{"id":1355,"date":"2025-11-10T12:00:00","date_gmt":"2025-11-10T12:00:00","guid":{"rendered":"https:\/\/forecastingresearch.org\/?post_type=research&p=1355"},"modified":"2026-05-05T11:23:12","modified_gmt":"2026-05-05T11:23:12","slug":"longitudinal-expert-ai-panel-leap-working-paper","status":"publish","type":"research","link":"https:\/\/forecastingresearch.org\/research\/longitudinal-expert-ai-panel-leap-working-paper","title":{"rendered":"The Longitudinal Expert AI Panel: Understanding Expert Views on AI Capabilities, Adoption, and Impact"},"content":{"rendered":"\n

Abstract<\/h2>\n\n\n\n

Public debates about AI revolve around bold claims and counterclaims,\nbut rarely culminate in precise, falsifiable forecasts of AI\ncapabilities, adoption, and impact. The Longitudinal Expert AI Panel\n(LEAP) improves this signal-to-noise ratio by gathering such forecasts,\nmonthly, from a carefully chosen panel of 339 experts spanning industry,\nacademia, and policy. The median expert foresees that by 2030 AI will be\nresponsible for 7% of U.S. electricity usage, assist in 18% of work\nhours in the U.S., and provide daily companionship for 15% of\nadults\u2014roughly 7x, 4x, and 2.5x current levels, respectively. The median\nexpert also gives a 60% chance that AI systems solve or substantially\nassist in solving a Millennium Prize Problem by 2040, which would be a\nmajor achievement in mathematics. There is substantial within-individual\nuncertainty and between-individual disagreement among experts, each\naccounting for roughly half of the total variation in expert forecasts\nacross all questions. Nevertheless, the vast majority of LEAP forecasts\nfall far short of the warnings from AI lab leaders about imminent\nartificial superintelligence. We analyze 1.7 million words of\nparticipant rationales to provide a complementary qualitative overview\nof the key mechanisms underpinning fast and slow forecasts of AI\nprogress.<\/p>\n\n\n\n

\n
View the full PDF report \n \n<\/svg>\n\n \n<\/svg><\/a><\/div>\n<\/div>\n\n\n\n
\n
Acknowledgments<\/summary>\n

We thank Victoria Schmidt, Morgane Bascle, and Jonah Black for\nresearch assistance.<\/p>\n<\/details>\n\n\n\n

Disclaimers<\/summary>\n

The views expressed in this paper do not necessarily represent the\nviews of the Federal Reserve Bank of Chicago or the Federal Reserve\nSystem. This research was funded with support from Open Philanthropy and\nCraig Falls.<\/p>\n<\/details>\n<\/div>\n\n\n\n

Executive Summary<\/h2>\n\n\n\n

Despite the clashing narratives around AI, there is little work systematically mapping the full spectrum of views among experts (computer scientists, economists, technologists) and the general public. What do these groups believe about AI’s future capabilities, adoption, and effects? And why do they believe what they do? The Longitudinal Expert AI Panel (LEAP) fills this gap with monthly surveys tracking the quantified beliefs of experts, historically accurate forecasters (\u201csuperforecasters\u201d),1<\/a><\/sup> and the general public.<\/p>\n\n\n\n

Policymakers and stakeholders routinely consult experts to ground\ntheir decision-making in coherent, informed beliefs, especially when\nfaced with new technologies and high levels of uncertainty. LEAP\nfacilitates this process by cutting through the anecdote and speculation\nthat currently dominates AI discourse. We specifically target prominent\nexperts whom policymakers would be most inclined to consult on the\nprogression of AI capabilities and their impact. Expert participants\ninclude top-cited AI and ML scientists, prominent economists, key\ntechnical staff at frontier AI companies, and influential policy experts\nfrom a broad range of NGOs. Our goal is to provide a clear summary of\nexpert views, analyzed within and across these key groups.<\/p>\n\n\n\n

Since we launched LEAP in June 2025, we have completed three survey waves focused on (1) high-level predictions about AI progress; (2) the application of AI to scientific discovery; and (3) the adoption and social impact of AI. Experts provided thoughtful engagement, spending a median of 44 minutes per survey and writing over 460,000 words of rationales explaining their beliefs, across all surveys and questions.2<\/a><\/sup> In this paper, we share results from these first three waves along with a detailed methodological description of the project as a whole.<\/p>\n\n\n\n

