Search
Close this search box.

Tag: NLP

AI Artistic Parrots and the Hope of the Resurrection

Guest contributor Dr. Scott Hawley discusses the implications for generative models and resurrection. As this technology improves, the generation of new work attributed to the dead multiply. How does that square with the Christian hope for resurrection?

“It is the business of the future to be dangerous.”

(Fake) Ivan Illich

“The first thing that technology gave us was greater strength. Then it gave us greater speed. Now it promises us greater intelligence. But always at the cost of meaninglessness.”

(Fake) Ivan Illich

Playing with Generative Models

The previous two quotes are just a sample of 365 fake quotes in the style of philosopher/theologian Ivan Illich by feeding a page’s worth of real Illich quotes from GoodReads.com into OpenAI’s massive language model, GPT-3, and had it continue “writing” from there. The wonder of GPT-3 is that it exhibits what its authors describe as “few-shot learning.” That is, rather than requiring of 100+ pages of Illich as older models, it works with a few Illich quotes. Two to three original sayings and the GPT-3 can generate new quotes that are highly believable.

Have I resurrected Illich? Am I putting words into the mouth of Illich, now dead for nearly 20 years? Would he (or the guardians of his estate) approve? The answers to these questions are: No, Explicitly not (via my use of the word “Fake”), and Almost certainly not. Even generating them started to feel “icky” after a bit. Perhaps someone with as flamboyant a public persona as Marshall McLuhan would have been pleased to be ― what shall we say, “re-animated“? ― in such a fashion, but Illich likely would have recoiled. At least, such is the intuition of myself and noted Illich commentator L.M. Sacasas, who inspired my initial foray into creating an “IllichBot”:

…and while I haven’t abandoned the IllichBot project entirely, Sacasas and I both feel that it would be better if it posted real Illich quotes rather than fake rehashes via GPT-3 or some other model.

Re-creating Dead Artists’ Work

For the AI Theology blog, I was not asked to write about “IllichBot,” but rather on the story of AI creating Nirvana music in a project called “Lost Tapes of the 27 Club.” This story was originally mis-reported (and is still in the Rolling Stone headline metadata) as “Hear ‘New’ Nirvana Song Written, Performed by Artificial Intelligence,” but really the song was “composed” by the AI system and then performed by a (human) cover band. One might ask, how is this is different from humans deciding to imitate another artists?

For example, the artist known as The Weeknd sounds almost exactly like the late Michael Jackson. Greta Van Fleet make songs that sound like Led Zeppelin anew. Songwriters, musicians, producers, and promoters routinely refer to prior work as signifiers when trying to communicate musical ideas. When AI generates a song idea, is that just a “tool” for the human artists? Are games for music composition or songwriting the same as “AI”? These are deep questions regarding “what is art?” and I will refer the reader to Marcus du Sautoy’s bestselling survey The Creativity Code: Art and Innovation in the Age of AI. (See my review here.)

Since that book was published, newer, more sophisticated models have emerged that generate not just ideas and tools but “performance.” The work of OpenAI’s Jukebox effort and artist-researchers Dadabots generate completely new audio such as “Country, in the style of Alan Jackson“. Dadabots have even partnered with a heavy metal band and beatbox artist Reeps One to generate entirely new music. When Dadabots used Jukebox to produce the “impossible cover song” of Frank Sinatra singing a Britney Spears song, they received a copyright takedown notice on YouTube…although it’s still unclear who requested the takedown or why.

Photo by Michal Matlon on Unsplash

Theology of Generative Models?

Where’s the theology angle on this? Well, relatedly, mistyping “Dadabots” as “dadbots” in a Google search will get you stories such as “A Son’s Race to Give His Dying Father Artificial Immortality” in which, like our Fake Ivan Illich, a man has trained a generative language model on his father’s statements to produce a chatbot to emulate his dad after he’s gone. Now we’re not merely talking about fake quotes by a theologian, or “AI cover songs,” or even John Dyer’s Worship Song Generator, but “AI cover Dad.” In this case there’s no distraction of pondering interesting legal/copyright issues, and no side-stepping the “uncomfortable” feeling that I personally experience.

One might try to couch the “uncomfortable” feeling in theological terms, as some sort of abhorrence of “digital” divination. It echoes the Biblical story of the witch of Endor temporarily bringing the spirit of Samuel back from the dead at Saul’s request. It can also relate to age-old taboos about defiling the (memory of) the dead. One could try to introduce a distinction between taboo “re-animation” that is the stuff of multiple horror tropes vs. the Christian hope of the resurrection through the power of God in Christ.

However I would stop short of this, because the source of my “icky” feeling stems not from theology but from a simpler objection to anthropomorphism, the “ontological” confusion that results when people try to cast a generative (probabilistic) algorithm as a person. I identify with the scientist-boss in the digital-Frosty-the-Snowman movie Short Circuit:

“It’s a machine, Schroeder. It doesn’t get pissed off. It doesn’t get happy, it doesn’t get sad, it doesn’t laugh at your jokes. It just runs programs.”

Short Circuit

Materialists, given their requirement that the human mind is purely physical, can perhaps anthropomorphize with impunity. I submit our present round of language and musical models, however impressively they may perform, are only a “reflection, as in a glass darkly” of true human intelligence. The error of anthropomorphism goes back for millenia, however, the Christian hope for resurrection addresses being truly reunited with lost loved ones. That means being able to hear new compositions of Haydn, by Haydn himself!

