Bill Gates Says AI Will Be As Dangerous as Nukes

Tuco_sl said:

Yeah, this is a huge issue for robotics. Existing chemistry using zinc/manganese dioxide, nickel-metal hydride or lithium-ion will probably never get us there. If you look at the pace of technology in the last 20 years nearly everything is making exponential gains except for batteries. We hear of new super-batteries now and then but nothing dramatic has happened in energy storage.

However, AI constructs that exist purely on the internet could be potentially very dangerous. I don't know enough about network security or distributed networks to know what would happen if a very capable and destructive AI virus were to leak onto the web and became resistant to removal, blockage or quarantine. A network of AIs all sitting in sleeper cells on the deep web until a new security vulnerability was exposed that they could exploit would be really annoying.

Click to expand...

THIS is the scenario that people like Bill Gates is talking about. You want to read some good fiction on it way before its time? Read some William Gibson. That man is a visionary.

I think a_skeleton_03 is trying to say we first must clone a soul to insert into a robit before it can have AI.

In all seriousness, there isn't much computing mass ina human brain, it's just concurrent connections and firings there are which allows us (and other animals) to have intelligence.

I would because trying to seperate out the mass if connective tissue, vasculature and whatnot would be a pain. But ya a brain is about 3 pounds though.

Edit since you edit

Your brain is always computing. I was just saying that from a "hardware" perspective, it is the connectivity and simultaneous functioning that makes it so amazing. From the interactions between individual neurons to the interconnectivity of different highly specialized regions.

Also since you mentioned the cerebellum and used "low-level", it is one of the most fascinating brain structures. One it accounts for about half of all your brain's neurons. It also has some of the most morphologically beautiful and highly specialized neurons (Purkinje cells) with many dendritic extensions (thousands per neuron) and there are tens of millions of these in the cerebellum (i think, it's been a while since I've studied the cerebellum).

Why mention all of this? It seems to me that our current technology just isn't capable of producing intelligence based on the structure and how the computation occurs. Hell, the small neuronal network found in drosophila is capable of amazing computation. I know at some point the ability of a fly to track an object on a complex environment was looked at for missle tracking capabilities for missle defense systems.

Malakriss_sl said:

The brain runs on 12 watts and has a lot of gaps in performance, but it's meant to function for the 99% routine and does it wonderfully. Once you get outside of that and start learning what the limitations are in perception, memory, left/right brain cooperation, etc you can see how much powerful computers are in the hardware department. The issue is simply efficiency and software, or whatever the AI equivalent algorithms will be in the future.

Click to expand...

Disagree entirely. Computers are more powerful for just straight calculation, sure, but they are severely limited elsewhere and isn't entirely compatible with learning, at least how it is understood to occur neurologically.

Tuco is right about the physical adaptations that a neuron or even organelles/epigenetics of a neuron can undergo overtime too. Neuron dendrites grow and adapt, receptors lose/gain sensitivity, new neurons are created and epigenetics affect gene expression based in environmental stimuli. Also, a synaptic connection isnt just on or off, the degree which it is activated can vary leading to different down stream affects. I just dont see a computer or any software doing any of this any time soon with how I computation is currently handled.