Among the reasons Pope Leo XIV gave for selecting the name he did was his desire to address the pressing human questions raised by artificial intelligence—just as his namesake and forerunner Pope Leo XII courageously and profoundly addressed the challenges of the Industrial Revolution. One such question is what A.I. does to our ability to communicate. As we rush to acquire a host of personal A.I. assistants that are designed to speak for us, we might pause and consider: What makes it so difficult for us to speak?
It turns out that the one thing we need help with is the one thing these assistants, by their very design, cannot help us with.
Though it may seem obvious when we consider human perception that we look out at the world and simply report what we see, there may be more than meets the eye. The philosopher Martin Heidegger said something peculiar about language: “We do not say what we see, but rather the reverse, we see what one says about the matter.”
So, for example, one says, “Majoring in philosophy is impractical,” or, along the same lines, “Poetry is useless.” By virtue of these routine phrases, we are primed to see only the supposed lack of utility. We might notice that philosophy majors do not qualify for high-paying jobs as petroleum engineers, but we simply cannot register the fact that philosophy majors end up making more money over the long term than such practical majors as biology and business.
Who is the “one” that scripts our speech for us? The one is each of us when we speak habitually without due consideration. We express what the philosopher Edmund Husserl called “sedimented” judgments, ones that are carried to us downstream from past occasions of thought but without having the full force of present insight behind them. Like the children’s game of telephone, much can be lost in the transmission.
Coding what one says
ChatGPT, we can see, gives voice to what one says. It generates new speech exclusively on the basis of having digitally analyzed 45 terabytes of text concerning what human beings have said previously. The averaged-out result is the voice of the one. The computer scientist Stephen Wolfram writes that “What ChatGPT is always fundamentally trying to do is to produce a ‘reasonable continuation’ of whatever text it’s got so far, where by ‘reasonable’ we mean ‘what one might expect someone to write after seeing what people have written on billions of webpages, etc.’”
If we look under the hood of ChatGPT, we don’t find any texts or knowledge of language. Instead, we find scores and scores of numbers, gleaned from these terabytes of data, that can be used by the system’s neural net to generate, word by word, what we speakers of language can recognize as intelligible prose.
The computers are blind. The speech they generate is not nourished by fresh experience or thoughtfulness. Instead, their prose is parasitic on the experience and intelligence that people have brought to speech. And they have been trained by human teachers to refine their algorithm so that instead of regurgitating nonsense echoed from the human conversation, they are able to regurgitate something that sounds sensible and appropriate.
Hence, these systems originate in a version of what one says, namely a mathematical representation of what has previously been said, and the output is measured by our native sense of what one might say.
Saying what we see
Though we often thoughtlessly repeat whatever one says and notice only whatever one happens to notice, we humans are free to do otherwise. There remains the possibility that experience will nudge us to grasp a little more of the truth, and thereby to bring to speech a little more of what can and should be said. There remains the possibility that we might invest ourselves a little more carefully and thoughtfully in considering the topic at hand.
In Phenomenology of the Human Person, Robert Sokolowski calls attention to how we use speech to take ownership of what is said. When we say, with meaning, “I think that such and such is the case,” we put our own credibility on the line and indicate that we have taken all reasonable precautions to ascertain the truth. When a chatbot uses the pronoun, I,by contrast, there is no self, no responsible person coming through. There is no “agent of truth,” to use Sokolowski’s term.
We might rattle off “Poetry is useless” as what one says about poetry, but we should pause before saying, “I think that poetry is useless.” That phrase carries with it the burden of having actually considered the matter, at least for a moment, and with that consideration comes the possibility of disconfirming what one says. “Hmm. Consider ‘The Iliad’ and the Psalms, and how the spiritedness of Achilles and the jubilation of David have buoyed the spirits of countless people. Poetry may be among the most beneficial of things for human life.”
The fact that A.I. systems are so good at generating text that sounds like human speech can lead us to believe that we are dealing with an individual who is responsible for what is said. But in fact these machines are expert only at echoing back to us what others typically say. The question of truth makes this deficient character plain. The machines, we say, “hallucinate” when they start fabricating truth. But there is no normative dimension to their processes. They follow rules mechanically. They cannot care for truth as we do; they are indifferent to the distinction between truth and falsity except insofar as the terms affect the way we measure their outcomes.
They cannot turn and face the truth or culpably fail to face the facts, not for the trivial reason that they do not have faces, but for the profound reason that they can only do what they in fact do; there is no striving to reach a measure they can fail to reach. They don’t hunger for truth or hanker after the good. We are dealing with creatures that simulate our behaviors rather than duplicating our powers of intellect and of will.
Five ways to live with A.I.
In The Language Animal, Charles Taylor details the central role of speech in human life: We are the animal that speaks and harkens to the voices of others. Yet today, our natural habitat is threatened. The dialogical character of speech is being replaced by an ever-louder monologue in which we are cast in the role of mere auditors for what “one says.”
Now, the switch from what one says to what I think is not automatic but requires effort. Instead of just following the ruts in the wagon trail, the way of least resistance, we have to goad ourselves to blaze a better path if necessary. And it is precisely that additional effort that might induce us to look for shortcuts provided by our growing legion of digital assistants.
In light of this situation, Rainer Maria Rilke’s advice to a young writer is particularly germane: “Go into yourself. Examine the reason that bids you to write; check whether it reaches its roots into the deepest regions of your heart, admit to yourself whether you would die if it should be denied you to write.…Then, approach nature. Then try, like the first human being, to say what you see and experience and love and lose.”
Following Rilke, here are five tips to help each of us realize our human vocation to say what we see instead of what one says.
First, humanize but anonymize the machine. The voice of a digital assistant is nothing more than an algorithmically averaged presentation of the millions of writers that composed the texts on which the assistant was trained. Its voice is at once that of many and of none. Each of us must therefore challenge its authority by returning its anonymous judgments to the truth of the matter.
Second, sink your words down into the soil of living experience. Talk face to face with others whenever possible, making a point to look into the eyes of your interlocutor and then to look with your interlocutor towards the things you are talking about, so that your speech might take its bearings from your joint experience of the things themselves. Be ready and willing to tarry with the real; impatience breeds superficiality.
Third, be mindful of lips and hands. Speech, whether spoken or written, is always the work of one’s own bodily agency. Handwritten notes express one’s thoughts so much more authentically and personally; the philosopher Ludwig Wittgenstein even said he thought with his hands rather than his head.
Fourth, speak off script. Say unexpected things and go out of your way to call attention to what is important and insightful rather than what is expected or typical. Don’t let algorithms write your text, even your pleasantries. Dare to be idiosyncratic and to forge and fashion the human conversation in new ways.
Fifth, rediscover the wellspring of speech. Poetry is the art of saying things beautifully, and philosophy the art of saying things truthfully. Commit several choice poems to memory, and try your hand at writing some yourself. Read some philosophical prose, and dare to bring to light essential truths in your own voice.
We endangered language animals know not only what to say but, more importantly, why. Although it takes a modicum of effort, this is nonetheless our birthright and great joy: to articulate truths freshly and compellingly, to invest our plain words with the substance of our intentions and to say all the words that we know really matter.
