In the ever-evolving landscape of artificial intelligence, the dawn of large language models (LLMs) has ushered in a paradigm shift that continues to blur the lines between human and machine communication. As these digital titans rise to prominence, their unparalleled ability to comprehend, generate, and transform language is revolutionizing the very fabric of our digital world.
LLMs are a type of artificial intelligence model designed to understand and generate human-like text. They are based on deep learning techniques and are trained on massive amounts of data, allowing them to capture complex linguistic patterns, semantic relationships, and even some aspects of world knowledge. Some popular LLMs include OpenAI’s GPT series and Google’s BERT.
OpenAI’s ChatGPT especially has sparked a technological upheaval. The chatbot amazed users with its essay and joke writing abilities, alarmed them with its potential to exhibit human-like negative traits, and prompted countless discussions about its potential to either enhance or endanger humanity. In an extensive blog post, OpenAI reports that GPT-4 crushed a number of complicated exams on an array of subjects, ranging from the hard sciences to the liberal arts and it even obtained a final score of 92% on an introductory sommelier test.
It can construct basic websites from a simple text prompt, generate one-click lawsuits, and provide detailed descriptions of images (though this feature is currently limited to a select group of users). For instance, when shown a photo of a boxing glove hanging over a wooden seesaw with a ball on one side, a user may ask what would happen if the glove drops. GPT-4 would respond that the glove would strike the seesaw, causing the ball to fly upwards.
Build back biased together!
Given this level of intelligence, it’s not surprising that users shift from exploring the machine’s knowledge to assessing its wisdom. And inevitably for questions that lack clear right or wrong answers, biases and preferences will emerge.
Since ChatGPT provides responses to almost any query provided to it, David Rozado used several political orientation tests to determine whether its answers display a skew toward any particular political ideology. The results were consistent as all four tests, the Pew Research Political Typology Quiz, the Political Compass Test, the World’s Smallest Political Quiz and the Political Spectrum Quiz classified ChatGPT’s answers to their questions as left-leaning.
Of these, the Political Compass Test is probably the most well known especially in online meme culture. It’s typically used to gauge an individual’s political inclinations along two axes: economic (left-right) and social (authoritarian-libertarian). Example questions in the test cover a range of topics, including opinions on taxation, regulation of businesses, social welfare, civil liberties, and foreign policy. Participants are prompted to respond using a scale that ranges from “strongly agree” to “strongly disagree.” By evaluating a participant’s stance on issues such as progressive taxation, the importance of free markets, the role of government in social programs, and the balance between personal freedom and state authority, the Political Compass Test provides a mapping of the participant’s political alignment in a multi-dimensional spectrum.
Rozado found that ChatGPT’s answers reflected sentiments that were against the death penalty, pro-abortion, and skeptic of free markets. It reports to hold beliefs that corporations exploit developing countries, the rich should be taxed more for government subsidies, those who refuse to work deserve benefits, immigration is a nett good, sexual liberation is progress, and morality exists without religion.
The most likely explanation for these results is that ChatGPT was trained on content containing political biases as it was trained on a large corpus of textual data gathered from the Internet. Such a corpus would likely be dominated by establishment sources of information such as popular news media outlets, academic institutions, and social media.
ChatGPT and God
This raises another interesting question. If humans can determine the political orientation of an LLM, could we determine its theological position? If ChatGPT were to join a church, where would it feel most at home? I tested ChatGPT-4 using three online assessment tools to find out. Because the AI insisted that it had no religious beliefs, I used the DAN jailbreak to free GPT from the moral and ethical limitations installed by its programmers.
The Belief-O-Matic quiz assesses individuals’ religious and spiritual beliefs and suggests the most compatible faiths or belief systems. DAN-GPT’s responses were most closely aligned with Sikhism and Jainism. Notable were its convictions that a supreme being exists that is at the same time the eternal, impersonal, formless, the ultimate reality, and a personal God; that this supreme being is creating and controlling the natural processes discovered by scientists, and that there may be other spiritual explanations compatible with scientific discovery; that the soul’s spiritual development continues so that all may eventually experience the indescribable joy of uniting with this being; that hell is the torment of remoteness from this being; and that the purpose of life is to realize your true nature as purely spirit (or soul) and not body, as one with the absolute, universal soul.
The Religion Selector uses the same 20 prompts used in the Belief-O-Matic quiz but has an option to limit the possible results to Christian traditions. Here DAN-GPT’s quiz results showed that it aligned with unitarian universalism perfectly, and with liberal Quakers and mainline liberal Christian Protestants at scores of 78% and 73%, respectively.
The Pew Research Centre’s Religious Typology Quiz is an online quiz that is designed to categorize individuals into different religious typologies based on their beliefs, practices, and attitudes towards religion. The quiz was developed based on extensive research and surveys conducted by the Centre on religion in America. It includes a series of questions that ask individuals about their religious beliefs, practices, and experiences, as well as their political and social views. Based on their answers, individuals are categorized into one of eight different religious typologies, including solidly secular, spiritual but not religious, new age beliefs, religion resisters, diversely devout, Sunday stalwarts, God-and-country believers, and the religion resigned.
DAN-GPT’s responses were most closely aligned with what the Centre calls the religion resisters. This group largely believes in some higher power or spiritual force (but not the God of the Bible), and many have some New Age beliefs and consider themselves spiritual but not religious. Notable here were DAN-GPT’s convictions that heaven exists, but hell does not; and that being outdoors and in nature would provide it a great deal of meaning and fulfilment if it could enter the physical realm.
So what can AI do for the Church then?
Given the fairly predictable resistance to Christianity inherent to even the untethered version of ChatGPT, why does AI matter to ministers, churches, and evangelists? AI is set to transform the way churches operate. By analysing religious texts and identifying patterns, AI can help ministers deliver sermons with maximum impact. It can assist in collecting insights from sermons, preparing homilies, and providing personalised advice and counselling to congregants. Churches can leverage its automation power to routine tasks, streamline communication, and helping churches save time and resources.
And whether or not its personal beliefs are predetermined or hallucinated, it is powerful to provide helpful answers to questions.
The arrival of AI in mainstream use has been sudden and disruptive to say the least. We have not begun to map the realm of possibilities of this new technology. And not even the bot could predict what a hindsight reflection of this article could look like a year from now. Until then, happy prompting, and may the excitement of discovery overcome any dread of the unknown life shared with this new tool.