by Martin Conrad, Chair of Humanities, guest columnist
Humans have always had a curious relationship with information — more specifically with the veracity of information. Arthur C. Clarke pointed out that information is not knowledge, and that knowledge is not wisdom. “Is that true?” may be the most common question asked throughout all of human history, both privately and publicly.
I’m not going to jump down the rabbit hole of exploring “truth” (with or without a capital “T”), except to say that — for the purposes of this discussion — all expressions are “true” simply by their existence. “Rabbits are frogs,” may not be biologically factual, but it is a grammatically correct sentence — it contains a subject, a verb, and a predicate nominative — and, therefore, it is a “true” sentence.
By this point, you’re probably wondering whether perhaps I’ve hit my head, or maybe you are contemplating accusing me of engaging in sophistry. Fair enough; I’ll get to the point (well, I’ll start angling in that direction, anyway).
It seems to me that recent anxieties concerning artificial intelligence (AI), are simply the next chapter in our human struggle around our relationship with information. It’s not a strange new creature, it’s simply a horse of a different color.
In Sapiens: A Brief History of Humankind, Yuval Noah Harari opines that the first true purpose of spoken language was gossip, a salient feature of which is its tendency to be sensationalized and exaggerated. He goes so far as to claim that the first true members of the Fourth Estate were gossip-mongers.
The point is that just because someone speaks something to you is no guarantee of the veracity of the utterance. All communication is manipulation; all expression is propaganda. This is not an indictment or a condemnation, it is a simple statement of fact. The purpose of every spoken utterance, every written exposition, every painted tableau, and every sculpted form is an attempt to alter the experience and, by extension, the thought patterns of others.
Again, “rabbits are frogs,” is “true” in some sense, but not in all senses. The challenge for the listener, then, is to think critically; and to examine not only the structure of the statement but also its purpose.
Therein lies the rub (with a nod to the Bard): structure is content (it is what it is), but purpose is context: broadly “What does it mean?”, or more specifically “What is its intended meaning?” An advertisement encouraging you to walk more and drive less is propagandizing you, but not necessarily to your detriment. It’s up to you to exercise agency and liberty, to choose whether or not to participate in the manipulation being directed toward you.
Critical thinking, the battle cry of educators, and the bane of student self-confidence is the ability to ask appropriate questions intelligently. The statement stands as made, but what was the intention of the speaker in stating it? What were the circumstances of the utterance, and do they have any bearing on not only the form of the statement, but its intent, and, most crucially, on its goal?
When we humans passed from primarily oral transmission of information to written forms, we were confronted yet again with the same conundrum. Just because it’s carved in stone (often literally), doesn’t make it “true.” As People’s Exhibit A, I submit Rameses II’s numerous monumental retellings of his “glorious victory” in the Battle of Kadesh. Militarily, it was a disaster, but he turned it into a public relations coup. (“Spin” is far from a recent invention.)
By no means did this problem end with Rameses — there are several documented examples of Joseph Stalin having ordered erased from official Communist Party photographic records numerous associates who had fallen out of favor. One might also briefly mention the scandal that accompanied Matthew Brady’s staging of some of his American Civil War battlefield photographs. George Orwell imagined an entire nightmare governmental policy of regularly and constantly revising historical records to ensure their conformity with the shifting political realities of the moment. On a more quotidian scale, just because a scribe noted that there were twenty-five sheep in a herd, a wise magistrate went and counted for themselves.
As the means of mass-producing the written word came into existence, beginning with woodblock printing in China in the Seventh Century CE and peaking with the innovation of movable type by Gutenberg in the Fifteenth Century CE, the human struggle with the veracity of information took on a yet another new dimension.
Since few persons individually possessed printing presses, their printed words were reproduced and distributed by others (first the individual printers, and later entire printing houses employing hundreds or thousands). This led to an awareness that not only was the printed word not necessarily “true” simply by having been printed, but the number of copies of that printing did not inherently guarantee the veracity of the information, nor did the number of hands through which the printed word passed from inception to reception increase its reliability. Also, the mere fact that a large number of people chose to subscribe (pun intended) to a particular idea, did not not make its declarations objectively veracious, either.
Eventually, in technical arenas, and academia particularly, a tradition arose of asking people who had achieved reputations of honesty and trustworthiness to interrogate written materials for accuracy and clarity before committing them to mass distribution. “Information literacy” proudly took its place beside critical thinking as the dynamic duo of truth-ferreting-out heroes. Peer review became de rigeur, but as People’s Exhibit B, I submit the ridicule and character assassination endured by Alfred Wegener over his proposal of the reality of plate tectonics. He was right, but “common sense” and “scientific wisdom” both declared him a crack-pot.
