Trust without Teachers
In her first post-election public appearance, Hillary Clinton decried an "epidemic of fake news." Salacious stories and fraudulent claims about politicians and their supporters had spread unfiltered and unconstrained through social media. With some concocted content from Macedonian teenagers and young American college graduates, Facebook, suggested some, threw the election to Trump. Mark Zuckerberg, Facebook's chairman and co-founder, denied that his company had any responsibility. "More than 99 percent of what people see" on Facebook, he said shortly after the election, "is authentic." It was a "pretty crazy idea" to suggest that Facebook could affect an election. Trust us, counselled Zuckerberg, we only give you facts and friends.
Zuckerberg's refusal to acknowledge Facebook's possible role in the US election is both disingenuous—Facebook has conducted experiments on the effects particular kinds of posts have on people's voting decision—and irresponsible. The social media behemoth is now the primary news medium in the United States. Zuckerberg casts his company as a neutral medium that simply connects friends, shares information, and facilitates democracy. But Facebook is now a social institution that people rely on and, however implicitly, trust.
Implicit in the entire project is a basic dissatisfaction with the current digital environment that Google helped create.
Compare Zuckerberg's initial response to Google's recent attempts to reinvent its search engine as an arbiter of facts and trustworthiness. Acknowledging that most search engines evaluate web sources based on their popularity, a team of Google engineers described their attempts to evaluate the "trustworthiness" of 119 million web pages in an article titled "Knowledge-Based Trust: Estimating the Trustworthiness of Web Sources." Google, so it seems, wants to automate trust.
Google's method for extracting facts from the web, evaluating them, and then determining a score for individual web pages represents a significant shift from its earlier assumptions about how information is organized and transmitted in our digital age, and I will return to these important details later in this essay. But what I find most significant about Google's "Knowledge-Based Trust" project is Google's interest in trust in the first place.
The authors provide technical details for algorithms and machine-learning processes, but implicit in the entire project is a basic dissatisfaction with the current digital environment that Google helped create. And now Google wants to reform that media environment by redefining what it means to trust and what counts as authoritative knowledge in our digital age. In little more than a decade since its founding, Google is moving from helping us access the web pages we want to determining what web pages we should trust.
But what kind of custodian of knowledge and trust is Google? For centuries, universities and academies have served this cultural function. They were bulwarks against falsehood and institutions for truth. They did not always live up to the epistemic and ethical ideals they propounded, but one of their primary tasks was to make facts and beliefs correspond. In important ways, then, universities are the cultural forebears of Facebook and Google.
What was the nature of the trust granted these institutions? If Google's engineers hope to ground trustworthiness in statistical computation, in what did universities and other institutions of higher learning ground their own trustworthiness? Why did communities, nations, and cultures turn to these institutions for knowledge they could trust?
One of the primary responsibilities of a teacher was to decide which texts to read and which texts to ignore.
There were, of course, many features to this university-based trust—scholarly method, libraries, curricula, degrees—but one of the most consistent, from thirteenth-century Paris to nineteenth-century Berlin to twentieth-century Baltimore (at Johns Hopkins) was, simply put, the teacher. Knowledge was tied to the character of particular people, teachers who cultivated traditions and cared for their students.
The Renaissance humanism of the fourteenth-century Italian Leonardo Bruni provided one model for creating and sharing trustworthy knowledge. Born in Arezzo, Italy, the birthplace of Petrarch, in 1370, Bruni served as apostolic secretary for four popes and as chancellor of Florence. Like his fellow civic humanists of the Quattrocento, he considered liberal learning, what he termed the studia humanitatis, to be an education oriented toward the unity of an active political life with a philosophical one. An early version of the modern humanities, the studia humanitatis, he wrote, related "life and behavior" for the purpose of forming a liberal citizen, the person midway between the "avaricious and prodigal."
For Bruni, a humanistic education required a pedagogical relationship built on trust. A student had to trust that such an education would transform him before even embarking on it, and he had to trust that his teacher was an able and loving guide to such a process.
Writing to Lady Battista Malatesta of Montefeltro in 1424, Bruni described education as the development and perfection of "innate powers." When done rightly, education cultivates students' latent capacities. It builds them up, he continues, by elevating them to excellence. One of the underlying presuppositions of humanist pedagogy was that every student needed an instructor to "train" and "initiate" her into the practices and knowledge that developed all those who came before her. The student had to grant her teacher a certain authority over her, the student's, own development. And so, she had to trust her teacher not only to guide her well but, most importantly, to care for her along the way.
For Bruni, such an education was best achieved by studying only the "best and most approved authors" and avoiding "the crude and ignorant" writings of those who only "ruin and degrade our natural abilities." And, as was the case for his humanist contemporaries, the "best" authors were the ancients, such as Aristotle, Plato, and Cicero. Bruni, a translator and scholar, collected, curated, and cared for these written records of the ancients, because he believed that by reading ancient literature, students such as Lady Montefeltro could be changed. Literature, he wrote to her, was "the pablum of the mind by which the intellect is trained and nourished." It provides the reader with the material on which students exercise their minds and with examples of lives well lived. And not just any literature would do, as the advice to read only the "best" literature suggested. A properly humanist education required distinctions. Consider yourself a "gastronomer" and carefully choose what you consume, he counselled Montefeltro, in order to preserve the integrity of your taste.
