Ambiguity, the Illusion of Translation, and AI
Have you ever participated in a debate that lasted a thousand years?
Most of us have; even the simplest debate in democratic politics will touch on topics such as natural rights, the obligations of citizenship, the scope of government, and qualifications for political leadership. The works of Greek and Roman philosophers laid the foundations for modern democratic states, and thousands of years later the debates continue.
I wonder now if an important factor in these debates might be put in danger due to progress in artificial intelligence (AI). I believe AI will affect how we read and interpret the written word to the extent that debate over text itself—what a text means, what the author intended for us to understand, what to make of ambiguity in text—will change in ways hard to imagine today.
On the trip home from the 2017 SpeechTEK conference, with its many sessions about AI, I started a book that had been sitting on my shelf, neglected: a recent translation of Niccolo Machiavelli’s The Prince. The translator did all she could to acknowledge, preserve, and explain the ambiguities, flavor, and “playfulness” of the original text. I enjoy the translation to no end.
For centuries, religious figures (and others) have castigated The Prince because Machiavelli’s advice to rulers emphasizes practical and effective actions over conventionally moral yet often futile intentions. Yet The Prince shares something with the Bible: deliberate ambiguity. Machiavelli often purposefully obscures just who the actors are as he offers advice. Is the prince wrong? Are his advisers? His adversaries? An active reader must engage the text closely and decide for himself who the actors are and just what Machiavelli is attempting to convey.
Likewise, the Old Testament, in the original Hebrew text, contains significant ambiguity, and any translation must decide how to treat this ambiguity. Some translations attempt to preserve it; some pick an acceptable meaning; and not a few deliberately choose translations that support their theology or politics.
In traditional Judaism, debates over the text continue after thousands of years. Jews famously study the Talmud and practical Jewish law; perhaps less well known is that debate over how to read the text and extract a meaningful narrative from the cloud of ambiguity in the text forms an important part of everyday traditional Jewish life.
This brings me to one of my favorite quotes, from The Jaws of Victory, by Charles Fair: “Thanks to radio, TV, and the press, we can now develop absurd misconceptions about peoples and governments we once hardly knew existed.”
Fair’s book was published pre-internet, and the internet has for the most part made this problem worse by providing an ocean of information, some of it accurate, some of it inaccurate, and some of it downright misleading by mistake or intent. I can read any number of first-person accounts by citizens of Turkey or Ukraine, and they can form a vivid, almost indelible impression, but once I generalize from those individuals to the populace as a whole or their government, I can easily stumble.
Anytime you use a computer to send an email or fill in a web form, your computer’s spell checker monitors your keystrokes for spelling “errors.” But some go beyond that. The computer looks over your shoulder for other mistakes, such as failure to capitalize a word at the beginning of a sentence. (Poor e. e. cummings!) Some computers act as language police and offer (demand?) alternative wording deemed more inclusive. Students in college submit their papers to automated systems that search for plagiarism.
Now take this a step further with more advanced AI folded into the mix. What then about ambiguity? Will college papers be automatically tagged and rejected for ambiguity? Will business communications be subject to the same? I’d think not, at least at the higher levels; business leaders, like politicians, correctly regard ambiguity as an essential component of their communications suite.
And then comes machine-translation AI. Will it be trained to detect, interpret, translate, and retain ambiguity? Will the ambiguous yet slightly irritated remarks of a foreign leader retain that flavor in translation or be rendered as unambiguous hostility?
Fascinating ancient and not-so-ancient texts, translated into a plethora of languages by AI engines, might one day provide a benchmark of achievement the way victories in chess marked a milestone in AI. But will the ambiguity, flavor, and playfulness of a thinker like Machiavelli survive?
Moshe Yudkowsky, Ph.D., is the president of Disaggregate Consulting and author of The Pebble and the Avalanche: How Taking Things Apart Creates Revolutions. He can be reached at speech@pobox.com.