Academia.eduAcademia.edu

Figure from:When Will AI Exceed Human Performance? Evidence from AI Experts

See full PDFdownloadDownload figure
Figure S2: Aggregate subjective probability of HLMI arrival for two framings of the question. Figure S2: Aggregate subjective probability of HLMI arrival for two framings of the question. The “fixed probabilities” and “fixed years” curves are each an aggregate forecast for HLMI predictions, computed using the same procedure as in Fig. 1. These two framings of questions about HLMI are explained in “Elicitation of Beliefs’ above. The “combined” curve is an average over these two framings and is the curve used in Fig. 1.

Figure 1 S2: Aggregate subjective probability of HLMI arrival for two framings of the question. Figure S2: Aggregate subjective probability of HLMI arrival for two framings of the question. The “fixed probabilities” and “fixed years” curves are each an aggregate forecast for HLMI predictions, computed using the same procedure as in Fig. 1. These two framings of questions about HLMI are explained in “Elicitation of Beliefs’ above. The “combined” curve is an average over these two framings and is the curve used in Fig. 1.

Related Figures (15)

Figure 1: Aggregate subjective probability of ‘high-level machine intelligence’ arrival by future years. Each respondent provided three data points for their forecast and these were fit to the Gamma CDF by least squares to produce the grey CDFs. The “Aggregate Forecast” is the mean distribution over all individual CDFs (also called the “mixture” distribution). The confidence interval was generated by bootstrapping (clustering on respondents) and plotting the 95% interval for estimated probabilities at each year. The LOESS curve is a non-parametric regression on all data points. Each individual respondent estimated the probability of HLMI arriving in future years. Taking the mean over each individual, the aggregate forecast gave a 50% chance of HLMI occurring within 45 years and a 10% chance of it occurring within 9 years. Figure 1 displays the probabilistic predictions for a random subset of individuals, as well as the mean predictions. There is large inter-subject variation: Figure 3 shows that Asian respondents expect HLMI in 30 years, whereas North Americans expect it in 74 years.
Figure 1: Aggregate subjective probability of ‘high-level machine intelligence’ arrival by future years. Each respondent provided three data points for their forecast and these were fit to the Gamma CDF by least squares to produce the grey CDFs. The “Aggregate Forecast” is the mean distribution over all individual CDFs (also called the “mixture” distribution). The confidence interval was generated by bootstrapping (clustering on respondents) and plotting the 95% interval for estimated probabilities at each year. The LOESS curve is a non-parametric regression on all data points. Each individual respondent estimated the probability of HLMI arriving in future years. Taking the mean over each individual, the aggregate forecast gave a 50% chance of HLMI occurring within 45 years and a 10% chance of it occurring within 9 years. Figure 1 displays the probabilistic predictions for a random subset of individuals, as well as the mean predictions. There is large inter-subject variation: Figure 3 shows that Asian respondents expect HLMI in 30 years, whereas North Americans expect it in 74 years.
Figure 2: Timeline of Median Estimates (with 50% intervals) for AI Achieving Human Per- formance. Timelines showing 50% probability intervals for achieving selected AI milestones. Specifically, intervals represent the date range from the 25% to 75% probability of the event occurring, calculated from the mean of individual CDFs as in Fig. 1. Circles denote the 50%-probability year. Each milestone is for AI to achieve or surpass human expert /professional performance (full descriptions in Table S5). Note that these intervals represent the uncertainty of survey respondents, not estimation uncertainty.
Figure 2: Timeline of Median Estimates (with 50% intervals) for AI Achieving Human Per- formance. Timelines showing 50% probability intervals for achieving selected AI milestones. Specifically, intervals represent the date range from the 25% to 75% probability of the event occurring, calculated from the mean of individual CDFs as in Fig. 1. Circles denote the 50%-probability year. Each milestone is for AI to achieve or surpass human expert /professional performance (full descriptions in Table S5). Note that these intervals represent the uncertainty of survey respondents, not estimation uncertainty.
Figure 3: Aggregate Forecast (computed as in Figure 1) for HLMI, grouped by region in which respondent was an undergraduate. Additional regions (Middle East, S. America, Africa, Oceania) had much smaller numbers and are grouped as “Other Regions.”
Figure 3: Aggregate Forecast (computed as in Figure 1) for HLMI, grouped by region in which respondent was an undergraduate. Additional regions (Middle East, S. America, Africa, Oceania) had much smaller numbers and are grouped as “Other Regions.”
Supplementary Figures Figure S1: Aggregate subjective probability of HLMI arrival by demographic group. Each graph curve is an Aggregate Forecasts CDF, computed using the procedure described in Figure 1 and in “Elicitation of Beliefs.” Figure Sla shows aggregate HLMI predictions for the four countries with the most respondents in our survey. Figure Slb shows predictions grouped by quartiles for seniority (measured by time since they started a PhD). Figure Slc shows predictions grouped by quartiles for citation count. “Q4” indicates the top quartile (i.e. the most senior researchers or the researchers with most citations).
Supplementary Figures Figure S1: Aggregate subjective probability of HLMI arrival by demographic group. Each graph curve is an Aggregate Forecasts CDF, computed using the procedure described in Figure 1 and in “Elicitation of Beliefs.” Figure Sla shows aggregate HLMI predictions for the four countries with the most respondents in our survey. Figure Slb shows predictions grouped by quartiles for seniority (measured by time since they started a PhD). Figure Slc shows predictions grouped by quartiles for citation count. “Q4” indicates the top quartile (i.e. the most senior researchers or the researchers with most citations).
Table S1: Median estimate (in years from 2016) for automation of human jobs by region of undergraduate institution
Table S1: Median estimate (in years from 2016) for automation of human jobs by region of undergraduate institution
Table S2: Robust linear regression for individual HLMI predictions
Table S2: Robust linear regression for individual HLMI predictions
Table S3: Demographic differences between respondents and non-respondents
Table S3: Demographic differences between respondents and non-respondents
Three of the questions below concern Stuart Russell’s argument about highly advanced AI. An excerpt of the argument was included in the survey. The full argument can be found here: gs Be ECE ere eee). ee. Ce 2 fe ty nnnare Table S4: Median survey responses for AI progress and safety questions S4: Survey responses on AI progress, intelligence explosions, and AI Safety
Three of the questions below concern Stuart Russell’s argument about highly advanced AI. An excerpt of the argument was included in the survey. The full argument can be found here: gs Be ECE ere eee). ee. Ce 2 fe ty nnnare Table S4: Median survey responses for AI progress and safety questions S4: Survey responses on AI progress, intelligence explosions, and AI Safety
Table S5: Descriptions of AI Milestones
Table S5: Descriptions of AI Milestones