AGI, exponentiality and superintelligence

Yesterday at the IA Conference Brasil event, I had the chance not only to give a talk, where I basically presented my human out of the loop article, but also to talk with professionals who are thinkers in the field of artificial intelligence. A mathematician, a computer scientist, people with PhDs, people with a lot of experience.

And there it’s interesting to see that there are already some points of consensus. Nobody likes to talk about the subject much, not even about this tweet from Jensen, CEO of NVIDIA. But it’s hard not to face that AGI is already here. Specific tasks, defined on the computer, can already be performed well by the machine, just as by a human.

I know it’s not fair to compare the technology of the future with today’s. But if you went back 10 years and, in 2016, looked at someone talking to their phone in fluent Portuguese to carry out operations and searches, you would certainly say that the Turing test had not only been passed, but that artificial intelligence was at a human level. It’s pretty hard to deny that, regardless of the criteria and benchmarks you use.

And talking with a mathematician who was there at the event, he said that even some kind of singularity, where the machine starts to improve itself, even if indirectly, already exists too.

I know these are stretched statements, because they have a marketing tone, a tone of astonishment. Just as much as being very afraid of the machines taking over. But I’d like to put on record that, in my perception, and not only mine, we have already reached a lot of what we used to consider impossible and superhuman.

Adaptability and autonomy are not yet at a point that sounds like the human mind, but something is happening.