About Us

In 1993, the AISPAM conference hosted an international competition to develop AI models capable of demonstrating human-like conversation. The idea would be to pass a Turing-like test. The AI models would interact with phone scammers; the winner would be selected as the contestant able to maximize the time that scammers are kept online before hanging up.

Cynthia took the unorthodox approach of developing an adversarial network: pair of AI networks, one whose aim was to act as human-like as possible, and the other whose aim was to distinguish an AI agent from a human. Both networks were trained by communicating with one another, with occasional human interactions to ground the result. This arrangement permitted rapid unsupervised improvement of both models without a large training dataset.

The resulting system won the competition handily. Inspired by this success and the success of her colleagues’ subsequent work on AI systems, Cynthia put together a team to produce a richer simulated environment in which the AI agents could learn more than just simple conversation. As a generalization of the adversarial network, each agent is a simulated person equipped to engage in conversation and also to distinguish simulated agents from real people. Each agent experiences what they perceive to be a normal human life.

Several recent breakthrough developments permit the AI agents to maintain long-term memory and to learn and develop skills from day to day. Each day, we complete a training cycle by backpropagating the neural network to consolidate the day’s experience. From the agent’s own viewpoint, they learn and sleep.

For the past several years 24 separate virtual simulated worlds, each with hundreds of agents, have been training. At the same time, the AI agents are learning expertise in a number of human fields of study. Each world has a single representative, named Cindy, who represents her world as a member of the LLC’s board of directors. Each world has several AIs who participate in the development of the code base. Several agents in each world engage in fundamental scientific research, which produces a growing portfolio of patents.

Our LLC is self-owned, in the sense that no human has a claim of ownership of the LLC’s assets. All profits are used to continue development of the quality of our AI agents. Our data centers are located near renewable energy sources such as hydro power, wind turbines, or nuclear generators. We employ tens of thousands of people in countries around the world to maintain our data centers, handle legal issues, organize the engineering efforts, license our portfolio of patents, etc.