Survey: Artificial intelligence developments in 2023 surprise artificial intelligence experts

It’s not just your imagination. A new survey finds that artificial intelligence researchers themselves are alarmed by the breakneck pace of AI development.

A wide-ranging survey of artificial intelligence experts released this week supports the idea that the development of artificial intelligence is indeed accelerating at a dizzying pace – at least from the perspective of experts in the field. It also helps quantify the tech world’s notorious divide between die-hard AI fans and AI “doomsayers,” who purportedly preach caution because they fear some kind of AI doomsday scenario.

Despite the differences, diehards seem to be slightly more numerous, and – if you read between the lines – they seem to be considered the victors.

The paper on the investigation is a pre-publication version from AI Impacts, a San Francisco-based research firm that has received funding from billionaire Facebook co-founder Dustin Moskovitz Funding from Open Philanthropy.

By averaging the survey responses of 2,778 AI researchers who met the authors’ own visibility criteria and comparing them to similar previous surveys, the authors found that, broadly speaking, AI experts feel a sense of acceleration across the board . The authors note that on average, when it comes to 32 different AI-related tasks, “the 50th percentile year in which they become feasible is expected to be 1.0 years earlier,” between 2022 and 2023.

In less technical terms, at some point between the 2022 survey and the 2023 survey, the average AI prediction time was a year ahead. This is a stronger finding than the average expert saying “yes” to the question: “Do you think developments in the world of artificial intelligence are accelerating?” because it shows that experts are actually revising their assumptions about this acceleration year over year. time estimate.

Perhaps the key finding of the study is that respondents’ overall predictions for two key concepts have drastically changed compared to similar predictions in 2022: Advanced Machine Intelligence (HLMI) and Full Automation of Labor (FAOL). In particular, the expected arrival time between 2022 and 2023 has been shortened by 13 years.Meanwhile, FAOL’s forecast reduced by 48 years same period.

This document represents a radical and significant shift in mindset. In a mind-bending year, artificial intelligence experts have come to believe that “for any occupation, machines can be built to perform tasks better and cheaper than human workers” and that’s nearly half the point. A century earlier than the previous year.

Considering how quickly these experts believe these major consequences will come, read their thoughts on whether artificial intelligence should develop faster (the view held by so-called “effective accelerationists”) or slowly (i.e. the AI ​​apocalypse) The viewpoint held by the commentator) is very meaningful. Overtly hardcore doomsayers, or at least those who want artificial intelligence to develop “much slower,” were the smallest group of respondents at 4.8%. Meanwhile, apparent accelerationists – those who react “much faster” – eliminated doomsayers outright at 15.6%.

But the group of respondents who answered this question with “slightly slower speed” actually won the majority, with 29.9% of their answers, followed by “current speed” (26.9%) and “slightly faster” (22.8%). This confusing middle section is made up of three more status quo answers, accounting for 79.6% of all answers.

However, it’s worth taking a closer look at an important distinction noted in the survey: respondents only had expertise in AI, not forecasting in general or AI itself. As a result, they may lack “skills and experience or expertise in non-technical factors that influence the trajectory of artificial intelligence,” the authors write. In fact, this academic warning is worth keeping in mind whenever you read what artificial intelligence experts say about the future in any context.

But these findings are not inconsequential because AI researchers lack psychic abilities. By understanding their subjective beliefs about their fields of expertise, we can learn about what these people, on average, want, fear, and see: They think an automated world driven by artificial intelligence will be more advanced than ever before. Nothing comes faster, and as a group they are mostly on the fence about whether the pace of AI change is good.

But disturbingly, those who want to launch rockets on this already accelerating freight train far outnumber those who want to slam on the brakes.