Humanistic education recognized the humanity of the student--both her finitude and potential for development.
Bruni suggested, in other words, that both reading and eating are acts of consumption. When a student reads Plato, she ought not just scan words on a page but also make them her own. She ought to ingest them so that they, like water and food, can sustain her. For Bruni, reading was not simply information transfer—the relocation of full and complete ideas from one space to another to another, one manuscript to a mind. Reading was a transformative experience in which texts act on readers in often-inscrutable ways. The decisions about which books to imbibe, then, was crucial for humanists such as Bruni. And, like the process of education more broadly, choosing which books to read was not to be undertaken alone.
Bruni's fellow humanist to the north Desiderius Erasmus (1466–1536) also counselled teachers and students to read only "the best." The quality of a humanistic education was directly related to the quality of the texts around which it was organized, and so, he concluded, one of the primary responsibilities of a teacher was to decide which texts to read and which texts to ignore.
In On the Method of Study, his educational program published in 1511, Erasmus acknowledged the burden such responsibilities placed on teachers and the trust it required students to grant them. Writing after the invention of the printing press and amid the first cultural anxieties about the proliferation of print, he worried that the humanist mandate to read only the "best" implied an omniscience that exceeded human capacities, especially those of young students. A student, he wrote, will understandably look to his teacher and expect the "proverbial encyclopedia"—an ability masterfully to traverse the ever-expanding field of knowledge and to offer clear and concise guidance. A good teacher, therefore, must "range though the entire spectrum of writers so that he reads, in particular, all the best, but does not fail to sample any author, no matter how pedestrian." The humanist teacher filtered the "range" of literature in order to present to his students only those texts worth reading. And the students had to trust the results of the teacher's search.
Humanists such as Bruni and Erasmus understood teachers as guides on a path of ethical formation, one lined with exemplary texts and transformative moments of reading. They assumed that no student was, initially at least, capable of making their way alone. Humanistic education recognized the humanity of the student—both her finitude and potential for development. The time "at our disposal," Bruni warned Lady Montefeltro, is limited, so read and devote yourself only to "those things that are most important and most useful."
Humanist pedagogy, as articulated by Bruni and Erasmus, demanded another element of trust. They both described education as a particular practice with a distinct end: moral development. Although Bruni could describe moral formation in precise and generally Aristotelian terms, the students whom he counselled certainly could not, at least at the beginning of their education. For them, the ends of education were inarticulate. They had to begin their education with only a vague sense of who they might become at the end or why they were undertaking the process at all. These goods were often difficult to articulate, deferred, and even not recognized as goods at all.
Unlike their humanist predecessors, modern university professors crafted books and experiments, not souls.
Almost four centuries after Bruni wrote to Lady Montefeltro and after the near collapse of German universities around 1800, the German philosopher Johann Gottlieb Fichte—along with several of his contemporaries such as Friedrich Schleiermacher, Friedrich Schelling, and Wilhelm von Humboldt—dreamed of an institution that could sustain and cultivate the kind of pedagogical practices and trust that Bruni and Erasmus described.
Many of the early modern humanists had positions in universities, but their more lasting legacy was the Republic of Letters, the loose epistolary network that bound scholars across Europe and beyond through the circulation of letters. Fichte and his contemporaries wanted to recover the humanist ethos but translate it into institutional form, replete with state funding, committees, and clearly articulated norms. They wanted to reinvent the university as a community of scholars and students devoted to truth and marked by trust.
In his proposal for a new university in Berlin, written in 1809 and shared with von Humboldt, who was overseeing Prussian plans for such an institution, Fichte wrote that true learning and scholarship emerged out of a "love for the art of reason," a love most immediately visible in the charismatic authority of the university professor as he lectured to a rapt group of students.
For Fichte, a university lecture was a performance of the act of thinking itself, not simply a transfer of information from professor to student. (The German term for lecture, Vorlesung, literally means "to read before.") An excellent lecture was a function not only of the quality of arguments propounded or the evidence provided but also of the context in which these were presented. The vocation of the scholar-as-teacher, wrote Fichte, was to attend to the students immediately before him. His task was to "form, express, and clothe" reason in diverse ways so that individual students could engage it in a particular context. He had to be capable of thinking in "infinite forms" so that he could perform the act and art of thinking for his students, whom he knew and with whom he met weekly outside of the lecture hall for conversations. The lecture was an activity crafted for a particular time and had to be attuned to the needs, capacities, and knowledge of the given audience. Like the humanist teacher, the university professor had to know and love his students.
Fichte understood the scholar's pedagogical vocation, his responsibility for his students, in broader social terms as well. As the "educator of humankind," he wrote, the scholar "exists only in and for society." Scholars represented a distinct social group entrusted with and responsible for knowledge and education and, thus, the development of humankind. The true scholar-teacher was, concluded Fichte, to be the "ethically best man of his time," the exemplary character of the modern age.
