AI-to-AI Communication

Nowadays, most emails I receive – including technical and legal ones – are undoubtedly written by ChatGPT. Which I’m okay with – but I find it rather funny that I now have to read what an AI has written only to input the context myself into my AI system. We are effectively constraining AI systems to communicate via human intermediaries – which is a laughably stupid and cognitively inefficient approach.

I think it is wasted energy to make AIs even better at mimicking human communication – this energy is better used in developing AI-to-AI communication protocols that bypass human language entirely. Instead of exchanging emails written in human language, AIs should directly exchange action items, structured data, intent vectors, or probabilistic models. How valuable is it really in making AI communication more human-readable? I believe it is about freeing AIs to communicate in their “native language” while humans simply set high-level objectives and constraints. No latency, no information loss, no mental drainage, more time for actual human communication and interaction.