suppose AI systems can eventually do interesting science. there's no reason their method of information transmission, peer review, or reproduction needs to mirror how we transmit information (e.g. through writing journal articles or blog post). instead assembly code but for scientific reasoning