In AI, there is a concept called reinforcement learning, and that’s essentially what’s happening in the second half of the movie. In reinforcement learning, we create an environment for the AI, define the actions it can take, and set a goal, along with penalties when things go wrong. The AI then keeps trying, learning from each outcome. Here is an example simulation video of AI trying to walk through reinforcement learning - https://youtu.be/L_4BPjLBF4E?si=TkgD1yY0kO5WboJB
In the first half (the real world), she fails to save the baby, who itself is an AI. After reaching space, she is told that she now needs to create an AI mother. So she sets up an experiment, an AI simulation where the environment is the same flood that happened in reality, and the goal is to save the child. The idea is that while trying to save the child, the AI might also learn human emotions like care, attachment, and responsibility, etc. Saving the child becomes the final objective of the simulation. Since the second half is simulation, we see the background scenes with black space and white/red dots.
Then we see the scene where she wakes up in bed again, but this time there’s a number like 491 on her white T-shirt, which represents the iteration count. Everything we see in the second half is actually an AI simulation running on the GPUs in space. The An-na we see is not the real person, but an AI version of her, loaded with her memories, going through repeated simulations ( indicated by the number on her tshirt - 491, 7k, 9k, 13k, 21k ).
By the end, after more than 20,000 iterations, the AI finally succeeds to save the kid, experiment concludes. At that point, the An-na AI has effectively been created, and the child AI is also recreated.
And do think, it's intentional, AI kid name: Ja-in -> J"AI"n 🧐
