The method is called Tree-of-Thought Prompting.
How does it work? Imagine you’re playing a board game where you have to choose your next move every time.
You don’t just make a move at random, right? Instead, you think about each possible move, imagine what will happen if you make a particular move, and then choose the best one.
In #prompt, we ask ChatGPT to create a team of experts in different fields that are close to the topic of the question. Then we tell them to act in several steps, taking turns speaking and giving themselves from 1 to 8 points, depending on the quality of the answer.
If an expert realizes that they are wrong, they leave, and the group brings in another expert who they think will be better. Experts can use available plugins for research and reasoning, and most importantly, carefully and logically explain their answers.
Here is a simple plugin for ChatGPT3 without access to plugins:
Imagine three different experts are answering this question.
All experts will write down 1 step of their thinking,
then share it with the group.
Then all experts will go on to the next step, etc.
If any expert realises they’re wrong at any point then they leave.
The question is «question»
Here’s a harder one for ChatGPT4 with access to plugins:
Imagine three different experts answering this question (give them appropriate roles and human-friendly names). All experts will write down the next step of their thinking (showing all work and equations) and then share that step with the group. Afterward, each expert will (in another round) share what their next step will be, considering what everyone has said. Then all experts will repeat the process and go on to the next step, etc. If any expert realizes they’re wrong at any point, they leave, and the group brings in another expert they think would be better. The experts use the available plugins for research and reasoning and think carefully and logically in explaining their answers. Based on their research and amount of search hits, they apply a rating to their statements between 1-8. They note this ranking in brackets. And they can contradict each other if the rating of another expert is higher or lower than their own statement. My question is «question»
Here is the answer to the query “What does God look like according to Ai?”
Expert Panel:
- Prof. David Robinson, a Theologian and expert in religious studies.
- Dr. Eleanor James, an AI Ethicist and computer scientist.
- Dr. Peter Watanabe, a cultural anthropologist.
Round 1:
Prof. David Robinson [Rating: 6]: “From a theological perspective, the appearance or form of God varies among religions and philosophies. Many traditions, especially the Abrahamic ones, often describe God as formless, infinite, or beyond human comprehension. Therefore, the answer would largely depend on which religious or philosophical framework you are referencing.”
Dr. Eleanor James [Rating: 8]: “AI does not possess beliefs, emotions, or consciousness. It works on data and algorithms. If you were to ask an AI to ‘imagine’ what God looks like, the AI would most likely produce an image based on the data it has been fed, which would consist of various artistic depictions and descriptions from diverse sources. However, this does not mean AI ‘believes’ in this representation.”
Dr. Peter Watanabe [Rating: 7]: “Throughout history, humans have depicted God in numerous ways based on cultural, geographical, and historical influences. These depictions are influenced by societal values, needs, and the environment. Thus, AI’s representation would be a culmination of these myriad representations.”
Round 2 (after sharing their findings):
Prof. David Robinson [Rating: 5]: “Given Dr. James’ point, AI might generate images that resemble famous artworks or symbols associated with divinity, like Michelangelo’s depiction on the Sistine Chapel or symbols like the Om or Cross. My next step would be to understand the most recurrent symbols of God across major religions.”
Dr. Eleanor James [Rating: 8]: “I concur with Prof. Robinson. If we were to use a neural network trained on religious art, the output would be a blend of these iconic representations. However, it’s crucial to underline that AI’s generated image isn’t a reflection of any inherent ‘knowledge’ or ‘belief’ of the divine but a mere data output. I’ll investigate AI-generated art projects on this topic.”
Dr. Peter Watanabe [Rating: 6]: “Building on both inputs, the depiction could also vary based on the data’s geographical origin. For instance, AI trained mostly on Western art might produce a different result than one trained on Eastern religious iconography. My focus now would be to discern how regional variations in depictions of the divine have evolved.”
Round 3:
Prof. David Robinson [Rating: 4]: “My research indicates that symbols like the Cross in Christianity, Crescent in Islam, and Om in Hinduism are recurrent. However, these are symbols and not direct representations of God. If AI were to rely on symbols, the result might be very abstract.”
Dr. Eleanor James [Rating: 7]: “I found a project where AI was trained on religious artworks. The generated images were often abstract amalgamations of colors and shapes, with fleeting resemblances to known depictions. It reiterates that AI does not ‘understand’ God but can only replicate patterns. I’d now like to see how different AI models interpret the same dataset.”
Dr. Peter Watanabe [Rating: 6]: “Regional variations are fascinating. In Eastern traditions, where pantheons of deities exist, the AI might produce a more diverse range of images compared to traditions with a singular concept of God. I’ll be exploring the anthropological reasons behind these diverse depictions next.”
As the experts continue their rounds, they further refine their understanding. The consensus remains that AI doesn’t possess a personal understanding of God but can generate images based on its training data. If an expert’s approach seems less relevant as discussions progress, they might step back, and the group could consider bringing in a new expert, perhaps an art historian or an AI programmer, for more insights.
