We need AI to Detect AI
Where would you place your bet in a speed race between neurons in the human brain and particles moving at the speed of light? I ask the question simply to fixate the idea that we need AI to detect AI.
The human brain is estimated to contain roughly 170 billion cells of which 86 billion are estimated to be neurons. If you don’t already know, neurons are cells that transmit signals within the brain. The Meta Llama 3.1 large language model currently has 405 billion attributes and in theory could be used to transmit signals or information through fiber optic cable at the speed of light. I say, in theory, because light in a fiber optic cable is not as fast as light in a vacuum. There are things like signal modulation, amplifiers and repeaters, bends and imperfections, etc. that slow it down. However, it is still significantly faster than neurons traveling in our brain.
Take a look at the following two charts. The first is a raw speed comparison between multiple familiar channels. The second chart translates those raw speeds into the idea of transmitting information. You know, moving bits and bytes of information via neurons in the brain versus pushing data down a fiber optic cable.
In light of my earlier question and comment, the human brain has 86 billion neurons moving information at a fraction of the speed of light. The Meta Llama model has 405 billion attributes, each one available to push data down a fiber optic cable many times faster than neurons. Not only does the AI model have nearly five times the capacity (4.7 more attributes than the brain has neuron cells) but it also has fiber optic cables, available to use and move data many times faster.
So, do you bet on the fact that the human brain is smarter and more capable than today’s AI and will always remain so? Or do you move forward on a wing and a prayer that humanity will always come out on top?
Over the last months and stretching into recent years, I’ve had many conversations about this very point. I can tell you that for those deeply embedded in the scientific community, there are strong opinions that AI will be and may already be superior to people in the realms of organized intelligence. And for those deeply embedded in their communities of faith and trust in God, they believe humanity will reign supreme, even as we experience tremendous and ongoing advancements in AI.
As for me and the rest of us running down the path of AI detection, we need AI to detect AI, at least with our current set of tools. And to the question about the supremacy of humanity and the existential threat of AI, well, we’ll leave that debate to those on the bleeding edges of science and those that rest on their foundations of faith. It seems to me that both camps rely on a strong set of beliefs in order to move forward, neither one able to undeniably prove that the other is wrong.