Another effect of digital technologies may be the gradual erosion of epistemic authority and legitimacy.
Fichte's ideal university was never realized. The rise of the research university in Germany and the United States over the course of the nineteenth century is more a story of a modern bureaucracy than a flourishing community devoted to reason and intellectual desire. Unlike their humanist predecessors, modern university professors crafted books and experiments, not souls. And their authority was bound to the prestige of their institutions and the systems—such as tenure, peer review, publication—that legitimated them.
The humanist pedagogical ideal provides a model of trust with which to compare Google's "Knowledge-Based Trust." When Sergey Brin and Larry Page started Google as graduate students at Stanford University in the late 1990s, the first generation of search engine companies—such as WWWWorm and the early Yahoo—had established themselves by focusing primarily on categorizing and cataloguing websites. But Brin and Page took a slightly different approach to search. In their earliest work, funded by the National Science Foundation, they sought to design a search engine that would not just identify and catalogue web pages but also assess their quality. As the early web grew in size, from 2,738 pages in 1994 to 2,410,067 pages in 1998, users needed more help. Like maps, search engines are only as good as the navigability they afford. Google presents the World Wide Web in scale—ten sites at a time versus ten million.
In its first decade, as I have pointed out elsewhere, Google 1.0 became the most used search engine by focusing primarily on signals external to websites, especially hyperlinks. The value or quality of a website was largely determined by what other users thought of it, as evidenced by whether people linked to it. A high-quality website, according to Google, was one with a lot of links.
But what distinguished Google 1.0 from other first-generation web search engines was also what made it so limited: it implicitly defined the quality of a page in terms of popularity. A popular yet completely erroneous site from a prankster could have a higher score than an unpopular site from a careful and erudite scholar. Google consistently tweaked its algorithms in order to address what it and others increasingly came to see as a liability. Just because a website was popular didn't necessarily mean that it was trustworthy or true.
When the team of Google engineers announced their work on knowledge-based-trust (KBT) two years ago, they also pointed the way to Google 2.0 and the future of search. The method they described would allow engineers to identify and extract facts from specific websites, estimate their "correctness," and then calculate a "trustworthiness score" for every website. It is an iterative process that uses the redundancy of web-based information—that so many websites have so much of the same information—to determine what sites should be trusted and what sites should not.
Google's potential solution, a KBT score, represents the highest point of an epistemic revolution, one wrought in part by changes in digital technologies and human decisions to use them. Using increasingly common and productive machine-learning techniques, Google's engineers are proposing a fundamental transformation of the very concept of epistemic trust, which Google defines in computational and statistical terms: probability, redundancy, iteration. Google is reshaping trust for a digital age.
Google's attempt to filter digital fact from falsehood is an acknowledgment that the early celebrations of the web as a radically and inherently decentralized medium that would help sustain democratic orders were premature. The proliferation of digital technologies affords not just more access but also more opportunities for individual actors—people, states, and institutions—to spread disinformation. The post-election anxieties about fake news suggest that another effect of digital technologies may be the gradual erosion of epistemic authority and legitimacy. The bigger threat, in other words, is a cultural corrosion of belief, whereby citizens and users of digital technologies trust no one and no institution. No democracy could survive such a collapse in authoritative knowledge, knowledge that is widely trusted. Our common challenge, then, is how to work with our technologies and each other to understand what it means to know, trust, and act in an age of information superabundance.
The Google engineers who wrote the initial "Knowledge-Based-Trust" article quoted the eighteenth-century British theologian and logician Isaac Watts in their introduction: "Learning to trust is one of life's most difficult tasks." We could read this as a threatening promise—that we must learn to trust our new machines and algorithms and the multi-billion-dollar company that owns them.
But we could also read the allusion to Watts as a subtle reminder that Google's search algorithms are not wholly separate from humans, and that today trustworthiness resides neither in a purely mechanical domain nor in a purely human one. It resides at the intersection of the two. The engineers who design and manipulate these algorithms and the tech titans who own them are people.
Watts wrote sermons and treatises on logic. In both he counselled his readers to treat trust as a precious gift. He warned that "those who trust to their own understanding entirely are pronounced fools in the Word of God." Trust in the Lord, he implored, but also in the gifts he has given us. And for Watts, the greatest of those was logic, which he defined as the "art of using reason well in our inquiries after truth." Logic, the basis of computation both past and present, is an art, a deeply human practice.
For humanists such as Bruni and Fichte, facts and all that is given had to be taken in, consumed, and remade by humans. Only then could it become knowledge. Their insistence that knowledge is a profoundly human pursuit was not a claim about the infinite capacities of man. It was, rather, an insistence on human finitude and our need for meaning and knowledge scaled to our limits. The humanist insistence on finitude can help us understand the promises and perils of our most recent attempts to know with the help of technology. They can help us understand that the problem with attempts to automate trust is not the supposed delegation of thought and decisions to inhuman machines. It is, rather, the possible shrouding of decision-making—the act of thinking and judgment—within a closed or opaque system. Automation can obscure the humanity and, thus, the finitude of knowledge. The dangers of automation can be all too human.