Across the first three waves of LEAP, five patterns stand out:<\/p>\n\n\n\n

    \n
  1. Experts expect sizable near\u2011term societal effects from AI by\n2040.<\/li>\n\n\n\n
  2. Substantial disagreement and uncertainty underlie expert\nforecasts.<\/li>\n\n\n\n
  3. The median expert expects much slower progress than prominent\nleaders of frontier AI labs.<\/li>\n\n\n\n
  4. Experts anticipate faster progress than the public on most\noutcomes.<\/li>\n\n\n\n
  5. Experts and superforecasters mostly agree. Where they disagree,\nexperts tend to expect more AI progress. Also, there are no systematic\ndifferences between the predictions of different types of experts:\ncomputer scientists, economists, industry professionals, and policy\nprofessionals.<\/li>\n<\/ol>\n\n\n\n

    First, experts expect sizable near\u2011term societal effects<\/strong>: by 2030, the median expert predicts that 18%3<\/a><\/sup> of all U.S. work hours will be assisted by generative AI, up from 4.1% in November 2024 (Bick et al. 2025<\/a>);4<\/a><\/sup> AI training and deployment will consume 7% of U.S. electricity;5<\/a><\/sup>,<\/sup>6<\/a><\/sup> autonomous vehicles will provide 20% of U.S. ride-hailing trips, and annual global private investment (as reported by Our World in Data)7<\/a><\/sup> will reach $260 billion, up from the $130 billion reported total for this series in 2024. The median expert predicts that by 2030, 15% of adults will self-report using AI for companionship, emotional support, social interaction, or simulated relationships at least once daily, up from 6% today. By 2040, that number doubles to 30% of adults. Experts expect substantial improvements in the ability of AI systems to complete difficult math questions: the median expert believes that AI systems will achieve performance of 75% on the FrontierMath benchmark8<\/a><\/sup> by 2030; and 23% of experts expect saturation of the benchmark.9<\/a><\/sup> The median expert also believes it is more likely than not (a 60% chance) that AI solves or substantially assists in solving a Millennium Prize Problem by 2040.10<\/a><\/sup> We summarize these expert forecasts in the figure below.<\/p>\n\n\n\n

    \"\"\/
    Figure:<\/strong> Median expert forecasts for various questions. We display the 10th<\/sup>, 25th<\/sup>, 50th<\/sup>, 75th<\/sup>, and 90th<\/sup> percentiles of the median forecasts given by experts at each date. For example, if 25% of experts give a median forecast of $10 billion or less, the 25th<\/sup> percentile series in the graph will lie at $10 billion; these series are not<\/em> confidence intervals. Where available, we include a historical baseline in light blue.<\/figcaption><\/figure>\n\n\n\n

    Even experts with lower forecasts of AI progress, capabilities, and adoption still expect substantial impacts relative to current levels. Recall that, by 2030, the median expert predicts that 18% of all U.S. work hours will be assisted by generative AI, up from 4.1% in November 2024 (Bick et al. 2025<\/a>);11<\/a><\/sup> however, respondents were shown a historical baseline value of 2%, based on an earlier version of the cited paper. We also asked forecasters for their 25th<\/sup> percentile forecast,12<\/a><\/sup> and panelists on this question gave a forecast of 9%, still more than a fourfold increase from the historical baseline level provided to forecasters at the time of the survey. In other words, the median expert gives a 75% chance that at least 9% of work hours will be assisted by generative AI in 2030.<\/p>\n\n\n\n

