AI isn’t just rapidly becoming smarter, but it can also transmit its smartness to other AI systems much faster than humans.
This observation comes from none other than Nobel Prize winner Geoffrey Hinton, a renowned computer scientist often referred to as the “Godfather of AI.” A key aspect of his concern revolves around the sheer speed at which AI systems can share information, dwarfing human capabilities by an almost unimaginable factor. He points out the fundamental difference between how humans and AI transmit knowledge, a difference that underlies his fear.
“Now, you and I, when I want to get some knowledge from my head into your head, I can’t just take the strengths of the connections between neurons and average them with the strength of the connections between your neurons. Our neurons are different; we’re analog, and we just have very different brains,” Hinton said.
This biological constraint forces humans to rely on indirect methods of knowledge transfer. “So the only way I have of getting knowledge to you,” he continues, “is I do some actions, and if you trust me, you try and change the connection strengths in your brain so that you might do the same thing.”
This process, Hinton argues, is incredibly inefficient. “If you ask, well, how fast is that?” he posits. “Well, if I give you a sentence, it’s only a few hundred bits of information, at most. So, it’s very slow. We communicate just a few bits per second.”
This limitation stands in stark contrast to the capabilities of current AI systems. “These large language models running on digital systems,” Hinton reveals, “can communicate trillions of bits a second.”
This difference in bandwidth has profound implications. “So they’re billions of times better than us at sharing information,” Hinton concludes. “That’s what got me scared.”
Hinton’s concern highlights a crucial difference between human and artificial intelligence: scalability and speed of knowledge dissemination. While humans are bound by the limitations of biological communication, AI can share updates and learnings at an astonishing rate. Imagine a scenario where a single AI develops a harmful capability. This capability could then be rapidly disseminated across the entire network of AI systems, creating a widespread and difficult-to-contain problem. Furthermore, this speed allows AI to evolve and adapt at a pace far exceeding human comprehension, potentially leading to unforeseen consequences.
Humans, on the other hand, are extremely slow at transmitting information. After a new human is born, it can’t walk or talk, and it takes several years before it can even communicate and operate with other humans. It then takes this new human decades of school, college and employment to operate at a level similar to other humans.
AIs, though, can spawn new AIs instantly by simply copying the parameters of their AI models. The recent exponential growth in AI capabilities, coupled with this efficient knowledge sharing, paints a picture of a future where humans may struggle to keep pace with, let alone control, the rapid evolution of artificial intelligence. This is not mere science fiction; it’s a very real and present concern voiced by one of the leading pioneers in the field.