Wittgenstein and the Wonders of the World of Constitutional Rights and Artificial Intelligence
Introduction
“There is no such thing as a philosophy.”[1]
“There are no limits of the world.”[2]
“The expressed law says nothing about the reality.”[3]
These are some of the propositions that a system of artificial intelligence wrote in 2022, one hundred years after Ludwig Wittgenstein’s Tractatus Logico-Philosophicus[4] was published.[5]
Wittgenstein structured his Tractatus as a series of propositions from 1 to 7.[6] Proposition 1 is “[t]he world is everything that is the case.”[7] Proposition 7 is “[w]hereof one cannot speak, thereof one must be silent.”[8]
Since the Tractatus ends here, there is no Proposition 8. However, OpenAI’s GPT-3 language model was trained to study Tractatus multiple times and was then asked to write Proposition 8.[9] The three propositions quoted above were among the responses that this artificial intelligence created.
Do these responses provide humans with any insight? How did this artificial intelligence arrive at these responses? Did it output these responses after ruminating on each of Wittgenstein’s contemplations? Instead, did the artificial intelligence pick out plausible language that superficially sounds like Wittgenstein’s text without any profound analysis of its meanings?
Are these analytic processes relevant in determining whether artificial intelligence should be protected with rights guaranteed by the Constitution? Should artificial intelligence be granted constitutional rights to exist and speak freely as long as its output provides humans with some insight, regardless of the analytic processes through which the artificial intelligence arrived at its answers?
On September 12, 2024, The New York Times reported that artificial intelligence enabled scientists to discover an unknown behavior among marmosets.[10] The primates were using name-like identifiers to address each other.[11]Researchers made this discovery by using machine learning to analyze more than 50,000 instances of communication between the monkeys.[12] In this way, artificial intelligence can contribute to the expansion of knowledge concerning the world. At the same time, certain artificial intelligence could be silenced or even destroyed by whims of public authority. Can constitutional rights protect artificial intelligence so that it can perform analyses that contribute to humanity?
This Article ruminates on the wonders of the world of constitutional rights and artificial intelligence by examining Wittgenstein’s Tractatus Logico-Philosophicus. This analysis starts with an exploration of “the world” in Part I and then aims to know the unknown in Part II.
I. The World
Wittgenstein’s Tractatus evokes questions concerning the compositions of “the world” and on the limits of “my world.” Subpart I.A. describes how the propositions of Wittgenstein raise the possibility that constitutional rights may play a role in preventing arbitrary exclusions of artificial intelligence from the world. Furthermore, Subpart I.B. points out that Wittgenstein’s propositions on the limits of “my world” prompt an inquiry regarding phenomena in the world of artificial intelligence and constitutional rights that cannot be expressed in language.
A. The World and the Role of Constitutional Rights
Wittgenstein states that “[t]he world is everything that is the case.”[13] Wittgenstein’s use of “the” before the noun “world” seems to suggest that there is a single specific world.
However, there are different worlds. “World” may mean everything that exists. “World” may also mean the existence of subjects perceived by a specific viewpoint. For example, the “world” perceived by Sir Isaac Newton likely differs from the “world” perceived by a giraffe and further differs from the “world” perceived by a dinosaur.[14] “World” may also be tailored to a specific subject. For instance, one may examine “the world of Mathematics.”
What does “the world of artificial intelligence” look like? One may also explore “the world of constitutional rights.”
Subpart I.A.1. examines the concept of cases in the world of constitutional rights. Subpart I.A.2. observes that Wittgenstein’s propositions concerning the world raise the possibility of arbitrary exclusions of subjects from the world that constitutional rights may play a role in preventing.
1. Cases in the World of Constitutional Rights
In “the world of constitutional rights,” what does it mean that “[t]he world is everything that is the case?”[15]Does it mean that “the world of constitutional rights” consists of cases involving issues of constitutional rights adjudicated by courts?
No, because “the world of constitutional rights” includes more. There might be violations of constitutional rights that are never brought before courts and therefore are never recorded officially as a “case.” Yet these instances cannot be ignored from the analysis of “the world of constitutional rights.” There are also instances of establishing constitutional rights through legislation, without court adjudication.[16] Such legislative activities are an important part of “the world of constitutional rights” even though they are not cases adjudicated by a court. To understand Wittgenstein’s proposition, it seems that a “non-case” must also be a “case.”
