Human-level Artificial Intelligence likely by 2050, Experts say (Patreon)

[This is a transcript with links to references.]

Artificial Intelligence is changing the world rapidly, and many of you have asked me to comment on the threat that AI poses for human civilization. Well thanks for your faith in me, but feel like I’m not the right person to ask. I think you’ll get more out of me telling you what the people think who actually work on the stuff.

And today I have very interesting results from a survey among 3000 AI experts. The most interesting result is that they now think AI is going to change the world even faster than they said just a year ago. Let’s have a look.

This survey was conducted by AI Impacts, that’s a project which has been around since 2018 or so. It's an interdisciplinary group of 6 researchers who, as the name suggests, study the impacts of Artificial Intelligence on society. This recent survey was conducted late last year. For this, they recruited about two thousand 800 AI researchers, selected from among authors of publications on the topic which had appeared in scientific journals.

Collectively, the experts give it a 50 percent chance that we will have High Level Machine Intelligence ,that is comparable to human intelligence or higher, by 2047.

This is down thirteen years from the predictions they made just a year earlier. It’s a similar thing with their predictions for the full automation of labour. They see it coming with 50 percent probability by 21 20 or so, but that’s down almost 50 years from last year’s prediction.

I think this basically means is that AI is developing faster than even most experts thought it would.

About half of them also think that the “intelligence explosion argument” is correct. This intelligence explosion is weaker version of what was formerly known as the technological singularity. That once you have programs that are intelligent enough to build more intelligent programs, intelligence will rapidly increase and go god knows where.

And some of them are quite worried about what this will bring. While most of them think that highly intelligent AIs will be mostly good, 5 percent think that it’ll be extremely bad, possibly leading to human extinction.

Though to me this raises more questions than it answers. First of all, there’s the question if you think that your work will lead to human extinction why are you working on it. And even more concerningly, as I’ve discussed in an earlier episode, about one in 5 people think that human extinction would actually be good, because humans are bad. This then makes me wonder how big the overlap is between the 10 percent wo think that AI would be good, and the 20 percent who think that human extinction would be good.  

Okay, but human extinction is still somewhat in the future, presumably. What’s coming up first is that we’ll be replaced by machines. The survey participants gave their best guesses about when artificial intelligence would be able to replace certain jobs, including the job of AI researchers themselves. They expect it will take something between 30 and 100 years until AIs research themselves.

In this figure, the dots are the median, so what 50 percent think, the lower end of the bar is 25 percent and the upper 75 percent. You can see that the experts say the first jobs to be replaced by AI are everything related to language. That includes reading, transcription, translation, and writing, reaching the level of NYT best−selling fiction by 2040 or earlier.

I guess if I ever want to write my great science fiction series, I better do this soon.

These predictions make sense because language generation is one of the AI tasks that have been most explored so far. The other most obvious advances we’ve seen recently is with visuals and audio. These researchers say that before 2040 we’ll see an AI generated song in the top 40. Personally, I think it’ll happen much sooner, within the next couple of years.

They also think that truck drivers will become obsolete by 2050 and surgeons by 2100.

One thing to note though is that these predictions are not just about artificial intelligence. They mix AI together with advances in robot technology and also sociology, politics, and psychology. For example, the question of whether retail salespeople will be replaced by AI isn’t just about whether it’s possible to do that. It also has a lot to do with whether humans will accept this change. And it’s a strange but true fact that humans are most interested in other humans.

Think about it. Most dogs can easily outrun the fastest humans. But we don’t really care, we want to know what humans can do. Humans are also fascinated by humans who produce photorealistic paintings. Which is incredibly odd given that you could just take a photo isn’t it.

Maybe it doesn’t make sense, but we do as a matter of fact care who did something and how it was done. It’s not just about the result, it’s about the effort. So I think that as long as there are human customers there will also be human salesmen and human creators. Until the AI decide that we’re really just an annoyance and get rid of us entirely.