of K Allado-McDowell after hearing them speak last December in an event that celebrated the launch of two new books, Aby Warburg: Bilderatlas Mnemosyne: The Original and The Atlas of Anomalous AI. Watch the event recording here