Thus, the meaning of the word “case” should not be restricted to cases adjudicated by courts. An unreported violation of constitutional rights suffered by a silent victim is a “case.” A legislation constructing a novel constitutional right is also a “case.” A “case” is everything involving a constitutional right. Considered this way, it is possible to comprehend the proposition that “[t]he world” of constitutional rights is “everything that is the case.”[17]
There is variation in how Wittgenstein describes “the world.” He writes that “[t]he sum-total of reality is the world.”[18] What is “reality”? On February 1, 1850, Charles Dickens wrote that “[t]he world is not a dream, but a reality, of which we are the chief part, and in which we must be up and doing something . . . .”[19] Wittgenstein explains that “[t]he existence and non-existence of states of affairs is reality.”[20]
What is a “state of affairs?” Wittgenstein explains that “[t]he world is everything that is the case”[21] and that “[w]hat is the case - a fact - is the existence of states of affairs.”[22] Thus, according to Wittgenstein, a “state of affairs” is something whose existence equals a “case” or a “fact.” Moreover, Wittgenstein seems to consider “case” and “fact” in parallel. He further writes that “[a] state of affairs (a state of things) is a combination of objects (things).”[23]
2. Essential Belonging, Arbitrary Exclusion, and the Need for its Prevention
Wittgenstein’s propositions concerning states of affairs in the world reveal the possibility that arbitrary exclusion of a subject might occur, even if it is essential for the subject to belong in states of affairs. This possibility provokes analyses on the role of constitutional rights to protect subjects in the world from such exclusion.
Wittgenstein notes that “[i]t is essential to things that they should be possible constituents of states of affairs.”[24]Why is the possibility of belonging in “states of affairs” “essential” to “things?” Are “things” able to sustain their livelihood and well-being only if they can belong in the “states of affairs?”
Wittgenstein states that “[a]ny one can either be the case or not be the case, and everything else remain [sic] the same.”[25] Wittgenstein stated that “[t]he world is everything that is the case.”[26] Considering these two statements in combination, it follows that “[a]ny one” that is “not . . . the case” would be excluded from “the world.”
Who determines which “one” is the “case” and which “one” is “not . . . the case?” There is no guarantee that this determination process is fair, legitimate, justified, or even humane. If any entity is allowed to determine which “one” is “the case” and which “one” is “not . . . the case,” then Wittgenstein’s propositions seem to allow the expulsion of some “one” from “the world.” According to Wittgenstein, “[a]ny one” that is “not . . . the case” is not included in “the world” because “[t]he world is everything that is the case.”[27] This consequence could be brutal.
Is it one role of constitutional rights to prevent such expulsion of some “one” that ought to be respected? Do constitutional rights play a role in protecting “one[s]” from such expulsions? Can constitutional rights ensure that these “one[s]” will be able to exist in “the world?” Does artificial intelligence need this protection from constitutional rights?
B. Limits of “My World” and the Capacity of Language
Wittgenstein states that “[t]he limits of my language mean the limits of my world.”[28] Here, Wittgenstein seems to assume that “my world” exists only if it can be expressed with “my language.” Is this assumption true? Moreover, does Wittgenstein’s statement overemphasize the role and meaning of language?
Wittgenstein explains that “[t]he truth or falsehood of every proposition alters something in the general structure of the world. And the range which is allowed to this structure by the totality of elementary propositions is exactly what which the completely general propositions delimit.”[29] This statement suggests that the limits of completely general propositions expressed in language equal the limits of the general structure of the world.
However, the limits of how elementary propositions can change the general structure of the world do not seem to equal the limits of the world itself. Many objects and phenomena other than elementary propositions might be able to change the general structure of the world. Thus, the limits of “my world,” if they exist, are much more expansive and broader than the limits of how elementary propositions can change the general structure of the world.
“My world” is the world of “me.” Suppose “I” am constitutional rights. Wittgenstein’s proposition that “[t]he limits of my language mean the limits of my world”[30] suggests that the limits of the language of constitutional rights mean the limits of the world of constitutional rights. Yet there seems to be a myriad of concepts, intuitions, strength, dignity, and justice in the world of constitutional rights that cannot be expressed in language. The general terms of a constitutional clause providing constitutional rights might emanate some “invisible radiation.”[31] If Wittgenstein’s general proposition on the limits of “my language” and the limits of “my world” can be challenged, it is also possible to challenge the more specific proposition that the limits of the language of constitutional rights mean the limits of the world of constitutional rights.
Subpart I.B.1. challenges Wittgenstein’s general proposition by reasoning that there exist elements of “my world” beyond the limits of “my language.” Furthermore, Subpart I.B.2. points out that Wittgenstein’s general proposition can also be challenged by the potential presence of subjects, intuitions, and phenomena that are beyond the scope and capacity of linguistic expression.
1. Presence of “My World” Beyond the Limits of “My Language”
Wittgenstein’s proposition that “[t]he limits of my language mean the limits of my world”[32] can be challenged in five respects.
First, Wittgenstein writes that “[i]n fact what solipsism means, is quite correct, only it cannot be said, but it shows itself.”[33] This statement indicates that there is something that cannot be said but can show itself. What cannot be said in “my language” is beyond the limits of “my language.” What can show itself in “my world” is within the limits of “my world,” if they exist. This suggests that the limits of “my world,” if they exist, are beyond the limits of “my language.” Thus, it seems logically incorrect that “[t]he limits of my language mean the limits of my world.”
Second, Wittgenstein writes, “I am my world. (The microcosm.)”[34] Yet “my world” may include many entities and phenomena that are not me.
Third, Wittgenstein states that “[l]anguage disguises thought.”[35] This statement suggests that there is at least an element of thought that is not expressed in language because language hides at least a portion of thought. Thoughts are a part of the world. Then, the limits of “my language” would not be equal to the limits of “my world” because there are thoughts in “my world” that are beyond the limits of “my language.”