Acknowledgement: The title is an homage to the “Stochastic Parrots” paper of the (former) Google AI ethics team.

Scott H. Hawley is Professor of Physics at Belmont University and a Founding Member of AI and Faith. His writings include the winning entry of FaithTech Institute’s 2020 Writing Contest and the most popular Acoustics Today article of 2020, and have appeared in Perspectives on Science and Christian Faith and The Transhumanism Handbook.

AI Reformation: How Tech can Empower Biblical Scholarship

In a past blog I talked about how an AI-enabled Internet was bound to bring a new Reformation to the church. In this blog, I want to talk about how AI can revolutionize biblical scholarship. Just like the printing press brought the Bible to homes, AI-enabled technologies can bring advanced study tools to the individual. This change in itself can change the face of Christianity for decades to come.

The Challenges of Biblical Scholarship

First, it is important to define what Biblical scholarship is. For those of you not familiar with it, this field is probably one of the oldest academic disciplines in Western academia. The study of Scripture was one of the primary goals for the creation of Universities in the Middle Ages and hence boasts an arsenal of literature unparalleled by most other academic endeavors. Keep in mind this is not your average Bible study you may find in a church. Becoming a Bible scholar is an arduous and long journey. Students desiring to enter the field must learn at least three ancient languages (Hebrew, Greek and usually Aramaic or Akkadian), German, English (for non-native speakers) and usually a third modern language. It takes about 10 years of Graduate level work to get started. To top that off, those who are able to complete these initial requirements face dismal career options as both seminaries and research interest in the Bible have declined in the last decades. Needless to say, if you know a Bible Scholar pat him in the back and thank them. The work they do is very important not only for the church but also for society in general as the Bible has deeply influenced other fields of knowledge like Philosophy, Law, Ethics and History.

Because of the barriers of entry described above, it is not surprising that many who considered this path as an option (including the writer of this blog) have opted for alternative paths. You may be wondering what that has to do with AI. The reality is that while the supply of Bible scholars is dwindling, the demand for work is increasing. The Bible is by far the most copied text in Antiquity. Just the New Testament alone has a collection of over 5,000 manuscripts found in different geographies and time periods. Many were discovered in the last 50 years. On top of that, because the field has been around for centuries, there are countless commentaries and other works interpreting, disputing, dissecting and adding to the original texts. Needless to say, this looks like a great candidate for machine-enhanced human work. No human being could possibly research, analyze and distill all this information effectively.

AI to the Rescue

As you may know, computers do not see the world in pictures or words. Instead all they see is numbers (0s and 1s to be more exact). Natural Language Processing is the technique that translates words into numbers so the computer can understand it. One simple way to do that is to count all the times each word shows up in a text and list them in a table. This simple word count exercise can already shed light into what the text is about. More advanced techniques will not only account for word incidence but also how close they are from each other by meaning. I could go on but for now suffice it to say that NLP starts “telling the story” of a text albeit in a numeric form to the computer.

What I describe above is already present in leading Bible softwares where one can study word counts till Kingdom come (no pun intended). Yet, this is only the first step in enabling computers to start mining the text for meaning and insight. When you add AI to NLP, that is when things start getting interesting. Think more of a Watson type of algorithm that you can ask a question and it can find the answer in the text. Now one can analyze sentiment, genre, text structure to name a few in a more efficient way. With AI, computers are now able to make connections between text that was only possible previously by the human mind. Except that they can do it a lot faster and, when well-trained, with greater precision.

One example is sentiment analysis where the algorithm is not looking for the text itself but more subjective notions of tone expressed in a text. For example, this technique is currently used to analyze customer reviews in order to understand whether a review is positive or negative. I manually attempted this for a Old Testament class assignment in which I mapped out the “sentiment” of Isaiah. I basically categorized each verse with a color to indicate whether it was positive (blessing or worship) or negative (condemnation or lament). I then zoomed out to see how the book’s  sentiment oscillated throughout the chapters. This laborious analysis made me look at the book in a whole different lens. As AI applications become more common, these analysis and visuals could be created in a matter of seconds.

A Future for Biblical Scholarship

Now, by showing these examples I don’t mean to say that AI will replace Scholars. Algorithms still need to be trained by humans who understand the text’s original languages and its intricacies. Yet, I do see a future where Biblical scholarship will not be hampered by the current barriers of entry I described above. Imagine a future where scholars collaborate with data scientists to uncover new meaning in an ancient text. I also see an opportunity for professionals that know enough about Biblical studies and technology becoming valuable additions to research teams. (Are you listening Fuller Seminary? How about a new MA in Biblical Studies and Text Mining?). The hope is that with these tools, more people can be involved in the process and collaboration between researchers can increase. The task of Biblical research is too large to be limited to a select group highly educated scholars. Instead, AI can facilitate the crowdsourcing of the work to analyze and make meaning of the countless text currently available.

With all that said, it is difficult imagine a time where the Bible is just a book to be analyzed. Instead it is to be experienced, wrestled with and discussed. New technologies will not supplant that. Yet, could they open new avenues of meaning until now never conceived by the human mind. What if AI-enabled Biblical Scholarship could not just uncover new knowledge but also facilitate revelation?