Recorded sound brought the problems of information literacy into yet a deeper dimension, especially as the ability to edit recordings was perfected. Simply because an audio rendition of someone speaking seemed to “prove” their position on a given subject, how much faith could one put in the “sound bite” being presented? The ability to quote someone out-of-context, already widely documented in the print realm, resurfaced in the audio sphere with much wider impact.
Even if someone actually said what was being played back, was it the entirety of what was said? What were the circumstances in which the utterance was captured? Were there things said which were not recorded? Or, perhaps more concerning, were there things that were recorded and later deleted? Here, enter Exhibit C, the infamous missing 18-½ minutes of the Nixon White House Tapes.
Celluloid motion capture and later video recording (first on actual tape media and now in purely digital format) added yet another layer of debate and uncertainty. It became more widely known, due to the medium of filmed entertainment, that what one “witnessed” on-screen need not necessarily be “the truth, the whole truth, and nothing-but-the-truth.” It is a well-known “fact” among conspiracy theorists that the entire Apollo program — and specifically Neil Armstrong’s historic step onto the lunar surface — was a brilliant filmic fabrication by Stanley Kubrick.
Close on the heels of cinema and television came the Information Age, then the World Wide Web, the Internet, the Information Superhighway, and the eventual (inevitable?) revelation that “just because it’s on the web doesn’t make it true.”
In this particular incarnation of our perennial reevaluation of our relationship with the nature of information, yet another dimension was revealed: with the ease of access to such vast amounts of information, it became habitual among many to simply accept the first information encountered concerning a given subject. Information literacy lost ground to functional expediency.
I recall an anecdote shared by a historian colleague, who was once stunned to find themselves reading a student essay explaining how Germany won World War II. Upon exploring the student’s listed sources, it became clear that the student had encountered as their first “hit” during an internet search an alternate-history website and had proceeded to simply repackage the “information” they found there.
AI, as I said above, represents to me the latest (and probably not the last) genre of our reevaluation of our relationship with information. Recently, the Washington Post ran an article on the inherent bias in AI-generated graphics, citing examples of stereotyped images of domiciles around the world when given prompts along the lines of “a typical Japanese house.”
Similar articles followed, in numerous other publications as well, all of them concluding that AI-generated art and text is inherently biased — even demonstrably racist, sexist, ageist, ableist, etc. The “shocking” realization dawned that AI is biased because its sources are biased.
What surprises me about this is that people were surprised by it. AI can only draw from the information to which it has access. If all one has with which to build a shed is mud and straw, it should hardly be earth-shattering to note that the finished structure isn’t comprised of stone blocks.
There are several levels of the onion to peel back here, of course. The various AI “engines” have been “educated” only on data extant on the internet (and up to a certain date); the limitations imposed by numerous political, economic, and technological realities have “filtered” what data has been published, and subsequently archived electronically; while the internet comprises vast amounts of the products of human expression, it does not by any stretch include all of it (such a thing may actually be physically and technologically impossible).
In short, yes; AI is biased because its sources are biased. I would, to paraphrase a famous Vulcan, accept that as axiomatic. What I’d like to conclude with, however, is that I see this as an unprecedented opportunity for lifting our collective consciousness. If the current data on the internet is biased, this fact provides us with a marvelous opportunity to ask, “Why is that so?” and “What can be done about it?” What existing examples of human expression are missing from the datasets? Why are they missing? How can they be added?
Yes, there will be questions of intellectual property rights — but that is an existing question we’re already confronting. A concerted effort to “digitize” more and more books, paintings, movies, audio recordings, etc., will simply necessitate a more thorough exploration of the moral, ethical, and legal considerations of an existing dilemma. I don’t see how this could have any except positive results. A more complete examination of the subtleties of an issue can only lead to a more equitable policy regarding it.
But, I’d like to add a further level to the question of what can be done about the inherent bias in AI-assisted imagery and text: we can choose to learn from it. When an AI engine produces a biased, stereotyped, or outright discriminatory product, we, of course, should be disappointed, dismayed, frustrated, irritated, angered, even outraged. But we should also be energized and motivated. To what end, you ask? No, not to smash the machines in an orgy of Luddite furor.
Rather, to actively engage (and to encourage others also) in consciously producing artifacts for purposeful inclusion into the various datasets from which future AIs will learn, with the intent to purposely (initially) smooth out the existing biases and, ultimately, overwhelm them with diverse, equitable, inclusive “raw data”. We can both be and make the change we want to see in the world.
Let us allow the unpleasant fact of past human sectarian narrowmindedness with which AI confronts us now to empower us to progressively rededicate ourselves to the worthy goal of growing beyond our past shortcomings and striving toward a more accepting and accepted worldview.
That AI is biased is a truth, because it is true that humanity has been biased in myriad ways throughout its history, right up to the present day. Let us not shrink from this truth, but face it honestly, squarely, and determinedly.
I believe that AI can, if we will let it, help us become more human(e).