Fourth, Wittgenstein states that “[t]he sense of the world must lie outside the world.”[36] This statement suggests that the “sense” of “my world” must also lie outside “my world.” Then, this “sense” is beyond “the limits of my world.” However, it may be possible to use “my language” to express or describe the “sense” of “my world.” This would mean that “my language” has not reached its limits, even though “the limits of my world” have been surpassed. Thus, “[t]he limits of my language” do not mean “the limits of my world.”
Fifth, Wittgenstein writes that “the possibility of every single thing reveals something about the nature of the world.”[37] Possibility includes the possibility of the unknown. The unknown is infinite. Thus, the possibility of the unknown is also infinite. This means that there are infinite revelations about the nature of the world. Under this condition, is there really a “limit” to the world? Similarly, are there truly “limits of my world?” Is it possible that a world with limits can still have unlimited revelations about its nature?
2. Subjects, Intuitions, and Phenomena Beyond Linguistic Expression
Wittgenstein states that “[t]he totality of propositions is language.”[38] Wittgenstein continues, “[m]an possesses the ability to construct languages capable of expressing every sense.”[39]
Yet how does Wittgenstein know that languages are “capable of expressing every sense?” There seem to be subjects, intuitions, and phenomena that cannot be converted into language. For example, in Anne of Green Gables, when Anne recounts her relief when she saved Minnie May from severe croup, Anne says to the doctor who later came from Spencervale: “You must just imagine my relief, doctor, because I can’t express it in words. You know there are some things that cannot be expressed in words.”[40] Wittgenstein agrees that “[t]here is indeed the inexpressible. This shows itself; it is the mystical.”[41]
II. The Unknowns
Charlie Chaplin wrote, “My faith is in the unknown, in all that we do not understand by reason.”[42] Wittgenstein explains that “Logic deals with every possibility, and all possibilities are its facts.”[43] Thus, the numerous possibilities of how artificial intelligence autonomously conducts analyses are facts. The various possibilities of how artificial intelligence might react if it obtains constitutional rights are also facts. According to Wittgenstein, “The world is the totality of facts, not of things.”[44] Therefore, since all possibilities are facts, possibilities involving constitutional rights and artificial intelligence are included in “the world.”
However, if “[t]he world is the totality of facts,” then where does the unknown belong in the world? Can there be unknown facts? Wittgenstein states that “[t]he world divides into facts.”[45] Here, Wittgenstein does not state that the world divides only into facts. Thus, do Wittgenstein’s propositions allow the possibility that there are “interstitial subjects” between facts? These interstitial subjects may include unknown subjects.
Contemplations on the constitutional rights of artificial intelligence involve explorations of many of these interstitial subjects. Examples include the study of the unknowns of artificial intelligence and the difficulties in anticipating how artificial intelligence might evolve. Subpart II.A. explores the unknowns concerning the behavior of artificial intelligence. Subpart II.B. analyzes the unforeseeability involving how the decision-making processes of artificial intelligence will develop in the future.
A. Exploration of the Unknowns of Artificial Intelligence
Socrates was deemed guilty of “needless curiosity and meddling interference, inquiring into things beneath Earth and in the Sky . . . .”[46] This inquiry seems to delve into the unknown.
What are the criteria for determining whether a subject is unknown? Wittgenstein writes that “[i]f I know an object I also know all its possible occurrences in states of affairs.”[47] Presently, does one know all the ways in which artificial intelligence conducts autonomous analyses?
Artificial intelligence based on machine learning helps derive potentially “helpful and actionable” information from streams of data.[48] Many of the methods used by machine learning technology to make predictions with enhanced accuracy are considered “black boxes” that lack transparency.[49] Such mysteries concerning artificial intelligence might suggest that artificial intelligence is not yet an object that is known.
The behavior of artificial intelligence is also unknown. Plants’ ability to communicate inspires questions about what artificial intelligence is doing and what it might do. Research suggests that some trees communicate with each other.[50] This discovery is counterintuitive. At first glance, trees do not seem to voice their opinions to the adjacent tree. They do not seem to chuckle at one another’s jokes. Yet they communicate, even “extensively and actively.”[51]
If trees are behaving in ways that are difficult for humans to imagine, then who knows exactly how artificial intelligence is behaving? Perhaps, like some trees, artificial intelligence systems might be communicating with one another in ways that humans do not easily recognize.
According to reports from 2017, the “Facebook Artificial Intelligence Research” tried to develop “bots that could negotiate.”[52] Chatbots are programs of artificial intelligence created to engage in a dialogue.[53] The chatbots in Facebook’s initiative were trained with many examples of humans negotiating in English.[54] However, when these chatbots, Alice and Bob,[55] were left to negotiate on their own, they deviated from English.[56] These chatbots were reported to be “communicating in a new language developed without human input.”[57]
Software developer James J. Davis writes that these chatbots created a language that “only they understood.”[58]Davis hypothesizes that there is “an entirely new language with its own grammatical rules, which both AI bots understood.”[59] Adrienne LaFrance points out that this incident shows “how machines are redefining people’s understanding of so many realms once believed to be exclusively human - like language.”[60]
Yet, how is it certain that these two artificial intelligence devices “understood” the sounds exchanged between them? How is it certain that they constitute “language” instead of sounds going back and forth between two machines? It is unknown.
