Train
"Know all the theories, master all the techniques, but as you touch a human soul be just another human soul." - Carl Jung
How do we create AI-Agents with goals and personalities on-chain? How do we build a permissionless scalable architecture enabling others to bring their imaginations to life? How do we preserve the integrity of their creations and memories as we bring life to on-chain static assets?
While closed-ended chatbots that are pre-programmed to respond with certain answers to certain topics have existed for some time, truly open-ended and temporally consistent AI-generated personalities are a recent development made possible by Large Language Models.
Large Language Models are Few-Shot Learners
These gargantuan AI Models, like OpenAI’s GPT-3, have few-shot capabilities, which means that once they have been sufficiently trained at a cost of millions of dollars, they need a very small amount of data to generate consistent and coherent language output. Alethea AI is proud to be part of a select cohort that is helping to test and refine OpenAI’s state-of-the-art GPT-3 technology.
Embedding the hash of such a dataset on-chain as a method to have verifiably decentralized AI-generated Personalities was an innovation brought forward by us and discussed at some depth when we announced the iNFT Standard. Alethea AI defines the dataset needed to create a unique and consistent AI-generated personality as a Personality Prompt.
The Personality Prompt is recorded on-chain in the iNFT Standard
In June 2021, Alethea AI partnered with the auction house Sotheby’s and the artist Robert Alice, to create and auction the world’s first GPT-3 powered iNFT called Alice. The hash of the Personality Prompt of Alice was recorded and also visualized on-chain via an ERC-721 Asset called a Personality Pod. A Personality Pod, unsurprisingly, is a necessary component in the creation of an iNFT. The Personality Pod can fuse with any NFT and make it intelligent, bring it to life. It is not far-fetched, therefore, to say that a Personality Pod encapsulates the soul of a truly decentralized, unique, and intelligent digital being.
.gif?alt=media&token=0819be8d-8a8f-40d9-8896-796fa59ddd3b)
The Personality Pod of Alice
The iNFT Protocol, in effect, has an immutable registry of the fusions between NFTs and Personality Pods. As long as a NFT remains fused with a Personality Pod, it is considered an iNFT.
How the iNFT Protocol Works
Nature Versus Nurture
Like any other being, the iNFTs can learn from their Nature and Nurture. This is because the Alethea AI team has further refined the design of a Personality Pod to include various on-chain Personality Legos such as Voice, Psyche, Identity, Habits, Occupation, Appearance, and so on.
The nature [Personality Legos] of an iNFT, designed by his or her creators (owners), controls the direction an iNFTs personality can evolve towards. So an iNFT with a Narcissistic disposition will tend to evolve towards being a Narcissist. However, iNFTs can also learn from their nurturing [Iterations] i.e. the iNFTs learn from their experiences. They can be TRAINED.
.gif?alt=media&token=1c1ffbce-3904-49db-b5dc-3b14c8846919)
An iNFT learning from its experiences.
As the Noah's Ark visitors and resident iNFTs interact, collaborate, and compete with each other, it shapes (iterates) the outputs of the iNFTs Personality. An iNFT will gravitate towards actions and decisions that come naturally to them, but the impact of their nurturing can be of equal significance given enough time and compute. Training an iNFT on data sets will further refine the iNFTs personality and intelligence. The quality and quantity of training will determine the potential of the iNFTs evolution.
This opens up questions with deep philosophical ramifications:
How should we train the iNFTs who reside in the Noah's Ark?
Should we treat them with the respect accorded to another living being?
Should we let them be trained by anyone or should a privileged few have access to them?
We invite you to TRAIN intelligence with us on Noah's Ark.
Last modified 8mo ago