On Sat, Sep 13, 2014 at 9:39 AM, John Rose via AGI <[email protected]> wrote: >> Of course not. There are two important problems for AGI to solve. >> 1. People don't want to work. >> 2. People don't want to die. > > I agree these are priorities. I also think that the concept of soul is > related to both.
To the second, anyway. > First off I'll say that I really respect guys like Stuart Hameroff that have > the cajones enough to talk about it scientifically and philosophically, in > his way, pertaining to quantum consciousness. Hameroff has zero evidence to support his theories. But it probably doesn't matter. Billions of people believe with zero evidence that they will go to heaven. If I can convince you with some quantum mumbo jumbo that your soul will be transferred to this robot after I slice up your brain, then that's good enough for me. I'm really only interested in convincing the people that know you. > So, for you just to say that .. well there's no such thing as soul. What do I > gain from trying to explain? I predict with a high degree of certainty it to > be a waste of energy :) Right. We don't need pointless philosophical distractions. AGI is an engineering problem. We start with a cost/benefit analysis. First, the benefits: 1. The value of automating human labor is world GDP divided by market interest rates, about USD $1 quadrillion. 2. The value of all human lives is world GDP times life expectancy, about $5 quadrillion. And these values will increase as the economy, population, and life expectancy all grow. A solution will require something with the computing power and complexity of 7 billion human brains and bodies. There are three components: hardware, software, and training. 1. Hardware. A human brain performs 10^16 operations per second on 10^14 synapses. The body performs 10^17 DNA base copy operations and 10^19 amino acid assembly operations per second on 10^23 DNA bases. There are 10^10 humans. Current global computing capacity is 10^20 operations per second on 10^22 bits of storage. If these increase by a factor of 10 every 5 years as Moore's Law predicts, then we will be able to model all human brains in 2045 (constrained by CPU) and all human bodies in 2070 (constrained by memory). 2. Software. The human genome has the same information content as 300 million lines of code [1] or $30 billion at $100 per line. This cost is not a significant factor. 3. Training. Human long memory capacity for words, pictures, and music is 10^9 bits according to Landauer [2]. Assume 99% of human knowledge is shared or written [3], leaving 10^7 bits per person (10^17 total) that must be extracted through speech or writing at 2 bits per second at a global average wage rate of $5 per hour. Cost is on the order of $100 trillion and can be expected to increase as wages rise, but could be mitigated by decreasing costs of surveillance. There are engineering tradeoffs such as spending more on software to optimize for slow hardware. The software-training split assumes the cost of coding an equivalent baby. This is probably optimal because writing one bit of code costs 1000 times as much as communicating 1 bit of natural language. References. [1] The cost of AI, appendix A. https://docs.google.com/document/d/1Z0kr3XDoM6cr5TgHH0GXQTjyikr7WpCkpWFn9IglW3o [2] Landauer, http://www.cs.colorado.edu/~mozer/Teaching/syllabi/7782/readings/Landauer1986.pdf [3] The U.S. Labor Dept. estimates it costs on average $15,000 (1% of lifetime earnings) to replace an employee. -- -- Matt Mahoney, [email protected] ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