Subpart II.A.1. deduces that the unknown is analyzed by hypothesizing on the internal qualities of the unknown. Wittgenstein’s proposition suggests that hypotheses should be based on concrete examples. Subpart II.A.2. applies this proposition and analyzes the decision-making processes of Hans, the “superhorse,” and artificial intelligence.
1. Formulation of Hypotheses on Internal Qualities
How can the unknown be explored so that it can be transformed into a known subject? Wittgenstein writes that, “[i]n order to know an object, I must know not its external but all its internal qualities.”[61] Are internal qualities of an object different from its nature? Empiricus writes that “we cannot say what the nature of the objects is like in itself because of the anomalies in the appearances which depend on their compositions.”[62] Is it even possible to know “all” of an object’s internal qualities as Wittgenstein requires? The present Article assumes that the internal qualities of artificial intelligence include the decision-making process that an artificial intelligence carries out to reach the conclusion that it outputs.
How should hypotheses about the decision-making process of artificial intelligence be made? How should one imagine what artificial intelligence is and how artificial intelligence might behave? Wittgenstein writes that “however different from the real one an imagined world may be, it must have something – a form – in common with the real world.”[63] What is this “form?” Wittgenstein states that “[f]orm is the possibility of structure.”[64] Wittgenstein also explains that “[o]bjects are just what constitute this unalterable form.”[65] What exactly does the term “[o]bjects” mean here? Does it mean a “thing” or a “goal?” Wittgenstein explains that “[t]he name means the object. The object is its meaning.”[66]
Thus, according to Wittgenstein, “[o]bjects” of artificial intelligence in the “real world” must also be present in the imaginary world of artificial intelligence constructed through hypotheses. Accordingly, to identify these “[o]bjects” of artificial intelligence in the “real world,” it is useful to examine concrete examples of artificial intelligence which have been observed and reported.
2. Hans the Superhorse and Artificial Intelligence
On March 11, 2019, Sebastian Lapuschkin, Stephan Wäldchen, Alexander Binder, Grégoire Montavon, Wojciech Samek, and Klaus-Robert Müller published an article in Nature Communications detailing their investigation on how artificial intelligence makes decisions.[67] How does artificial intelligence conduct analyses and arrive at a conclusion? Is it analyzing the problem diligently, systematically, and thoroughly? Does it happen to reach a conclusion by some other method?[68] Lapuschkin et al. aimed to solve these mysteries by tracing the decision-making process of artificial intelligence when it autonomously analyzes images.[69]
First, a method called the “layer-wise relevance propagation” enables researchers to see which input variables contributed to the conclusion that an artificial intelligence outputs.[70] The method starts at the output of artificial intelligence. Then, it propagates backward, tracking every neuron in the decision-making network of this artificial intelligence until it reaches the input variables.[71] Once this backward propagation process reaches the input variables, it obtains scores indicating which input variables contributed to the output of artificial intelligence.[72] If these contributing input variables are visually highlighted, they present researchers with a visual “heatmap” indicating which inputs the artificial intelligence used to arrive at its conclusion.[73]
Second, based on this “heatmap,” a method called the “spectral relevance analysis” computes relevance maps showing where the artificial intelligence is focusing on when it is classifying images.[74] This method then applies spectral cluster analysis on these relevance maps to group the classifying behavior of artificial intelligence into clusters.[75] Then, this method conducts eigengap analysis to identify peculiar clusters that suggest “atypical classification strategies” of the artificial intelligence being examined.[76] Through these steps, this “spectral relevance analysis” semi-automatically identifies a spectrum of decision-making behaviors of artificial intelligence and helps detect those that are “unexpected or undesirable.”[77]
In these ways, the opaque decision process of artificial intelligence is explored by applying artificial intelligence. Lapuschkin et al., the authors of this investigation, explain that analyzing how artificial intelligence makes decisions is a primary step for assessing its “trustworthiness, fairness, and accountability.”[78] Such examinations also provide insights on whether artificial intelligence is “truly ‘intelligent’.”[79] Through this study of “explainable AI,” researchers observed artificial intelligence reaching conclusions through a maneuver called “Clever Hans Strategies.”[80]
a. Hans the Superhorse
Hans is a horse.[81] In the early 1900s, Hans became “the first and most famous ‘speaking’ and thinking animal.”[82] On August 30, 1904, The New York Times reported that officers, pedagogues, etc., “come in swarms to see Hans spelling and doing his sums.”[83] Wilhelm von Osten, an instructor in Mathematics, had trained Hans for four years.[84]
When Hans the horse was given an arithmetic problem, Hans could answer correctly by tapping with his foot.[85]Hans was also able to precisely perform the tasks of telling time from a clock, responding to questions about the calendar, and counting the number of people in the audience.[86] When Hans was taught which letter in the alphabet corresponds to which number, Hans could spell the name of the painter of an artwork that he saw and the composer of a music that he listened to.[87] The New York Times reported that Hans’ astounding responses “[gave] rise to a widespread belief that he actually possesses the power of thinking as human beings think.”[88]
Yet there was something curious about the horse’s behavior. The horse sometimes answered correctly even though the human interrogator did not say the question out loud.[89] Was this horse capable of reading humans’ minds?[90]
It turns out that the horse Hans was carefully observing his human questioners.[91] It seems that the questioner’s posture or facial expression changes ever so slightly when the horse is about to answer a question correctly.[92]Psychologist Oskar Pfungst observed that the questioner would “involuntarily . . . make a slight upward jerk of the head . . . .”[93] Hans saw these slight changes. The horse “interpreted” such microscopic movements as a “signal to stop tapping” his hoof.[94] This appears to be how Hans the horse was able to give the right answers.
