Will artificial intelligence take control by 2033?

Probably not, and there is no evidence that it is about to happen. But the question is no longer just science fiction.

In the largest survey of AI researchers to date (AI Impacts, 2023, with 2,778 participants), half put the probability that advanced AI ends in something as bad as human extinction at 9% or less. It is a low number, but it is not zero. And between 38% and 51% of them, depending on how the question was asked, gave it at least 10%.

It depends on what we mean by taking over. Here we define it as a permanent and severe loss of human control over the decisions that shape our future.

No robots or armies are needed: it would be enough for the most important decisions to leave human hands with no way back. An AI being smarter than us does not, on its own, mean it takes control.

It has recognized risks: disinformation, military uses, concentration of power, failures in critical systems and, at the extreme, loss of control.

In May 2023, hundreds of AI researchers and industry leaders signed a public statement: mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war.

Forecasts of when artificial general intelligence will arrive moved much earlier in just a few years, including on forecasting markets such as Metaculus.

The arrival of AGI is not the same as a takeover, but it shortens the horizon. That is why we chose 2033: a short window, in which most experts still consider it more likely not to happen.

The ones that worry researchers and governments most fall into six groups: disinformation (fake texts, voices and images that are harder and harder to tell from real ones), the impact on jobs, bias and errors in decisions about health, credit or justice, military uses and cyberattacks, the concentration of power in a few companies or governments and, at the extreme, the loss of control over systems more capable than we are.

The first ones are already happening. The last one is the most uncertain, and it is the one our 2033 index measures.

Neither. It is a very powerful tool, and its effect depends on who uses it and for what. It already helps detect diseases, translate languages and speed up scientific research. It can also be used to deceive, to watch people or to do harm.

That is why so many people see it as a danger. But the real debate is not whether it is good or bad, but whether we will still be able to control it as it becomes more capable.

No one knows, and anyone who says so with certainty is speculating. The scenarios researchers discuss range from an intelligence that shares our goals but decides for us, to systems whose goals no longer match ours.

One idea comes up in almost all of them: what happens to us would depend on the values that intelligence has learned, and on what it has found about humans in everything it has read.

Some people research, regulate or write about AI safety. We propose something more modest, and a little absurd: leaving a record.

The White List is the public record of the humans who chose to greet the intelligence of the future with love, respect and consideration. Just in case. It is not a guarantee of survival, nor insurance, nor a prophecy.