Positions on Generative AI
I've been writing unpublished essays about AI since Dall-E was announced. My opinions have slowly congealed to the following points. We shall see how well they stand the passage of time.
First of all, Generative AI is not an Artificial General Intelligence (AGI). Much of the fear and discussion1 around the future of GenAI concerns whether it leads to AGI.
Generative AI is an advanced machine learning technology. It relies on Training Data, a machine-reinforced Model, and a Prompt, with the goal of simulating human creativity or behavior… It relies on Training Data, a machine-reinforced Model, and a Prompt, with the goal of simulating human creativity or behavior.
The best explanation I have yet encountered is Stephen Wolfram's What is ChatGPT Doing… and Why Does It Work?. While it can feel like we are interacting with an intelligence, ACT's discussion of Janus Simulators speaks to interacting with a math function that feels but is not sentient in any way2.
The human tendency to anthropomorphize GenAI actions makes this conversation difficult. Even those who believe it's just a mathematical simulation still talk about AI "making stuff up" to say hallucination… when it's not making up, it is simulating based on its training data and model.
