The Reality Bandwidth

Commonly mentioned upsides of AGI include curing cancer, reversing aging, overcoming antibiotic resistance and a plethora of humanity's most urgent problems being solved. Well, if intelligence is all we need, then why don’t we point all the agentic intelligence towards these problems? Because…science isn’t just a reasoning problem.

Unlike mathematics, which exists in a rule-based environment where correctness can be established from previous stated assumptions, reality can’t. Where in maths you can reason through a problem, a proof doesn’t need validation from nature, you cannot simulate reality to a tee. Simulations of reality are only faster because they leave something out. Simulations compress experimental reality only to the extent that experiments have told us which details can reliably be compressed. This creates a constraint on scientific advancement. The rate at which we can obtain new, reliable information from the physical world. This constraint I call the Reality Bandwidth.

The Reality Bandwidth = the rate at which a scientific system can reduce uncertainty through reliable interaction with the physical world.

Recently, I have been trying to make progress on ectogenesis. A solution fundamental to solving the falling fertility rates and the oncoming population collapse. Yet, the barriers that I crash into aren’t reasoning barriers (my Codex plan gives me plenty of that) but instead, empirical data barriers. There just simply hasn’t been enough experiments done.

Science isn’t only thinking. A group of 10,000 intelligent researchers knows less about subatomic particles than 10 researchers with access to a Collider. We cannot just rely on reasoning to solve these tricky problems, we need to venture out into the real world. Build complex physical structures, experiment and accelerate our accumulation of novel and valid scientific data, which we then get swarms to reason through.

What is stopping us from curing cancer? At this point, it is difficult to argue that reasoning bandwidth is the only (or even dominant) constraint. The problem isn’t even our amount of data. It’s the rate at which we solve uncertainty through interaction of the real world.

Ok, so how can we fix this?

For us to reap the rewards of AGI, we must widen the Reality Bandwidth.

Here’s how:

Gather more novel information per experiment. If we care about how much uncertainty we can eliminate every second, then asking better questions will lead to more uncertainty eliminated. With the abundance of intelligence we now possess, we will produce and weed out thousands of hypotheses instantaneously whilst scouring all previously documented data to produce incredibly high uncertainty-eliminated/time ratioed experiments.

More experiments. Self-driving labs, microfluidics, biofoundries, robotics and high throughput examination. The future of science will be a closed loop of agents bumping into data gaps, reaching out to agentic labs who autonomously performs experiments with robotic hands, captures all data through sensors and sends it back to the agent who needed that data.

Extract more information from reality. Put scientific instruments and sensors everywhere. Deep, novel data from every interaction that occurs in the physical world. Gather troves upon troves of valuable data which is then made accessible to agents.

Eventually, query reality less often. Simulations are not a bad idea. In fact, they can be brilliant when used and developed appropriately. Giving impeccable simulations to swarms of agents is when we will see an absolute boom in scientific progress. Yet, as I mentioned previously, impeccable simulations are hard, and shouldn’t be weighed heavier than reality in which the simulations are based off of.

Imagine those same 10,000 agents who solved Navier-Stokes. Not confined to a world of symbols, proofs, and simulations, but given hands and eyes in the real world. Microscopes. Sequencers. Bioreactors. Automated cell cultures. Manufacturing equipment. Instruments that can see further, measure finer, and conduct experiments faster than any laboratory before them.

Intelligence wouldn’t be waiting on reality any longer.

An unanswered question leads to an experiment which leads to an answer which leads to progress which leads to better questions. Thought and reality would form a closed loop, turning faster and faster. Hypothesis into experiment, experiment into evidence, evidence into understanding.

Perhaps that is when the promise of AGI truly arrives. Not when machines think without limit, but when their thoughts reach into the world and the world answers back.

For most of human history, intelligence was scarce. We’ve always had more questions than minds capable of pursuing them. We may soon find ourselves in the opposite world: one with an abundance of minds, all waiting for nature to reveal its beautiful, beautiful truths.

— Erik

p.s.

If you liked this, I write a personal newsletter about the frontier, weird ideas and where I think the money is.

I also write ijustvibecodedthis.com, the fastest growing AI coding newsletter with over 90,000+ readers already.