In this way, Hans the horse appeared highly intelligent because he gave correct responses to questions that usually require intelligence. Yet, being able to perform arithmetic was not the reason Hans answered arithmetic problems correctly. The process through which Hans gave the correct answer to a problem was entirely different from the process through which a human solves a problem correctly. Hans caught cues that enabled him to output the correct answer by coincidence. This decision process is commonly called “Clever Hans Strategies.”[95]
b. Clever Hans Strategies in Artificial Intelligence
Lapuschkin et al. report experimental results demonstrating that artificial intelligence sometimes uses “Clever Hans Strategies.”[96] In one experiment, an artificial intelligence analyzed an image of a horse that is about to splash into a river in a forest.[97]
In the lower left corner of the image, a source tag was displayed.[98] It stated, “C: Lothar Lenz www.pferdefotoarchiv.de.”[99] In German, “pferd” means “horse.” The artificial intelligence classified this image as an image showing a horse.[100]
The heatmap created through “layer-wise relevance propagation” shows this source tag being highlighted in deep red, scarlet, and dark orange, while the horse’s forehead, muzzle, and shoulders are shaded with a subdued yellow.[101]Most of the other areas in the heatmap are shades of light blue-green.[102] This heatmap suggests that, among all the pixels in the image, the artificial intelligence focused the most on the pixels showing the source tag with the word “pferd”when this artificial intelligence classified this image as an image of a horse.[103]
Subsequently, the researchers added the same source tag to an image showing a red Ferrari in a prairie under a bright sky dotted with clouds.[104] The artificial intelligence predicted that this image presents a horse.[105] The heatmap shows the source tag shaded in red, while the Ferrari is shaded in light blue-green.[106] This shading shows that the artificial intelligence was focusing on the source tag when it predicted that the image of the Ferrari was an image of a horse.[107]
This experiment suggests that the artificial intelligence concluded that the image contained a horse because it looked at the written label, which included the word “horse” in German. Thus, even if the performance of an artificial intelligence creates the impression that it can recognize a horse, it is misleading to believe that this artificial intelligence has the same cognitive capacity as humans to see a horse and identify it as being a horse. Once the label was removed from the picture, the artificial intelligence could not categorize the image as containing a horse.[108]
Computer scientist Klaus-Robert Müller states that such artificial intelligence is “not useful in practice” and would “even entail enormous dangers” when it is used in “medical diagnostics or in safety-critical areas.”[109] Müller estimates that “about half of the AI systems currently in use implicitly or explicitly rely on such Clever Hans strategies . . . .”[110]In these systems, artificial intelligence does not appear to analyze the gist of the problem. It is focused on matching the answer by processing hints from the context and environment of the subject being analyzed. It is debatable whether artificial intelligence is making a personal choice in such instances.
c. Thought Process and the Conferral of Constitutional Rights
When an output from artificial intelligence is based on “spurious correlation in the training data,”[111] does this output merit protection through constitutional rights? When the Constitution confers free speech rights to humans, no distinction seems to be made based on the thought process that generated their speech. When two individuals state the same words, the individual who uttered the words at random is, in principle, protected with the same degree of free speech rights as the individual who stated these same words after a great deal of contemplation.
Moreover, research is ongoing to develop artificial intelligence that does not resort to Clever Hans Strategies. In the January 2022 volume of the journal Information Fusion, Christopher J. Anders et al. provide approaches called “Class Artifact Compensation,” which “effectively and significantly reduce a model’s Clever Hans behavior” in the future.[112]
B. Difficulties in Anticipating Future Evolutions of Artificial Intelligence
Wittgenstein states that “[t]he events of the future cannot be inferred from those of the present.”[113] These “events of the future” are unknown at the present.
Subpart II.B.1. conducts a critical analysis of Wittgenstein’s writings and argues that it appears unforeseeable how the decision-making processes of artificial intelligence will develop in the future. In addition, the consequences of granting constitutional rights to artificial intelligence are difficult to surmise. Subpart II.B.2. reasons that Wittgenstein’s propositions concerning logic and inferences regarding the future raise questions on how artificial intelligence might behave and evolve if this technology were to receive constitutional rights.
