Artificial intelligence that surpasses our own intelligence looks like stuff from science fiction books or movies. What do experts in the field of AI research think of such scenarios? Do they dismiss these ideas as fanciful, or do they take these perspectives seriously?
A human-level AI would be a machine, or a network of machines, capable of performing the same range of tasks that we humans are capable of. It would be a machine that is “capable of learning to do anything a human can do,” as Norvig and Russell put it in their AI handbook.1
He would be able to choose actions that allow the machine to achieve its goals, and then perform those actions. He would be able to do the job of a translator, a doctor, an illustrator, a teacher, a therapist, a driver, or the job of an investor.
In recent years, several research teams have contacted AI experts and asked them about their expectations for the future of artificial intelligence. These expert surveys are one of the insights we can draw on to get an idea of what the future of AI might look like.
The graph shows the responses of 352 experts. This is taken from the most recent study by Katja Grace and colleagues, conducted in the summer of 2022.2
Experts were asked if they thought there was a 50% chance that human-level AI exists.3 Human-level AI has been defined as autonomous machines that can perform every task better and more cheaply than human workers. Further information on the study can be found in the box at the end of the text on this page.4
Each vertical line in this table represents an expert’s response. The fact that there are such large differences in responses clearly shows that experts disagree on how long it will take before such a system can be developed. A few believe that this level of technology will never be developed. Some think it is possible, but it will take a long time. And many believe that it will be developed in the coming decades.
As the annotations point out, half of the experts gave a date before 2061, and 90% gave a date within the next 100 years.
Other surveys by AI experts come to similar conclusions. In the following visualization, I’ve added timelines from two previous surveys conducted in 2018 and 2019. It’s useful to look at different surveys because they differ in how they asked the question and how they defined AI at the human level. You will find more details on these studies at the end of this text.
In all three surveys, we see broad disagreement among the experts and they also express great uncertainty about their own individual predictions.5
What should we do with the timelines of AI experts?
Expert surveys are information to consider when thinking about the future of AI, but we should not overestimate the results of these surveys. Experts on a particular technology are not necessarily experts in making predictions about the future of that technology.
Experts in many fields do not have a good track record of forecasting their own field, as researchers such as Barbara Mellers, Phil Tetlock, and others have shown.6 The history of the flight includes a vivid example of such a failure. Wilbur Wright is quoted as saying, “I confess that in 1901 I told my brother Orville that the man would not fly for 50 years.” Two years later, “the man” was not only flying, but it was these same men who achieved the feat.7
In addition, these studies often find significant “framing effects”, where two logically identical questions get answered very differently depending on the exact wording of the questions.8
What I take away from these surveys, however, is that the majority of AI experts take the prospect of very powerful AI technology seriously. It is not true that AI researchers view extremely powerful AI as mere fantasy.
The vast majority believe that in the coming decades there is an equal chance that we will see AI technology that will have a transformative impact on our world. While some have long lead times, many believe that we may have very little time before these technologies arrive. Across all three surveys, more than half believe there is a 50% chance that human-level AI will be developed before some point in the 2060s, a time well into the lives of young people today. .
Metaculus Community Predictions
In the large visualization on AI timelines below, I’ve included forecasts from the Metaculus forecasting community.
Forecasters on the online platform Metaculus.com are not AI experts but people who devote their energy to making good predictions. Forecasting research has shown that groups of people can assign surprisingly accurate probabilities to future events when given the right prompts and feedback.9 To receive such feedback, the Metaculus online community tracks how they perform in their forecasts.
What does this group of forecasters expect for the future of AI?
As of this writing, November 2022, forecasters estimate that there is a 50/50 chance that “artificial general intelligence” will be “designed, tested and publicly announced” by 2040, in less than 20 years now.
At their page on that specific issue, you can find the precise definition of the AI system in question, how their forecast timeline has changed, and individual forecasters’ arguments about how they arrived at their forecasts.ten
Metaculus community timelines have become much shorter recently. Expected timelines shortened by about a decade in the spring of 2022, when several impressive AI breakthroughs happened faster than many expected.11
Forecasting by Ajeya Cotra
The latest forecast presented comes from research by Ajeya Cotra, who works for the non-profit organization Open Philanthropy.12 In 2020, she published a detailed and influential study asking when the world will see transformative AI. His timeline is not based on surveys, but on the study of long-term trends in the computation used to train AI systems. I present and discuss long-term trends in training computing in this companion article.
Cotra has estimated that there is a 50% chance that a transformative AI system will become possible and affordable by 2050. This is its central estimate in its “mid-case scenario”. Cotra points out that there are significant uncertainties around this midpoint scenario and has also explored two other, more extreme scenarios. The timelines for these two scenarios (his “most plausible aggressive” scenario and his “plausible most conservative” scenario) are also displayed in the visualization. The period from 2040 to 2090 in Cotra’s “plausible” forecast shows that it thinks the uncertainty is large.
The visualization also shows that Cotra has updated its forecast two years after its initial publication. In 2022, Cotra released an update in which it shortened its median schedule by ten years.13
It is important to note that the definitions of the AI systems in question differ greatly from study to study. For example, the system Cotra talks about would have a much more transformative impact on the world than the system Metaculus forecasters focus on. More details can be found in the Annex and in the respective studies.
What can we learn from forecasts?
The visualization shows forecasts from 1,128 people – 812 individual AI experts, aggregated estimates from 315 forecasters from the Metaculus platform, and detailed study results from Ajeya Cotra.
There are two big takeaways from these AI timeline predictions:
- There is no consensus and the uncertainty is great. There is huge disagreement among experts on when human-level AI will be developed. Some think it’s decades away, while others think it’s likely that such systems will be developed in the next few years or months. It’s not just disagreements Between experts; individual experts also point to the great uncertainty surrounding their own individual estimate. As always when uncertainty is high, it is important to emphasize that it goes both ways. It may be a very long time before we see human-level AI, but that also means we may have little time to prepare.
- At the same time, there is broad agreement in the overall picture. Many experts’ timelines are shorter than a century, and many have timelines much shorter than that. The majority of those who study this question believe that there is a 50% chance that transformative AI systems will be developed in the next 50 years. In this case, it would likely be the biggest transformation in our children’s lives, or even in our own lives.
Public discourse and decision-making in major institutions have not caught up with these perspectives. In discussions about the future of our world – from the future of our climate, to the future of our economies, to the future of our political institutions – the prospect of transformative AI is rarely at the heart of the conversation. Often it is not mentioned at all, not even in a footnote.
We seem to be in a situation where most people barely think about the future of artificial intelligencewhile the few who devote their attention to it find it plausible that one of the greatest transformations in human history is likely to occur in our lifetime.
Thanks : I would like to thank my colleagues Natasha Ahuja, Daniel Bachler, Bastian Herre, Edouard Mathieu, Esteban Ortiz-Ospina and Hannah Ritchie for their valuable comments on drafts of this essay.
And I would like to thank my colleague Charlie Giattino who calculated the timelines for the individual experts based on the data from the three survey studies and supported the work on this trial. Charlie is also one of the authors of the study cited by Zhang et al. on the timelines of AI experts.