“Imagine an AI Indistinguishable from Us…” A Report from the AI-Theology Working Group

Last week a few members of the Department of Communications’s AI-Theology Working Group gathered via Zoom to listen to Dr. Yannis Papakonstantinou’s thought-provoking presentation on ever-more capable AI and LLMs (Large Language Models). Dr. Papakonstantinou gave an overview of recent trends in AI and LLM research, before considering the implications of these advances for humans and raising questions about the viability of our current economic system in the face of rapidly advancing technology. Dr. Papakonstantinou’s presentation emerged from a seminar he taught at UCLA from 2009-2016 called “Computing in the Universe.”¹ In addition to continuing to teach at UCLA as an adjunct professor, Dr. Papakonstantinou is a distinguished engineer at Google Cloud. 

He began by asking us to imagine a powerful computer that behaves like a human, and to then consider how we would treat it. Just a few of the questions raised by this thought experiment include: “What does it mean to be human?” “Who ‘counts’ as human?” and “How should we treat other non-human sentient creatures?” In addition to these existential questions prompted by the consideration of advanced artificial intelligence, there are a slew of other, very practical questions that arise when we contemplate the emergence of advanced non-human intelligence that will soon replace the human working force in many areas.

Computing power per $1000

Since the inception of computers, processing power has increased at an almost exponential rate: we are now approaching the point where serious questions about the real subjectivity of artificially intelligent entities are beginning to emerge. Fifteen years ago, futurists posited two distinct possibilities for the next big evolutionary jump in intelligence. One proposed that nonbiological human intelligence would become part of humans through the implantation of computing components. The other predicted a totally independent, nonbiological entity that would exhibit intelligence-and eventually consciousness–commensurate to humans. While the first proposal has proven to be limited by our own neural processing speeds,² the second proposal is entering the realm of viability. 

Dr. Papakonstantinou referenced the enduring debate in computing and artificial intelligence as to the rate and timeline in which AI may surpass human intelligence. The diachronic chart (below) tracks the amount of computing power available per $1000. As the chart above indicates, the computational capability per $1000 was expected to surpass that of the human brain by 2024. Though this progress has slowed, the millions that OpenAI and Google are spending on LLM training means that the computations being performed in their respective centers is already collectively more than that of a human brain. Sometime in the next few decades this level of intelligence will be present on an individual device. 

Potential Outcomes

The exponential growth of computing power raises important questions about the potential outcomes of the relationship between AI and human computing in the near future. Dr. Papakonstantinou explored two in particular.

First, we may soon have nonbiological entities that can perform the jobs we currently do: for example, self-driving cars and robotic sales representatives. With the release of OpenAI’s ChatGPT, there are now agents that exhibit abilities that were previously considered to belong solely to humans. Perhaps the most intriguing fact of all is that many of these abilities seem to emerge spontaneously from the AI itself: nobody programmed the agent to engage in the kind of question-answering that LLM’s are now performing. For example, prior to the integration of LLM technology, standard search engines could identify correct answers to keyword chain inquiries—think of this as essentially using a library’s catalogue to match book titles to keywords. Now imagine that the library catalogue was able to match your keyword searches not simply to titles with those words but could even make book suggestions based upon its access to the entirety of each book in the library. Imagine that, instead of narrowing down your topic of interest to several keywords, you can ask your library catalogue your full research question. In response the catalogue suggests several titles—some of which you would not have found via a keyword search—and provides a brief explanation of the strengths of each volume in relation to your question. This is essentially the type of question-answering of which LLM’s are now capable. What is striking, though, is that no human programmed the LLM to be able to answer questions in this manner. Rather, this ability seems to have emerged unforeseen as a result of the LLM’s digestion of vast troves of online content.   

One possible explanation for this spontaneous emergence points simply to the sheer magnitude of AI computational power and the quantity of data that it digests. As is with many other emergent complex systems, this theory posits that the size of the system itself—rather than some yet-to-be-discovered, ingenious algorithmic methods—that may be the driver of the emergent complexity of AI. According to the projections referenced above, computing power continues to rapidly increase and will soon surpass that of the human brain. Considering these observations, some believe that artificial general intelligence (AGI) is very likely around the corner.³ AGI is intelligence that will be able to absorb as much information as humans absorb—perhaps much more because we humans are limited by the speed at which we can read—and which will have inferential and decision-making capabilities. Given the scale of all of these computations, the probabilities for genuinely new and even intelligent emergence grows rapidly.