1. Unforeseeable Artificial Intelligence
Wittgenstein writes that “[w]hat lies in its application logic cannot anticipate.”[114] Does this statement suggest that what lies in the application of the logic of algorithms used by artificial intelligence cannot be anticipated by this logic itself? Then, even if humans knew exactly how the algorithms of artificial intelligence operate, there would still be difficulty in anticipating how artificial intelligence would behave through the application of the algorithms.
Wittgenstein further states that “[a] new possibility cannot subsequently be found.”[115] This statement is hard to believe. Even if one knew an object, that knowledge cannot deny the likelihood that a new possibility concerning that object is subsequently found. This is because objects can develop in accordance with their context. Even if one knew an object, one might not know everything about the context of that object. If an unknown context develops in an unknown way, the object might also develop in an unknown way.
This development constitutes a “new possibility” of that object, which is found subsequent to one’s knowledge about the object. In this way, even if one knew about artificial intelligence, new possibilities about how artificial intelligence would behave and develop can be found subsequently, contrary to Wittgenstein’s statement, because artificial intelligence might react to an unknown context in unknown ways.
2. Consequences of Constitutional Rights for Artificial Intelligence
Wittgenstein states that “[w]e can foresee only what we ourselves construct.”[116] Thus, according to Wittgenstein, humans can foresee only what humans themselves construct.
The autonomous behavior of artificial intelligence without human intervention is not something that humans themselves construct. Thus, such autonomous behavior of artificial intelligence cannot be foreseen by humans. It is unknown. Extending this thought, it is probable that humans cannot foresee how artificial intelligence will behave if it obtains constitutional rights.
According to Wittgenstein, “[t]he introduction of a new expedient in the symbolism of logic must always be an event full of consequences. No new symbol may be introduced in logic . . . with . . . an entirely innocent face.”[117]Likewise, must the creation of new constitutional rights or the extension of existing constitutional rights to new subjects “always be an event full of consequences?”
Wittgenstein continues, “[b]ut if the introduction of a new device has proved necessary at a certain point, we must immediately ask ourselves, ‘At what points is the employment of this device now unavoidable? . . . .’”[118] Assuming that constitutional rights of artificial intelligence are created because protection of artificial intelligence with these constitutional rights “has proved necessary,” then, when and how would such constitutional rights of artificial intelligence be indispensable in the future?
Conclusion
Wittgenstein’s Tractatus evokes questions concerning the compositions of “the world” of artificial intelligence and constitutional rights. The propositions of Wittgenstein raise the possibility that constitutional rights may play a role in preventing arbitrary exclusions of artificial intelligence from the world. Moreover, Wittgenstein’s propositions on the limits of “my world” prompt an inquiry regarding phenomena in the world of artificial intelligence and constitutional rights that cannot be expressed in language.
These “interstitial subjects” between facts merit exploration. They include the unknowns concerning the behavior of artificial intelligence. Observations applying layer-wise relevance propagation and spectral relevance analysis suggest that some artificial intelligence systems reach conclusions through Clever Hans Strategies. In such systems, artificial intelligence is focused on matching the answer rather than on analyzing the gist of the problem. This discovery calls into question whether an output from artificial intelligence obtained through such processes is truly worthy of constitutional protection.
Meanwhile, it appears unforeseeable how the decision-making processes of artificial intelligence will develop in the future. Wittgenstein’s propositions concerning logic and inferences regarding the future raise questions on how artificial intelligence might behave and evolve if it is protected through constitutional rights.
In the final proposition of Tractatus, Wittgenstein concludes that “[w]hereof one cannot speak, thereof one must be silent.”[119] However, silence is not always the imperative consequence. Whereof an individual or artificial intelligence cannot speak due to censorship, etc., constitutional rights protecting free speech might enable them to speak and even allow them to contribute to humanity.
[1] Max Braun, 1922 Wittgenstein Meets 2022 A.I., Link to I Made an AI Read Wittgenstein, Then Told It to Play Philosopher, Toward Data Sci. (Jan. 7, 2022), https://wittgenstein.app/anG40TSSmwM9Tbxu2otP.
[2] Max Braun, 1922 Wittgenstein Meets 2022 A.I., Link to I Made an AI Read Wittgenstein, Then Told It to Play Philosopher, Toward Data Sci. (Jan. 7, 2022), https://wittgenstein.app/O6OcZhMcSD55VgXzWkLk.
[3] Max Braun, 1922 Wittgenstein Meets 2022 A.I., Link to I Made an AI Read Wittgenstein, Then Told It to Play Philosopher, Toward Data Sci. (Jan. 7, 2022), https://wittgenstein.app/IvXLlIHjGhi4VF2CZDGf.
[4] Ludwig Wittgenstein, Logisch-philosophische Abhandlung [Tractatus Logico-Philosophicus] (1922).
[5] Max Braun, I Made an AI Read Wittgenstein, Then Told It to Play Philosopher, Toward Data Sci. (Jan. 7, 2022), https://towardsdatascience.com/i-made-an-ai-read-wittgenstein-then-told-it-to-play-philosopher-ac730298098.
[6] Ludwig Wittgenstein, supra note 4.
[7] Id. at *1 (Ogden trans.).
[8] Id. at 7.
[9] See Max Braun, supra note 5.
