Mission and Purpose
Deep Discourse exists to make philosophy more accessible, engaging, and interactive through AI-driven dialogue. There is certainly great value to reading a book from start to finish, but sometimes we read books looking for something that catches our eye or relates to our current conditions
“No, Sir, do you read books through?”
Samuel Johnson (“Life of Johnson”)
There is additional value to the AI discussion format here – it allows anyone to ask questions without knowing what chapter and page they need to look towards for an answer. The AI characters are also given some freedom to combine ideas from different areas in the text, and to infer how their writings apply in situations that they never explicitly talked about.
By creating a forum of AI characters with widely different views of the world, Deep Discourse serves as an educational tool that brings historical wisdom to life in a conversational way.
How it Works
Each AI Character model is built by pairing a modern AI model with a collection of source texts. The character model is trained to review the philosopher’s corpus of written work, then designed to respond in the voice and thought process of each philosopher. This approach is unique among AI character models because we prioritize education, accuracy, and engagement rather than basic citation.
Do the AI philosophers on this website actually understand what they are saying? Large Language Models today are just the industrial-scaled application of a math operation called dot-product attention: In the human sense of knowledge and learning, they do not understand anything. But that doesn’t mean they aren’t incredibly useful for recalling and synthesizing content based on the corpus of human knowledge, as Plato reminds us:
“Poets utter great and wise things which they do not themselves understand.”
Plato
The model is constrained to generate responses only if the information in their alignment text supports it. To help verify this, the texts are carefully indexed and the AI is required to cite text references in the messages it generates. But, this has to be balanced with making an interactive experience – otherwise the AI character is no better than a search engine! So the models are encouraged to paraphrase, combine, and deduce new ideas based on the text, rather than just repeat it. It is discouraged from extrapolating and editorializing concepts for which it cannot find self-references. Each model has a hand-tuned novelty level (“temperature”) that we believe provides a good balance of free discussion and staying true to the text.
What Makes It Unique?
Versus Traditional Book
First, it’s easy to access the most relevant wisdom by asking your specific questions – you don’t have to search a book’s index, watch a full video or podcast, or look for opinions in an online forum. Deep Discourse presents the real words of real philosophers translated into a chat format. The carefully designed characters are true to their text, no different (and perhaps more faithfully) than a human translator from one language to another.
Second, it puts the collected works of thousands of years of thought, from around the world, in a single place. You can compare their thoughts side-by-side when asked the same question, and soon we will be able to create virtual forums where you can submit a topic to multiple AI characters at once, and they will discuss it among themselves to present their own views and possibly gain new insights from one another.
Versus Other AI Tools
Our modeling approach is centered around Verification of ideas against the source text. Other AI character models are given a persona and voice but are otherwise free to wander away from the source material. To represent our historical characters fairly and maximize the value of this website, we have ensure the statements that our AI philosophers make are closely linked to the original writings.
Who This Is For
- Anyone curious about life’s big questions
- Writers and researchers looking for inspiration
- Casual learners
- Philosophy students
Future Plans & Expansion
Today, we are working to build a more complete collection of philosopher AI characters. We’re starting with the most popular and influential names, balanced with a broad representation of different views and historic periods. Based on familiarity we are mostly pursuing the thread of Western philosophy, probably biased 90% or more towards it. The current goal is to embody most of the philosophers that can be found on sqapo.
After the individual chat characters are built, we will be building an interactive tool for the citations the AI presents. First, it will just help scroll to the right passage in the text that matches the citations in the chat.
Once the characters are stable and vetted for conversation individually, we will create a forum where they can interact with eachother. First, by simply allowing a user to ask a question to all philosophers of all time, arbitrate the ones who have strong opinions on the topic, and provide the different perspectives from each relevant character. Then, to allow the AI characters to move beyond single question/answers to a full discussion or argument with each other that you can read and supervise.
We are also prototyping a “zoomable” layered abridgement of source texts. It starts with raw text and uses a commercial AI model to recursively summarize the text to half the previous length. This is allowing us to create abridgements that range in length from half the original, to a few short paragraphs, even down to a single phrase or word which is the epitome of the entire document. You will be able to “zoom” in and out of the text like a map of the writing. For instance, you could skim the short summary for an idea of an interesting topic, zoom in a little to see a section that relates to your own life or work, then zoom all the way in to read the original words. Another example – you could start out reading the original text, zoom out a little to start moving faster if you get bored, then if something catches your eye you can zoom back down to the original again.
How to Get Involved
Currently, the most valuable way to contribute is just interacting with the characters and reporting on how you feel they don’t measure up to your expectations. Do they cite the wrong source? Do they claim not to know about a topic you are sure they wrote about? Is the conversation feeling too strict to the text? Too loose an interpretation? Any feedback helps us improve the way models are guided to create useful dialogue.
Please make suggestions for adding new names on the requests page!
If you are a technical user who is already working with LLM models, please reach out if you’re willing to help with concepts, training techniques, or computing resources. We appreciate any help.