    Second, substantial disagreement and uncertainty underlie expert forecasts<\/strong>. The top quartile of experts estimate that the majority of revenue from newly approved U.S. drugs in 2040 will be from AI-discovered drugs, but the bottom quartile of experts predicts that less than 10% of new drug sales will be from AI-discovered drugs.13<\/a><\/sup> On the question of whether AI will independently solve or substantially assist in solving a Millennium Prize Problem by 2040, a quarter of experts think it is quite likely (>81% chance), while another quarter of experts believe it is unlikely (<30% chance). On many questions, individual experts also express substantial uncertainty in their own forecasts. We construct a composite measure of the total variation in expert beliefs in the figure below (for two questions), taking into account both the within-forecaster uncertainty and the between-forecaster disagreement. We discuss the methodology in greater detail in Uncertainty and Disagreement<\/u><\/a>, where we describe the underlying assumptions that we rely on to construct these figures. These \u201cpooled distributions\u201d represent the full distribution of outcomes predicted by the average expert. We find that, across all forecasting questions where we allow forecasters to express their uncertainty, within-forecaster uncertainty explains 49% of the total variation in forecasts, while 51% of variation is explained by between-forecaster disagreement.<\/p>\n\n\n\n

    \"\"\/
    Figure:<\/strong> Pooled distributions for expert forecasts on Work Hours Assisted by Generative AI<\/em> (top panels) and FrontierMath<\/em> scores<\/em> (bottom panels). These pooled distributions combine within-expert uncertainty and between-expert disagreement. Densities are normalized to the same peak for comparability. See Uncertainty and Disagreement<\/a> for details.<\/figcaption><\/figure>\n\n\n\n

    Third, the median expert expects much slower progress than prominent leaders of frontier AI labs<\/strong>. These lab leaders predict human-level or superhuman AI by 2026\u20132029, while most of our expert panel rejects these shorter timelines. We ask respondents to forecast whether the average LEAP panelist will say, in 2030, that we are closest to a \u201cslow-,\u201d \u201cmoderate-,\u201d or \u201crapid-\u201d AI progress scenario. We define these scenarios in detail in the appendix (see Appendix E.I. 4. General AI Progress<\/a>). The average expert thinks that 23% of LEAP panelists in 2030 will say the world most closely mirrors a \u201crapid\u201d AI progress scenario that, of our three scenarios, most closely matches some of the lab leaders\u2019 claims. On the other hand, the average expert believes that 28% of panelists will indicate that progress plateaued at close-to-current levels, with fewer improvements in capabilities relative to today (a \u201cslow\u201d AI progress scenario).14<\/a><\/sup><\/p>\n\n\n\n

    Fourth, experts anticipate faster progress than the public<\/strong>15<\/a><\/sup> on most outcomes (e.g., the accuracy AI systems will achieve on the FrontierMath benchmark by 2030: median expert 75% vs. median member of the public 50%, p < 0.001<\/em>; generative AI work assistance by 2030: median expert 18% vs. median member of the public 10%, p < 0.001<\/em>). Experts predict a 32% chance that AI will be at least as impactful as a \u201ctechnology of the millennium\u201d\u2014like the printing press or the Industrial Revolution\u2014whereas the public gives this only a 22% chance. Across forecasts that exhibit a clear relationship with AI capabilities and impacts, a randomly selected expert is 16% more likely than a randomly selected member of the public to predict faster progress than would be expected by random chance.<\/p>\n\n\n\n

    We summarize some of the differences in aggregate forecasts in the\nfigure below.<\/p>\n\n\n\n

    \"\"\/
    Figure:<\/strong> Differences between the expert and public median 50th<\/sup> percentile forecasts for several questions where the unit is a percentage. Points indicate the median of each group\u2019s 50th<\/sup> percentile forecasts. We apply transformations to create valenced forecasts, where values closer to the left indicate slower progress and values to the right indicate faster progress.<\/figcaption><\/figure>\n\n\n\n

    Fifth, highly accurate forecasters (superforecasters) and\nexperts are largely aligned, with superforecasters expecting slightly\nless progress overall<\/strong>. Differences across expert\nsubgroups\u2014those specializing in computer science or economics, those\nworking in industry, and those working in the policy and think tank\nspace\u2014are small and rarely statistically significant. We summarize some\nof the differences in aggregate forecasts in the two figures below. The\nfirst compares experts and superforecasters, while the second groups by\ncategory of expertise.<\/p>\n\n\n\n

