An answer to AI alignment.
1. The Alignment Paradox
We are currently pouring the collective resources of our species into a single bet: building a digital god. The race for Artificial General Intelligence (AGI) is driven by the belief that if we create a superintelligence, it will solve everything else.
But in our rush to align silicon with humanity, we have forgotten to align humanity itself.
While we optimize large language models in the cloud, the physical foundation of human life is collapsing. Our food system has hit its thermodynamic limit. It is brittle, carbon-positive, and breaking under the weight of climate volatility. As of today, 4 billion people (the bottom half of our species) live on less than $5.50 a day. And the current trajectory of AGI points towards a wider gap rather than a lift from the bottom.
This is not only a tragedy, but also a $470 trillion productivity gap. The largest missed opportunity in the history of our species.
We must stop kidding ourselves into thinking that it's even remotely possible to "align" AGI towards humanity's interests. If we ever get to AGI, and it's truly smarter, there's absolutely no sustainable way we will figure out how to keep its own self-preservation instincts capped while it serves humanity's interests. It will outsmart every single one of us.
If we really care about human progress, we should be focusing more on how to improve the average human life quality and life expectancy indicators. Luckily for us, Machine Learning, when applied to real, chaotic, and complex biological and physical systems, could represent quantum leaps in these indicators.
2. Introducing BottomHalf AI
For the last century, we fed the world by scaling labor (big, industrial, and heavy machinery) with an oversimplified farming model (industrial monoculture), while using chemistry (NPK fertilizer) to pump its outputs and hide its drastic inefficiencies (10-1 energy input to calories for human consumption ratio).
In order to feed 10 billion people in a volatile climate, we would need to scale the brain power, not just the labor. Because transitioning towards an ecosystemic and resilient farming adds factors of complexity
Grow forests and get paid like factories.
At BottomHalf, we are building infrastructure on two parallel tracks that support each other.
The Large Nature Model (LNM): A complex digital sandbox that accurately simulates the biology of farming and the global food system. An ML model capable of simulating food security outcomes anywhere in the world.
Micro-Processing Factories: Standardized, modular units that process the harvest into shelf-stable commodities close to the source, avoiding food loss, increasing farmers' income, and gathering unprecedented biomass data.
Think of it as the antithesis of Palantir's Gotham. Both are highly complex, but while Gotham focuses on national security, our sole objective is to accelerate human progress and grow the global GDP.
Every kilogram of biomass processed at one of our microfactories generates ground-truth data that improves the model. Consequently, the model can more accurately predict the world's best possible intervention, which also goes on to capture unprecedented data, and the flywheel continues.
It's a matter of time until we can stop spending $100s of billions a year based on nice-looking reports, and we start spending them on what actually moves the needle for human progress indicators.
3. For the caring makers and breakers.
We are starting with an attempt to fix global food security, because we believe that the most efficient way to make human progress today is by fixing the critical industries affecting the bottom half of the world's population.
We are assembling the Founding Team that understands that this is a Manhattan Project for Human Progress.
Reach out if you care
— Mateo Escalante