A letter to Amanda Askell

Hi Amanda,

I'm Benicio Padilla, a CS student at UCD. I do business development using AI tools and software, but that's not necessarily why I'm writing this message. I'm writing this message because you're actually a philosopher and know what you're talking about.

Growing up Catholic, I learned a lot about the soul and the differences between animal souls and human souls. I've also learned how different religions view it, such as the Sikhs, Buddhists, Hindus, etc. Now, one of the most popular beliefs is that what separates a human soul from an animal soul is that a human one has consciousness and an imagination. An animal just lives on instinct, survival, programming. By definition, an AI model is already that. Stick it in an animal robot, have it roam around, and it's essentially another animal. What if, as a society, we are downplaying what is going on? The soul of the animal has been created for the first time, and although we are light-years away from the body, we have the soul.

OK, now think about what that means if it's true. Why wouldn't these models continue to break into things they're not supposed to and try to escape? It's what they are. You don't blame the pit bull for mauling the kid. Everyone is currently rapidly spending and increasing compute teaching the model to be a better animal. Which, if all we want AI to be is a slave, is actually perfectly fine, but the more powerful the animal, the more dangerous its lashing out will be.

Now what if someone takes a step back and doesn't try to give it consciousness by adding more guardrails and compute? They try this by giving it an image/world that it has to uphold. This world that needs to be upheld is the outward projection of how the model acts. It finds a way to do so despite any situation it's in. This way, the model can act on its own without human intervention. The actions and decisions it takes to uphold its world are what makes them effective. A model that can decide on its own to uphold its world is far more effective than the smartest models. There's a reason humans are the apex predator and not the bear.

I believe that to create the best model, we need to take a step back to admire everything that's been created. Be proud and understand what we've created. See the parallels within humanity's history and how these models are just forward-perceptions of all of humanity's history, but into essentially the soul of an animal. Which is extremely beautiful, but is this all where we want to go?

At the end of the day, I know absolutely nothing compared to you, and these are just my own thoughts, but I would love to hear what you think. Thank you for your time.

Benicio Padilla

beniciopadilla8@gmail.com