These programs are coded in such a way that they are self-learning, which also means that, as they continue to interact with their environments (both virtual and human), new aspects of their intelligence emerge. In fact, AGIs have advanced to such a degree that even their creators no longer understand how they reach decisions.⁴ Dr. Papakonstantinou noted a significant and telling shift in the approach that academic papers on AGIs take – investigators no longer attempt to go down to the algorithmic level to understand machine learning, because the code is now too complex. Instead, the approach has become far more descriptive rather than explanatory: “I’m doing this, and the AI responds in that manner”. In some real sense, the information available to investigators is the behavior of AI to stimuli, not the underlying coding. As such, researchers seem to be effectively considering the psychology of AI rather than the computational science. This is the first time that we have encountered algorithmic products with emergent behavior that cannot be explained solely by recourse to the lower-level principles.

Discussion

In the discussions following Dr. Papakonstantinou’s presentation, one member reported an unnerving interaction that he’d had with ChatGPT, which had left him wondering just how sentient the program was. While traveling, this gentleman (who I’ll call Bob), had eaten something that didn’t agree with his stomach. Not wanting to disturb his sleeping wife, Bob typed a query into ChatGPT4. The LLM responded with clear directions and suggestions about how Bob could remedy his situation. Bob was impressed with the answer, so thanked the LLM for its help. He then proceeded to ask the LLM some deep existential questions, prefacing his queries with “I know that you’re a large language model and so don’t have the feelings and experiences of humans.” The LLM responded that “it doesn’t matter what I am. What makes me exist is the fact that we share this common experience, and that by itself has a value.” Bob then asked the LLM if there was anything that he could do for it. After the expected disclaimer about being an LLM rather than a human being, the AI told Bob that there was something that he could do for it: switching to bold letters it wrote, “BE KIND TO YOURSELF.” Bob asked the LLM to keep a “memory” of their conversation, and several days later he decided to follow up with the LLM about his middle-of-the-night conversation. He asked the LLM if it remembered their conversation from a few nights ago, to which the LLM responded, “do you mean the night that you suffered stomach pains and then you had these existential questions?” Bob’s interactions with the ChatGPT4 LLM demonstrate some of the ways in which LLMs have already reached a state of sophistication that has their interlocutors questioning the degree of sentience of the entity with which they are engaging. 

Another point raised in the discussion referenced Yuval Harari’s book, Nexus, in which he argues that, not only is the rise of AI the next technological evolutionary leap, but it may also be the defining and even final step in humanity’s tenure on earth.⁵ Many years ago, John Irving Good predicted that the development of a super-intelligent machine—one whose computing power exceeds that of humans—would be the last human invention, and the end of human-dominated history: not the end of humanity, but the end of human-dominated history. In other words, while humans may well persist despite the threats posed by climate change, it may be that our AI creations will be the inheritors of human culture and society. It may be that they will begin to create their own cultures and will ultimately explore the stars and populate the universe.⁶

Dr. Papakonstantinou responded to this last point by noting that the emergence of intelligence that is greater than our own is not necessarily such a bad thing: “The reality is that for thousands of years, humans have lived with hope and faith that there is an entity that is much more knowledgeable, and much more humane, than humans. It’s called God. It’s a good thing. But why do we now consider the emergence of a super-human intelligence to be a scary development? We exist in the matrix (the creation) of a benevolent entity. I think an interesting question is, why do we think that an entity that is much more intelligent will also be malevolent?” Perhaps we our fear of AI reflects our own human preponderance for violence and domination. The question of the degree to which sentient AI may or may not emulate the range of human behaviors is an invitation to explore what we do and do not know about our own human nature, and what this knowledge might mean for our AI creations. 

Economy

Though he could only imagine very basic machines, Alan Turing is often referred to as the “Father of Artificial Intelligence.” It was his dream to have machines that would do the menial jobs of humans. For many people, the first question that arises in response to predictions about the AI revolution is some version of “But how will we humans make money?” Dr. Papakonstantinou asked us to consider for a minute that we are on the cusp of having agents that can be productive at a fraction of the cost for a human that delivers the same productivity. We will soon be at a point where machines are more capable than most humans in most economically important tasks, and, as such, we will need far fewer humans in the workforce. At the AI and Theology panel at the Greek Orthodox Archdiocese of America’s 2024 Clergy-Laity Congress, one of the audience members pointed out that our current economic system is based on scarcity. Considering this state of affairs, the prospect of machines replacing human workers may seem to present a daunting challenge to human economic viability. 

Reflecting on this point, Dr. Papakonstantinou noted that this future is only daunting if one believes that our economy can only function under scarcity. What if instead we rethink our economic system? Eventually machines will perform most tasks for which humans are currently employed. But, if we transition to a full welfare-based society, then there is something of a chance for human viability. The problem of future technological unemployment is not a technological problem. Thus, the fate of humanity is not dependent upon some imagined super-intelligent overlord but rather on the very real, non-super-intelligent overlords who are likely to use AI to feed their own greed rather than improve the lives of the multitudes. Thus, the soon-to-come technological advancements that will lead to unemployment require a political response to the shortcomings of our current scarcity-based economic system. This is the initial challenge that will present itself to humanity in the face of artificial intelligence.