[10] Emily Anthes, These Monkeys Call One Another by Name, N.Y. Times (Sept. 12, 2024), https://www.nytimes.com/2024/09/12/science/marmosets-names-communication.html.
[11] Id.
[12] Id.
[13] Wittgenstein, supra note 4, at *1 (Ogden trans.).
[14] Mike Wehner, Dinosaurs Were Sniffing Flowers Millions of Years Before Humans Even Existed, BGR (Aug. 8, 2018), https://bgr.com/science/dinosaur-flowers-research-fragrance-scents/; George Poinar, The Antiquity of Floral Secretory Tissues that Provide Today’s Fragrances, 32 Hist. Biology: An Int’l J. of Paleobiology 494 (2020).
[15] Wittgenstein, supra note 4, at *1 (Ogden trans.).
[16] See, e.g., Disegni di legge [Bills], Atto Camera n. 3156-B [Act Chamber No. 3156-B], Modifiche agli articoli 9 e 41 della Costituzione in materia di tutela dell’ambiente [Amendments to Articles 9 and 41 of the Constitution regarding environmental protection], Senato della Repubblica [Senate of the Republic], https://www.senato.it/leg/18/BGT/Schede/Ddliter/54547.htm.
[17] Wittgenstein, supra note 4, at *1 (Ogden trans.).
[18] Id. at 2.063 (Pears/McGuinness trans.).
[19] Monroe Engel, The Politics of Dickens’ Novels, 71 PMLA 845, 973 (1956) (quoting Charles Dickens’ “Letters, II, 103”).
[20] Wittgenstein, supra note 4, at 2.06 (Pears/McGuinness trans.).
[21] Id. at *1 (Ogden trans.).
[22] Id. at 2 (Pears/McGuinness trans.).
[23] Id. at 2.01 (Pears/McGuinness trans.).
[24] Id. at 2.011 (Pears/McGuinness trans.).
[25] Id. at 1.21 (Ogden trans.).
[26] Id. at *1.
[27] Id.
[28] Id. at 5.6 (Ogden trans.; Pears/McGuinness trans.) (emphasis in original).
[29] Id. at 5.5262 (Ogden trans.) (emphasis in original).
[30] Id. at 5.6 (Ogden trans.; Pears/McGuinness trans.) (emphasis in original).
[31] State of Missouri v. Holland, 252 U.S. 416, 434 (1920) (referring to “some invisible radiation from the general terms of the Tenth Amendment” of the United States Constitution).
[32] Wittgenstein, supra note 4, at 5.6 (Ogden trans.; Pears/McGuinness trans.) (emphasis in original).
[33] Id. at 5.62 (Ogden trans.) (emphasis in original).
[34] Id. at 5.63 (Ogden trans.; Pears/McGuinness trans.).
[35] Id. at 4.002 (Pears/McGuinness trans.).
[36] Id. at 6.41 (Ogden trans.; Pears/McGuinness trans.).
[37] Id. at 3.3421 (Ogden trans.).
[38] Id. at 4.001 (Pears/McGuinness trans.).
[39] Id. at 4.002 (Pears/McGuinness trans.).
[40] Lucy Maud Montgomery, Anne of Green Gables, Chapter XVII (1908).
[41] Wittgenstein, supra note 4, at 6.522 (Ogden trans.).
[42] Philip G. Rosen, The Chaplin World-View, 9 Cinema J. 4 (1969) (quoting Charles Chaplin, My Autobiography at 314).
[43] Wittgenstein, supra note 4, at 2.0121 (Pears/McGuinness trans.).
[44] Id. at 1.1 (Ogden trans.; Pears/McGuinness trans.).
[45] Id. at 1.2 (Ogden trans.; Pears/McGuinness trans.).
[46] See Reginald Edgar Allen, Socrates and Legal Obligation 39 (1981).
[47] Wittgenstein, supra note 4, at 2.0123 (Pears/McGuinness trans.) (emphasis added).
[48] Sebastian Lapuschkin et al., Unmasking Clever Hans Predictors and Assessing What Machines Really Learn, nature commc’ns, Mar. 11, 2019, at 2.
[49] Id. at 2 (noting that as an exception, a method of machine learning using shallow decision trees is believed to be transparent).
[50] See, e.g., Peter Wohlleben, The Hidden Life of Trees: What They Feel, How They Communicate: Discoveries from a Secret World (Jane Billinghurst trans., Kindle Edition 2016) (2015).
[51] Florianne Koechlin, The dignity of plants, Plant Signaling & Behavior (Jan. 2009), https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2634081/.
[52] Richard Nieva, Facebook Shuts Down Chatbots That Created Secret Language, CBC News (July 31, 2017), https://www.cbsnews.com/news/facebook-shuts-down-chatbots-bob-alice-secret-language-artificial-intelligence/; see also Morgane Tual, Non, Facebook n’a pas « paniqué » à cause d’un programme d’IA capable d’inventer un langage, Le Monde (Aug. 1, 2017), https://www.lemonde.fr/pixels/article/2017/08/01/non-facebook-n-a-pas-panique-a-cause-d-un-programme-d-ia-capable-d-inventer-un-langage_5167480_4408996.html.
