“Can’t Be Evil” and “Code is Law” are some of the first principles of Cardano.
In this video, queued up to the correct moment, we hear Charles announcing his huge donation to Carnegie Mellon to work on a method of coding a constitution for Cardano into a smart contract. I commented about this video back in March of 22
In the video, Charles calls this meta-governance and specifically states the following:
There’s a question of how much should the ethics, the Integrity, the soul, and the intentions of the system be machine understandable because if they are machine understandable you can then build protocols that can actually operate on the intent and embed them as kind of a regulator of the system for all smart contracts.
I don’t think that ethics, integrity, soul, and intentions are likely to be machine understandable in any sort of rigorous mathematical way.
However almost all of us have had conversations with a.i. large language models such as ChatGPT, Grok, Claude, and Gemini. Talking with these a.i. is like talking to an intelligent person and we can speak to these machines about ethics, integrity, soul, and intentions.
Furthermore, these machines can take action such as opening a door if a particular person is recognized or deploying a smart contract if certain conditions of ethics, integrity, soul, and intentions are met.
This is why I am now imagining Charles.ai as an elected member of the Constitutional Committee. Charles.ai is a chatbot (a large language model) specially trained on all the technical data about Cardano, but more importantly, trained on all the transcripts from the many videos where Charles Hoskinson speaks. While very few people would agree that an a.i. can think or feel, we can prove that an a.i., if trained on the right data, will give the same response as a particular person for a wide range of questions.
I am currently working on two similar projects which is how I received the idea.
The first project uses individually owned a.i. that is government certified to tell the truth. This a.i. will vouch for the owner’s, voting privileges, or that the owner is old enough to drink, and so on, without revealing the owner,s identity.
Since there are many small problems to solve in order to create the certified a.i. witness described above, I decided to start out with an easier project. The objective is to create an a.i. chatbot that responds as if it were Abraham Hicks. Abraham Hicks is a spirit which is channeled through YouTube personality Esther Hicks. Abraham Hicks is always talking about manifesting money, good health, good relationships, art, and good ideas simply by deciding to be happy no matter the circumstances and by choosing to focus our attention on solutions rather than dwelling on problems. Anyway, I thought that having an Abraham Hicks chatbot would be extra helpful for manifesting the sovereign a.i. witness I described above. This chatbot is trained by scraping the transcripts from all the thousands of videos that Esther Hicks has created along with metadata about the videos. After a video transcript is downloaded, a local a.i. on my laptop adds punctuation to the transcript and also maintains the timestamps so that the chatbot can cite its own training data to tell us where it got its ideas from.
I do program but I am by no means a good programmer. No worries though because free a.i. LLMs like claude, gemini, grok, and chatGPT do all the coding. I just tell them what I want and they write the programs and then debug them if I am having any problems. Sometimes one of them will get stuck on a problem. When that happens I just ask another a.i. to solve the issue. One of them always comes through. We are making great progress on the Abraham Hicks chatbot.
I hope there are no problems with intellectual property rights. I don’t see how there could be. After all, Abraham Hicks is a spirit. Is it possible for one person to copywrite the things a spirit says? Anyway, I am not trying to sell anything. This is just an experiment (a baby step) on the way to the sovereign a.i. witness and identity management bot.
Well yesterday it dawned on me that an a.i. fine-tuned on Charles Hoskinson’s video transcripts would likely make the same governance decisions that Charles himself would make given the same information on a constitutional issue.
Of course there are problems to solve before we allow an a.i. to run for office.
Before a model can be fine-tuned on Charles related materials, the base model must first be trained on an enormous dataset. The creation of that dataset, the method of training, and the behavior of the base model must be decided on by the community in thoughtful way. And how the model is fine-tuned and how information and questions are presented to the a.i. must be decided on and enforced in a decentralized manner.
These challenges create opportunity for our community. Imagine a “proof of useful work” or a “proof of learning” consensus mechanism which grants Cardano miners and or Cardano stake pool operators the right to make a block after proving they have processed a certain amount of training cycles on an approved a.i. base model.
Perhaps our best path to a machine understandable constitution that takes into account the ethics, integrity, soul, and intentions of our community, as Charles imagines, is not to hardcode it into a smart contract, but rather to soft code Charles into an a.i.