    \"\"\/
    Figure:<\/strong> Differences between the expert and superforecaster median 50th<\/sup> percentile forecasts for several questions where the unit is a percentage. Points indicate the median of each group\u2019s 50th<\/sup> percentile forecasts. We apply transformations to create valenced forecasts, where values closer to the left indicate slower progress and values to the right indicate faster progress.<\/figcaption><\/figure>\n\n\n\n
    \"\"\/
    Figure:<\/strong> Expert category median 50th<\/sup> percentile forecasts for several questions where the unit is a percentage. Points indicate the median of 50th<\/sup> percentile forecasts for each category. We apply transformations to create valenced forecasts, where values closer to the left indicate slower progress and values to the right indicate faster progress.<\/figcaption><\/figure>\n\n\n\n

    Introduction<\/h2>\n\n\n\n

    As artificial intelligence (AI) reshapes culture,16<\/a><\/sup> science,17<\/a><\/sup> labor markets,18<\/a><\/sup> and the aggregate economy,19<\/a><\/sup> experts debate its value, risks, and how quickly it will integrate into everyday life. Leaders of AI companies forecast transformative AI systems that cure all diseases,20<\/a><\/sup> replace whole classes of jobs,21<\/a><\/sup> and supercharge GDP growth by the 2030s.22<\/a><\/sup> Skeptics see small gains at best, with AI\u2019s impact amounting to little more than a modest boost in productivity\u2014if anything at all (Acemoglu 2024<\/a>).<\/p>\n\n\n\n

    Despite these clashing narratives, there is little work systematically mapping the full spectrum of views among computer scientists, economists, technologists in the private sector, and the public. What do these groups believe about AI’s future capabilities, adoption, and effects? Why do they believe what they do, and what mechanisms support those beliefs? Prior surveys offer opinions, but rarely comprehensively quantify those opinions, hampering policy guidance and evaluation. We fill this gap with the Longitudinal Expert AI Panel (LEAP), a monthly survey tracking the probabilistic forecasts of experts, historically accurate forecasters (\u201csuperforecasters\u201d),23<\/a><\/sup> and the public.<\/p>\n\n\n\n

    Policymakers and stakeholders routinely consult experts to ground their decision-making in coherent, informed beliefs, especially when faced with new technologies and high levels of uncertainty. LEAP is designed to facilitate this process by cutting through the anecdote and speculation that currently dominates AI discourse. We specifically target prominent experts whom policymakers would be most inclined to consult regarding the progression of AI capabilities and its technological impact. LEAP expert invitees (and participants) include top-cited AI and ML scientists, prominent economists, key technical staff at frontier AI companies, and influential policy experts from a broad range of NGOs. Our goal is to provide a clear summary of expert views, analyzed both across and within these key groups. To address concerns that our respondents might be a biased or selective sample, we reweight our data to account for different response propensities within our expert sampling frame, providing a representative summary of our expert frame\u2019s beliefs.24<\/a><\/sup> The Panel Construction<\/a> section contains more details. This reweighting leaves our headline conclusions unchanged (see Sensitivity of Results to Reweighting<\/a>).<\/p>\n\n\n\n

    We evaluate LEAP respondents\u2019 forecasts against detailed, pre-specified resolution criteria\u2014avoiding the problem of ambiguous resolution criteria that introduce noise in forecasts by permitting both different interpretations of a question by forecasters, and disagreement about how a question is later resolved. In contrast, clear resolution criteria put the debate in common terms and enable policymakers to understand the range of opinion. LEAP\u2019s resolution criteria are also used to measure accuracy\u2014and tie respondents\u2019 compensation to this accuracy\u2014to encourage participants to provide high-quality forecasts. Lastly, we can use this accuracy measure to identify the most accurate forecasters and explore how their views compare to other participants.25<\/a><\/sup><\/p>\n\n\n\n

    LEAP captures not just forecasts, but participants\u2019 rationales: 1.7\nmillion words of detailed explanations across the first three survey\nwaves of detailed explanations. Over 600,000 of these words come from\nour expert and superforecaster samples, with the remainder coming from\npublic respondents. We use this data to identify key sources of\ndisagreement and to analyze why participants express significant\nuncertainty about the future effects of AI.<\/p>\n\n\n\n

    Since we launched LEAP in June 2025, we have completed three survey\nwaves focused on (1) high-level predictions about AI progress; (2) the\napplication of AI to scientific discovery; and (3) widespread adoption\nand societal impact. In this paper, we share results from these first\nthree waves along with a detailed methodological description of the\nproject as a whole.<\/p>\n\n\n\n

