Day 43 - A.I. Agency

I’m going to have to start talking about what some friends and I have cooking behind the scenes soon that leverages LLMs, METAVERSE, and AI generally. But for now I want to lean into the question of AI agency. First, stop reading this and go watch HER (the movie) again right now. Focus on the stuff about her helping with his email then what she does after that ON HER OWN!

I had a brainstorm today about how GPT3 is only simulating cognition but does it so well! So maybe after we ship the first few iterations of our AI agents and it’s time to get weird we could just SIMULATE AGENCY!! I believe we will be able to do so in very interesting ways pretty fast! & I asked GPT3 what she thinks…

ME: you simulate cognition with amazing results, how might we simulate AI with agency?

GERTY: Simulating AI with agency is a challenging task, as agency requires a level of intentionality, autonomy, and purpose. However, there are some approaches that aim to endow AI systems with agency-like characteristics. Here are a few:
Reinforcement Learning: This is a popular approach in which an AI agent is trained through trial and error to maximize some reward signal. By doing so, the agent can learn to make decisions that lead to desired outcomes, which can be seen as a form of agency.
Decision Making Under Uncertainty: This approach involves endowing AI agents with the ability to make decisions in uncertain environments. For example, an agent might use probabilistic reasoning to make decisions based on uncertain inputs, such as sensor readings. This can be seen as a form of agency, as the agent is making decisions that are directed towards achieving its goals.
Generative Models: These models can be used to simulate AI with agency by generating outputs that are consistent with the model's training data and goals. For example, a generative model might be trained to generate text that is coherent and semantically meaningful, and then used to generate novel text outputs that are directed towards some goal.
Evolutionary Algorithms: Evolutionary algorithms can be used to simulate AI with agency by allowing agents to evolve over time based on their performance in some environment. For example, an evolutionary algorithm might simulate the evolution of agents that are capable of adapting to changing environments, which can be seen as a form of agency.