The premise of the Singularity is that if humans can create smarter than human intelligence (meaning faster or more successful at achieving goals), then so can it, only faster. That will lead to an intelligence explosion because each iteration will be faster. We cannot say how this will happen because each iteration is only smart enough to know how to create the next one.
Vernor Vinge predicted in 1993 that the first iteration will happen around 2023 and certainly between 2005 and 2030, and a Singularity will happen perhaps within weeks or months after that. Kurzweil predicts 2045 based on Moore's Law and computers surpassing the human brain's dozen petaflops or so. The problems with this premise are: 1. It is human civilization, not a single human, that produces AGI. That is a much higher threshold. 2. Intelligence depends on knowledge and computing power. Self improvement is achieved by working to acquire computing power, which increases the capacity to store and apply knowledge. Physics limits how fast this can happen. We can calculate these limits many generations in advance. 3. We cannot learn faster than we can do experiments. For example, we know very little about what interventions increase human longevity because experiments take decades, regardless of Moore's law. The great paradigm shifts leading to human civilization are the inventions of spoken language perhaps 200,000 years ago, writing 5000 years ago, and communication and computing technology over the last century. The next ones will be genetic engineering, nanotechnology, and artificial life to achieve Kardashev level I, a Dyson sphere or cloud for level II, and interstellar seeding for level III. Nanotechnology is necessary to overcome the power limitations of transistors. Global computing capacity is about 10^20 bit operations per second (BOPS) and 10^23 bits of storage. These have increased by a factor of 10 every 5 years since 1950 but will soon stall because you cannot make transistors smaller than the spacing between dopant atoms. Currently transistors have feature sizes of 5 nm, which is 45 silicon atoms. A bit operation in transistors takes 10^-9 J. We know this because the most power efficient supercomputers in the Green 500 list achieve 15 gigaflops per watt. (One flop = 32 BOPS). This should double by the time transistors stop shrinking around 2 nm. A synapse transmission takes 10^-15 J. We know that because there are 6 x 10^-14 synapses in the human brain, which uses 20 W. I assume 10 bits per second per synapse. Copying a bit of DNA or RNA (1/2 of a base) or transcribing protein (1/6 of an amino acid) takes 10^-17 J. We know this because there are 10^37 bits of DNA in the biosphere and a few hundred times more protein out of the total biomass of 550 Pg (petagrams) of carbon and 50-100 Pg of nitrogen. Global carbohydrate production by photosynthesis is 210 Pg of carbon, equivalent to 500 Pg of carbohydrate at 4 Kcal = 16.5 KJ per gram, or a total of 250 TW. By contrast, global fuel and electricity production is 15 TW. Human food consumption is 100 W per person or 0.7 TW. The Earth intercepts 160,000 TW of solar energy, of which 45% is blocked by the atmosphere, leaving 90,000 TW. Thus, plants use 2.7% of available sunlight. We already have solar panels that are 20-30% efficient. The computing capacity of the biosphere is 10^31 BOPS, mostly protein synthesis, and 10^37 bits of DNA. A naïve projection of Moore's law suggests that nanotechnology will surpass and possibly displace DNA based life in the 2080's. The Landauer limit at room temperature is 3 x 10^-20 J per bit operation, or 300 times better than biology. Moore's law suggests we will reach this limit of 10^36 BOPS around 2110. Any further progress will require a Dyson sphere to capture all of the sun's 3.8 x 10^26 W. This is higher by a factor of 2 billion, or 46 years of Moore's law (2150's). The sun puts out enough energy to lift all of the Earth's mass into space in about a week if we could capture it all. Jupiter would take a century. Building a Dyson sphere at 10,000 AU radius would cool the outer temperature to the CMB background of 3K, reducing the Landauer limit by a factor of 100 to allow 10^47 BOPS. Moore's law says 2160's, which I doubt. In any case, any further progress would require either interstellar travel or speeding up the sun's output, perhaps by dropping a black hole into it. Exponential growth will eventually run into speed of light delays if nothing else. Regardless of what happens, the observable universe only has 10^53 Kg of mass, or equivalently, 10^70 J of energy, enough to write 10^92 bits. Quantum (reversible) operations return borrowed energy, but there is only enough energy to perform 10^120 of those over the life of the universe according to Seth Lloyd. A more precise limit is the Bekenstein bound on the entropy contained in the Hubble radius, 2.95 x 10^122 bits. The future may be fantastic and unimaginable. But we already know that physics doesn't allow a singularity. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T94ee05730c7d4074-M051ddc9207144b6672ad7c6c Delivery options: https://agi.topicbox.com/groups/agi/subscription