Limits

Every AI system has boundaries: not just technical, but fundamental constraints that shape what these tools can and cannot do.

Although generative AI feels seamless when we’re using it, it’s grounded in our physical world. Every AI system has built-in limits: finite context windows, knowledge cutoff dates, and static knowledge that doesn’t update. These constraints aren’t necessarily flaws - they’re just part of how the system works.

It’s hard sometimes to think of what makes up the material infrastructure behind what we call the internet “cloud.” Somewhere, real computers are running models, consuming electricity, generating heat, and all of this is built using resources mined from the planet.. The “cloud” really is just someone else’s computer in a massive data centre.

Computation requires energy, and that energy becomes heat. So we’re literally warming the planet with every bit of text we generate from an AI. Every interaction represents a choice about how we use our planet’s finite resources.

All this creates profound inequalities. The enormous computational costs mean only wealthy communities can build and use advanced AI, whilst billions are excluded from these tools. Ironically, many of the excluded communities are the same ones bearing the environmental burden; mining rare earth elements, processing e-waste, and facing the climate impacts.

Understanding AI’s limits helps us ask better questions: not just what AI can do, but what it should do, when we walk away from it, and what we’re willing to sacrifice to build these systems.