Superhuman Augmentation vs Artificial General Intelligence (AGI) : 無料・フリー素材/写真
Superhuman Augmentation vs Artificial General Intelligence (AGI) / jurvetson
| ライセンス | クリエイティブ・コモンズ 表示 2.1 |
|---|---|
| 説明 | Which will come first? Green vs. Grey, as they say. Thanks to Ariel Poler for hosting a SF Salon on the subject with Erik Torenberg of Village Global and Silicon Foundry.I think we will build a superhuman AGI before we understand our own brain well enough to radically improve it or upload it to a silicon substrate. The complex creations of iterative algorithms (like evolution and deep learning) are inherently inscrutable. It is easier to push evolution forward than to reverse engineer the products of evolution.We are in the middle of a sea change in how the vanguard of engineering will be done. Building complex systems that exceed human understanding is more like parenting than programming. The locus of learning shifts from end products to the process of their creation. An ever-growing percentage of software will be grown and an ever-growing percentage of compute will run on infrastructure that resembles the brain (massively parallel, fine grained architectures with in-memory compute and a growing focus on the memory and interconnect elements). This is the path to AGI, IMHO. I’ve been working with a neural plasticity company for 14 years now (Posit Science). One of my concerns with uploading is the extreme plasticity of the sensory cortex and the recruitment of neighboring regions in the face of external changes (like phantom limb pain in amputees). Cut and paste of brain state to a foreign substrate may require a deep understanding of the analog domain, where structural topology and functional spike train variation is immense (there are over 300 types of neurons in neocortex that are structurally and electrically different. And each neuron has ~200 ion channels from a pool of 20-40 variations). Furthermore, our mostly 2D silicon substrates lack the interconnect density for a direct map of the synaptic fan-out of the brain. Without a deep understanding of what elements can be ignored or abstracted, a simulation of brain function explodes in combinatorial complexity.Going back a decade, in talks about AI futures, I was fond of advising to “augment early and often.” I worry that people want to believe in extreme augmentation and uploading, not because it is likely, but because it offers a mental model for “humanity” maintaining the mantle of supremacy, perpetually perched at the pinnacle of evolution. The idea that evolution will eventually progress way beyond us is hard to internalize. We seek transcendence, as the antidote for obsolescence. I’ll be brainstorming more about storming the brain this evening at a follow up salon.My 2006 musings on these topics. |
| 撮影日 | 2017-12-14 16:04:55 |
| 撮影者 | jurvetson , Los Altos, USA |
| タグ | |
| 撮影地 |