Estimates of the expected utility gain of AI Safety Research
When thinking about AI risk, I often wonder how materially impactful each hour of my time is, and I think that this may be useful for other people to know as well, so I spent a couple of hours making a couple of estimates. I basically expect that a tonne of people have put a bunch more time into this than me, but this is nice to have as a rough sketch to point people to. I'm going to make 3 estimates: an underestimate, my best-guess estimate and (what I think is) an overestimate. Starting facts [1] : Currently 8.3 Billion people on planet earth Current median age: 31.1 years Current life expectancy: 73.8 years I am going to commit statistical murder and assume this means that everyone on the planet lives ~42.7 years from this point onwards. Underestimate: 40 years of life left/person Median: 42.7 years + ~15 years' increase in life expectancy (20 years' growth in the past 60 years) = about 60 years of life left Overestimate: Everyone gets life extension and lives to heat death of universe: 10^100 years Since the population is growing, we should take that into account: Underestimate: We only care about the lives of people currently alive Median: We keep growing at current ~1% growth rate per year Overestimate: Population growth of 2% per year until the heat death of the universe Given these parameters, we can figure out the total expected years of life we care about for each scenario: Under:
When considering the impact of AI safety research, it's common to wonder how much each hour of time contributes to mitigating potential risks. To address this, an individual spent a few hours estimating the expected utility gain of such research, providing three scenarios: an underestimate, a best-guess estimate, and an overestimate. These estimates take into account the current population, life expectancy, and growth rates.
Starting with the current facts, there are 8.3 billion people on Earth, with a median age of 31.1 years and a life expectancy of 73.8 years. The author assumes that everyone will live approximately 42.7 years from now, based on these averages.
For the underestimate, the author considers 40 years of life left per person. This scenario assumes a static population, focusing only on the lives of people currently alive. The total expected years of life in this case would be 40 years multiplied by 8.3 billion, resulting in 332 billion years.
The median estimate factors in a growing population. It assumes a 1% annual growth rate, leading to an additional population that will also benefit from increased life expectancy. The total expected years of life in this scenario would be the product of the current population, the additional population, and the extended life span, which includes a linear approximation of increased life expectancy.
The overestimate takes an extreme approach, assuming that everyone achieves life extension and lives until the heat death of the universe, estimated at 10^100 years. However, this scenario quickly becomes intractable due to the exponential growth of the population. The author suggests skipping this overestimate, as it leads to calculations beyond practical limits.
In the underestimate, the research aims to achieve a 1% chance of a 1% decrease in the final risk for the entire field within 20 years. Assuming an imminent extinction event 30 years from now, the utility gain of this research would be significant.
These estimates serve as a rough framework to gauge the potential impact of AI safety research. While other experts may have conducted more in-depth analyses, these calculations provide a starting point for those interested in understanding the material impact of their efforts. The three scenarios offer a range of possibilities, from conservative to extreme, allowing individuals to assess the potential value of their contributions to the field.