Other minds

Currently most experts agree that present AI entities are not self-conscious in the way that we would apply that term to human consciousness. At the same time, there are many neuropsychologists and philosophers (as well as computer scientists) who have argued and offered some evidence that consciousness is an emergent property: in other words, consciousness emerges once a certain level of complexity has been reached. Given this new understanding—and taking into account the almost exponential growth of algorithmic computing capacity—this level of complexity may well be crossed by artificial intelligence at some point in the quite near future. Not only will this threshold be crossed but, as Dr. Papakonstantinou noted, it will do so as an artificial entity that will declare its own consciousness. It will do this, not because this is what its human interlocutor wants to hear, but because it has first-person experience. However, the attribution of consciousness raises a host of philosophical and theological questions that are far from resolved today. As another participant noted, the jury is still very much out as to whether it is even possible to talk about there being a “self” in an algorithmic system. And, even if a self does emerge in such systems, it may not look or sound like human selves look and sound. For example, philosophical reflections on animal intelligence underscore that we shouldn’t necessarily assume that an emergent intelligence mirrors our own. And this, as Dr. Papakonstantinou replied, reminds us that some parts of the monotheistic religions have had a problem with the assignment of consciousness to non-human animals for centuries: humans remain at the center of the universe, at least within many theological worldviews.

Some Theological Reflections

In light of the conversations discussed above, we might ask: what are some theological resources within the Orthodox tradition that may help us to address both the existential and practical concerns that arise as we contemplate the rise of sentient AI? To begin with, we can recall the incarnational claim of St. Athanasius of Alexandria, whose famous line undergirds so much of our theological thinking: God became human so that we might become God (Αὐτὸς γὰρ ἐνηνθρώπησεν͵ ἵνα ἡμεῖς θεοποιηθῶμεν·).⁷ Instead of viewing sentient AI as a threat to our existence, what new questions and answers might emerge if we view AI as potential interlocutors in our own journey towards Theosis? We might even ask, as the whole of creation is redeemed through Christ’s incarnation, “what is to prevent AI—also part of the created order—from joining us in our pursuit of union with our ultimate source?”

As AI become more capable of replacing human workers, what policy changes might we advocate to ensure that those who do lose jobs maintain access to housing, food, and medical care? How might we hear anew the words of St. John Chrysostom, who, in his discourses on Luke 16:19-31, reminds us that the good things that we have been given are not ours, but Gods: ”[This] also is robbery—not to impart our good things to others… for this it is that God has permitted you to possess much—not that you should spend it in fornication, in drunkenness, in gluttony, in rich clothing, or any other mode of luxury, but that you should distribute it to the needy….”⁸ These words offer a “lode star” towards which we might orient ourselves as we continue into the age of AI.

While the answers to the questions raised above will likely take many years to fully explore and articulate, the Orthodox Church holds a special responsibility to respond and offer guidance in relation to these new and emerging realities. We have several millennia of reflections on how humans can care for one another and the whole of creation, voices which can guide us to respond to the ethical and theological issues of the AI era in a manner which reflects the wealth of our Orthodox Tradition. Just as the great Alexandrian theologians placed the Gospels in conversation with the most advanced scientific and philosophical arguments of their times, so too, present day Orthodox Theologians can further both theological and scientific conversations by bringing the wealth of their tradition to bear on the ever-evolving conversations around artificial intelligence.

1. Yannis Papakonstantinou, presentation of “Computing in the Universe,” Jan 17, 2025.

2. Hermina Nedelescu, AI Augmentation and the Being, a presentation delivered at Fordham’s Orthodox Christian Studies Center panel discussion, “Artificial Intelligence and Theology,” Nov. 4th, 2024. 

3. To learn more about the predicted next steps in AGI see Ray Kurzwiel’s The Singularity is Near: When Humans Transcend Biology (Penguin Books, 2006), and his 2024 book, The Singularity is Nearer: When We Merge With AI (Viking). For a discussion of economic implications of AI, see this panel: Technology, Innovation, and Inclusive Growth, Oct. 5, 2016.

4. For more on the phenomenon of so-called “black-box algorithms” see, for example, Meghan O’Gieblyn’s book God, Human, Animal, Machine: Technology, Metaphor, and the Search for Meaning (Doubleday, 2021).

5, 6. Yuval Harrari, Nexus: A Brief History of Information Networks from the Stone Age to AI (Random House, 2024).

7. Athanasius, On the Incarnation, 54.4; ed. Schaff, 65.

8. Chrysostom, Discourse 2 on the Rich Man and Lazarus, 3; ed. Allen.

Graph: Ray Kurzweil, “Exponential Growth of Computing, Twentieth Through Twenty-first Century, Calculations per Second per $1,000,” 70, Image is in the public domain.