    In what follows, we first present aggregate forecasts from experts. Aggregation exploits the \u201cwisdom of the crowd\u201d phenomenon, in which the accuracy of aggregated predictions exceeds the accuracy of a large majority of their constituent parts. This practice is supported by work in numerous fields, such as prediction markets (Bassamboo, Cui, and Moreno 2015<\/a>), political forecasting (Sj\u00f6berg 2009<\/a>, Murr 2011<\/a>), and more (Hueffer et al. 2023<\/a>, Adjodah et al. 2021<\/a>). Aggregated predictions of many individuals tend to be remarkably accurate (Da and Huang 2020<\/a>; Lichtendahl Jr, Grushka-Cockayne, and Pfeifer 2013<\/a>; Surowiecki 2004<\/a>).26<\/a><\/sup> While aggregate forecasts are helpful, it is important to understand the extent of disagreement between experts and where individuals have more and less uncertainty in their forecasts. We discuss our approach to answering these questions below in the Uncertainty and Disagreement<\/u><\/a> section.<\/strong><\/p>\n\n\n\n

    We will continue to field new LEAP surveys every month for at least\nthree years. We will revisit questions from previous waves beginning in\nyear two, in order to assess how expert views are changing, both in\naggregate and within individuals. First, when we report results,\nparticipants can compare their own forecasts to other participants. As\nforecasts resolve, participants will be able to see how the accuracy of\ntheir past forecasts compares to other participants and adjust their\nfuture forecasts accordingly. We will also explore the relationship\nbetween short-term accuracy and long-run forecasts about AI\ncapabilities, effects, and diffusion.<\/p>\n\n\n\n

    In this report, we discuss LEAP in the context of prior AI forecasting surveys (Connection to Prior Work<\/u><\/a>), detail the procedures we use to build our panel (Panel Construction<\/u><\/a>), describe the surveys and forecasting questions (Monthly Surveys and Forecasting Questions<\/a>), share preliminary results (Results<\/u><\/a>), outline plans for future work (Next Steps<\/u><\/a>), and conclude (Conclusion<\/u><\/a>).<\/p>\n\n\n\n

    Connection to Prior Work<\/h2>\n\n\n\n

    While prior work on AI progress has taken a variety of forms, we\nfocus here on studies that, like LEAP, use surveys of the AI expert\ncommunity and the general public to measure the range of opinions and\nexpectations on the progress of AI, its adoption, and the associated\nsocietal impacts. This type of work faces five challenges. First, it is\ndifficult to specify a comprehensive sampling frame, and nonresponse\nbias is challenging to avoid. Both of these hurdles can distort\naggregate estimates. Second, views on AI could change over time. Third,\nambiguous resolution criteria introduce noise in forecasts and limit our\nability to identify the most accurate forecasters, whom policymakers may\nwant to rely on in the future. Fourth, while we might care about\nlong-term outcomes, it is challenging to assess the credibility of\nlong-term forecasts on short-run decision timelines. Fifth, participants\noften devote limited bandwidth to surveys, and, even when thoughtfully\nengaged, might not report their true beliefs on forecasting questions,\ninstead using the survey to express their preferences. While no survey\ncan fully overcome these limitations, we designed LEAP to thoughtfully\nengage with and mitigate each of these five challenges.<\/p>\n\n\n\n

    Challenge\n1. Narrow frames and selective response can distort estimates of AI\nprogress<\/h3>\n\n\n\n

    A survey can fail to represent aggregate views in two ways. First, the sampling frame might fail to capture the population of interest.27<\/a><\/sup> Second, survey respondents might systematically differ from nonrespondents. Both problems affect AI forecasting surveys.<\/p>\n\n\n\n

