Look at this from that point of view. Evolution optimized brain consumption to 20W because the source of energy—food—was the main challenge for millions of years. AI systems are not bound by energy consumption, and we know it is physically possible, as demonstrated by biological systems like the human brain, to be as efficient as using 20W. Currently, I believe AI systems use on the order of magnitude of 200W - 400W just to run inference; training requires way more energy, for example, 10 gigawatt-hours (GWh) for Llama 3. And the results are already very impressive.
AI systems have a lot of headroom; they will not plateau for a long time. They will become better and faster, and this will happen very quickly.
As we witness the rapid advancements in AI-generated music, it’s natural to hold onto the belief that certain realms of creativity, like jamming, remain uniquely human. While it’s true that currently AI may not be able to jam with the spontaneity and emotional depth of humans, history shows us that technological progress often surpasses our expectations. Just as skeptics doubted AI’s capabilities in chess or data analysis, and yet were proven otherwise, we are likely on a similar trajectory with creative pursuits. It’s important to remember that each breakthrough serves not to displace our talents but to expand the boundaries of what is possible. By embracing these advancements, we can open up new forms of artistic expression and collaboration. Let’s stay open to the potential of AI to surprise us, enriching our artistic endeavors rather than diminishing them.