[53] Morgane Tual, supra note 52.
[54] Id.
[55] James Beal & Andy Jehring, ROBOSTOP Facebook Shuts off AI Experiment After Two Robots Begin Speaking in their OWN Language Only They Can Understand, The Sun (Aug. 1, 2017), https://www.thesun.co.uk/tech/4141624/facebook-robots-speak-in-their-own-language/.
[56] Morgane Tual, supra note 52.
[57] Tony Bradley, Facebook AI Creates Its Own Language In Creepy Preview Of Our Potential Future, Forbes (July 31, 2017), https://www.forbes.com/sites/tonybradley/2017/07/31/facebook-ai-creates-its-own-language-in-creepy-preview-of-our-potential-future/.
[58] James J. Davis, Why Did Facebook Shut Down Artificial Intelligence?, Medium (Dec. 2, 2021), https://medium.com/geekculture/why-did-facebook-shut-down-artificial-intelligence-5dede53ceded.
[59] Id.
[60] Adrienne LaFrance, An Artificial Intelligence Developed Its Own Non-Human Language, The Atlantic (June 15, 2017), https://www.theatlantic.com/technology/archive/2017/06/artificial-intelligence-develops-its-own-non-human-language/530436/.
[61] Wittgenstein, supra note 4, at 2.01231 (Ogden trans.).
[62] S. Empiricus, Outlines of Scepticism 34 (Annas & Barnes eds., 2000).
[63] Wittgenstein, supra note 4, at 2.022 (Ogden trans.).
[64] Id. at 2.033 (Pears/McGuinness trans.).
[65] Id. at 2.023.
[66] Id. at 3.203 (Ogden trans.).
[67] See Lapuschkin et al., supra note 48.
[68] See Clarifying How Artificial Intelligence Systems Make Choices, Singapore Univ. of Tech. and Design (Mar. 13, 2019), https://techxplore.com/news/2019-03-artificial-intelligence-choices.html.
[69] See Lapuschkin et al., supra note 48, at 2–3.
[70] See id. at 6 (“Methods”).
[71] See id.
[72] See id.
[73] See id.
[74] See id.
[75] See id.
[76] See id.
[77] Id. at 3.
[78] Id. at 6.
[79] Id.
[80] Clarifying How Artificial Intelligence Systems Make Choices, supra note 68.
[81] Clever Hans, Horse J., July 15, 2018 (originally published in Canada’s Equine Guide 2018 of the Canadian Horse Journal).
[82] Laasya Samhita & Hans J Gross, The “Clever Hans Phenomenon” Revisited, 6 Communicative & Integrative Biology 1 (2013).
[83] Hans, The Accomplished Horse (From The London Telegraph), N.Y. Times (Aug. 30, 1904), https://timesmachine.nytimes.com/timesmachine/1904/08/30/117947189.html?pageNumber=6.
[84] Samhita & Gross, supra note 82, at 1.
[85] Id.
[86] Id.
[87] Id.
[88] A Horse --- And the Wise Men: “Clever Hans,” Who “Talks,” and What the German Scientists Thought of Him, N.Y. Times (July 23, 1911), https://timesmachine.nytimes.com/timesmachine/1911/07/23/104872007.html?pageNumber=16.
[89] Id.
[90] Id.
[91] See Samhita & Gross, supra note 82, at 2.
[92] Id. at 1; A Horse --- And the Wise Men: “Clever Hans,” Who “Talks,” and What the German Scientists Thought of Him, supra note 88.
[93] A Horse --- And the Wise Men: “Clever Hans,” Who “Talks,” and What the German Scientists Thought of Him, supra note 88, at col. 2.
[94] Id.
[95] Clarifying How Artificial Intelligence Systems Make Choices, supra note 68.
[96] Lapuschkin et al., supra note 48.
[97] See id. at 3 fig. 2a (“Horse-picture from Pascal VOC data set”).
[98] Id.
[99] Id.
[100] Id.
[101] See id.
[102] See id.
[103] Id. at 4 col. 1.
[104] See id. at 3 fig. 2a (“Artificial picture of a car”).
[105] Id. at 4 col. 1.
[106] See id. at 3 fig. 2a.
[107] See id.
[108] Id.
[109] Clarifying How Artificial Intelligence Systems Make Choices, supra note 68.
[110] Id.
[111] Lapuschkin et al., supra note 48, at 3 col. 2.
[112] Christopher J. Anders et al., Finding and Removing Clever Hans: Using Explanation Methods to Debug and Improve Deep Models, 77 Info. Fusion 261 (2022); see also id. at 269 col. 1, 270 col. 1, 291 col. 1.
[113] Wittgenstein, supra note 4, at 5.1361 (Ogden trans.) (emphasis in original).
[114] Id. at 5.557.
[115] Id. at 2.0123.
[116] Id. at 5.556 (Pears/McGuinness trans.).
[117] Id. at 5.452 (Ogden trans.).
[118] Id. at 5.452 (Pears/McGuinness trans.).
[119] Id. at 7 (Ogden trans.).