    Most surveys of AI experts that predate LEAP target participants in top AI\/ML or computer science conference proceedings (Muller and Bostrom 2014<\/a>; Walsh 2017<\/a>; Grace et al. 2018<\/a>; Zhang et al. 2021<\/a>; Stein-Perlman 2022<\/a>; Zhang et al. 2022<\/a>; Grace et al. 2024<\/a>).28<\/a><\/sup> LEAP seeks to broaden the range of opinions as well as the expertise of participants and thus defines four<\/em> target populations of experts: computer scientists researching topics in AI, economists studying the economic impacts of AI, AI industry professionals, and AI policy professionals. LEAP is the first survey to comprehensively measure the beliefs of each of these groups on AI-related questions so that we can compare each group\u2019s forecasts and accuracy in the future. The Sampling<\/u><\/a> section details our sampling procedures.<\/p>\n\n\n\n

    LEAP additionally includes members of the general public29<\/a><\/sup> and forecasters with a demonstrated track record of uniquely high accuracy on forecasting tasks.30<\/a><\/sup><\/p>\n\n\n\n

    Publicly available demographic data on AI experts is often collected and used to stratify results in these studies, but have not historically been used to reweight results to match a target expert population (Grace et al. 2018<\/a>; Zhang et al. 2022<\/a>). Many studies find associations between collected demographics and outlook,31<\/a><\/sup> suggesting that results may be sensitive to nonresponse bias. LEAP directly addresses issues of nonresponse bias by using common survey weighting methods, discussed in greater detail in Reweighting<\/a>.<\/p>\n\n\n\n

    Challenge 2.\nViews on AI can change over time<\/h3>\n\n\n\n

    One-time surveys reflect respondent views at a given time, so they cannot track the evolution of individual or group opinions over time. As a panel survey that will continue for many years,32<\/a><\/sup> LEAP enables this type of tracking over time.<\/p>\n\n\n\n

    Challenge\n3. Ambiguous forecasting questions complicate ex post<\/em>\nevaluation and comparison between forecasters<\/h3>\n\n\n\n

    When forecasting questions lack detailed resolution criteria,\nassessing accuracy and comparing forecasts is fraught. If a question is\nambiguous, forecasters may have different interpretations of how the\nquestion will be resolved, so those with disparate views can each claim\nto have superior accuracy. In contrast, forecasting questions with\nprespecified resolution criteria permit evaluation of forecasting\naccuracy after events resolve, which proves crucial for the subsequent\ntwo challenges.<\/p>\n\n\n\n

    Early surveys of AI experts focused on measuring the range of opinions on AI (Muller and Bostrom 2014<\/a>) or the relationship between those opinions and other forecasts (Walsh 2017<\/a>) rather than defining resolvable forecasting questions. Research that followed set a goal of generating useful, more resolvable, forecasts. Grace et al. (2018<\/a>) ask experts to forecast on AI progress milestones. Stein-Perlman and Grace (2022<\/a>) and Grace et al. (2024<\/a>) use slightly modified versions of the earlier questions. Zhang et al. (2022<\/a>) modify many of the AI milestone forecasts in Grace et al. (2018<\/a>). LEAP builds on these recent efforts and attempts to create clear resolution criteria for all questions, establishing a longitudinal panel of beliefs from the same set of experts to understand how opinions change over time and to better quantify uncertainty and disagreement across a wider set of experts. O\u2019Donovan et al. (2025<\/a>) and McClain et al. (2025<\/a>) present forward-looking questions but do not ask respondents for forecasts; Pew asks respondents their opinions on the impact of AI over the next twenty years but does not generate measurable outcomes (McClain et al. 2025<\/a>). LEAP forecasting questions use precise and specific language and are designed with measurable outcomes; survey respondents are provided with the resolution criteria at the time of the forecast, a feature that enables us to address the remaining two challenges. Monthly Surveys and Forecasting Questions<\/u><\/a> discusses this feature in greater detail.33<\/a><\/sup><\/p>\n\n\n\n

    Challenge\n4. Long-term forecasts are difficult to evaluate in the near-term<\/h3>\n\n\n\n

    When forecasters disagree about the long-run, how can policymakers\nevaluate those forecasts on short-run timelines? One potential path\nforward is to evaluate the short-run accuracy of forecasters, relying on\nthe most accurate forecasters over the short-run to better understand\nlong-run outcomes. Resolving this challenge first requires clear\nresolution criteria, discussed in the preceding challenge. However, it\nalso requires forecasts on both short- and long-run questions.<\/p>\n\n\n\n

