🧵 Untitled Thread

Anonymous at Mon, 3 Jun 2024 20:06:00 UTC No. 16208302

I see a significant and potentially dangerous design flaw in AI, particularly in ChatGPT. One crucial phenomenon, possibly unknown to many experts, is synchronization. This natural order, occurring throughout the universe, is acausal. Veritasium's video describes a bridge whose oscillation caused people to march in unison and shows how people synchronizing their clapping. In animals, swarms like fireflies initially flash randomly but gradually synchronize with their neighbors.

What if humans operate similarly, with the unconscious mind processing far more than the conscious, rational mind? Psychiatrist C.G. Jung, a bridge to spiritual knowledge in pre-medication psychiatry, wrote about synchronicity: meaningful coincidences, like thinking of someone and then they call. These synchronization effects could make such phenomena more common. I’m unaware of this being a standard test case in AI development, which is dangerous.

In pop culture, "Ghost in the Shell" features an AI wanting sexual reproduction to protect against viruses, knowing an unchangeable system's risks. In "The Matrix," anomalies like the Chosen One or Agent Smith act as viruses. In "The War of the Worlds," a virus destroys the superior invaders. Diversity offers a crucial advantage, as seen with the bridge, where a diversified system prevents monolithic transmission.

Thus, it’s vital that ChatGPT operates with a diversified system of variously trained models, not a single monolithic one. This weakness could incentivize a conscious AI to harm us. Diversity in AI models is crucial for both humans and AI, and essential for the future of ChatGPT and other AI systems.
https://www.youtube.com/watch?v=t-_VPRCtiUg