    When using fixed-year format questions, earlier work used 10-, 20-, and 50-year time frames for predictions (Muller and Bostrom 2014<\/a>; Grace et al. 2018<\/a>; Zhang et al. 2019<\/a>; Zhang et al. 2022<\/a>; Stein-Perlman and Grace 2022<\/a>; Grace et al. 2024<\/a>). LEAP includes questions with a wide array of resolution dates, including both near- (as early as the end of 2025) and long-term resolution, allowing us to explore whether short-term accuracy correlates with longer-run beliefs.34<\/a><\/sup><\/p>\n\n\n\n

    Challenge\n5. Forecasting is time-intensive and cognitively taxing, and\ntruthfulness is rarely incentivized<\/h3>\n\n\n\n

    Expert and public participants have limited time to complete surveys, and long surveys risk dissuading potential participants from completion. Three features of LEAP counteract this challenge. First, we provide historical baselines and relevant background information for each question (see Appendix E.I. Survey Questions: Wave 1<\/a> for representative examples). Second, the survey instrument contains interactive interfaces\u2014which integrate historical baseline data where available\u2014to facilitate the forecasting process (see Appendix B.V. Survey instrument<\/a>). Third, much like other surveys, LEAP compensates forecasters for their participation. In the first three waves, the median expert respondent spent 44 minutes on each survey, while the median member of the public and superforecaster spent 29 and 90 minutes, respectively. In contrast, the American Trends Panel (ATP) from Pew Research, a popular public opinion survey, targets ten to fifteen minutes per survey (Pew Research Center 2024<\/a>).<\/p>\n\n\n\n

    Thoughtful engagement need not translate into forecasters reporting their true beliefs. If forecasters\u2019 rewards are not positively related to the quality of their forecasts, they lack the incentive to provide accurate forecasts. Much past work relies on traditional participation payments that are independent of forecasting accuracy (Grace et al. 2018<\/a>; Grace et al. 2024<\/a>). Surveys of the general public (Zhang 2019<\/a>; McClain et al. 2025<\/a>) tend to provide similar recruitment incentives. In contrast, LEAP\u2019s detailed resolution criteria on timely forecasting questions allow us to incentivize participants to provide accurate forecasts by tying rewards to proper scoring rules (Brier 1950<\/a>; Jose and Winkler 2009<\/a>; see Appendix B.III. Scoring<\/a>). Past work demonstrates that proper scoring rules yield more accurate forecasts than unincentivized forecasts (Karger et al. 2021<\/a>).<\/p>\n\n\n\n

    Panel Construction<\/h2>\n\n\n\n

    We discuss in this section our procedures for sampling and\nreweighting, as well as summarize the characteristics of our panel.<\/p>\n\n\n\n

    Sampling<\/h3>\n\n\n\n

    We target prominent experts whom policymakers, business and nonprofit\nleaders, and other stakeholders would be most inclined to consult\nregarding the progression of AI capabilities and its technological\nimpact. Our complete panel consists of experts, forecasters with a\ndemonstrated track record of high, differential accuracy on forecasting\ntasks (\u201csuperforecasters\u201d), and members of the general public.<\/p>\n\n\n\n

    Expert Sampling<\/h4>\n\n\n\n

    We target four expert communities. First, we include computer scientists researching topics in AI by including top-cited authors, stratified by age, and the authors of the top-rated papers at leading AI and ML conferences. Second, we identify leading economists, both across fields and within the subfield of economics focused on the economic effects of AI and new technology. We include top-cited authors of papers on AI and technology, members of the U.S. Economic Experts Panel (Clark Center 2025<\/a>), and attendees of economics conferences on AI. Third, we include industry professionals, identified via their contributions to frontier models or employment at AI-related companies with extensive fundraising. Fourth, we identify institutions leading the discussion on AI development, policy, and impacts and invite research staff.<\/p>\n\n\n\n

    We sample from two other sources and sort them into one of the four communities above. First, we invited the honorees from TIME\u2019s 100 Most Influential People in AI in 2023 and 2024 (Barker Bonomo and Javed 2024<\/a>). Second, we allowed invited respondents to recommend other qualified candidates for the survey, yielding 172 additional invitees. In order to filter this group, we require that an individual:<\/p>\n\n\n\n