Lasse Schultebraucks

How long before superintelligence?

22 Jan 2018

Superintelligence is an AI which can outperform humans in every field. There are already AIs like AlphaGo (Zero) which had defeated the world best Go player, so AlphaGo seems like a smart AI, which in terms of a superintelligent AI it is not. Today there are only AIs which are at the very beginning. There are already AIs which are production ready and help experts in different branches like medicine, but there is not even close to an AI which can substitute an expert like a doctor. And occupation is not everything. Human intelligence is spread over many disciplines. Actually the term intelligence is not quite defined. In my opinion intelligence in terms of human intelligence can be measured by different aspects like problem solving and creativity, but also emotional intelligence.

Anyway, there attempts to reconstruct human intelligence by so called neural networks aka a deep learning. Computer Science and Neuroscience are working here together to solve human intelligence. Many researches are sure that if human intelligence is solved it is not far to superintelligence. The reason for this is the singularity aka a intelligence explosion concept. An AI on human level can build a better AI by itself and after many iterations there will be a superintelligence. So the task seems quite easy, we just have to build an AI on human level by e.g. reconstructing the brain with e.g. software.

So will there be superintelligence soon? I hope so. I wish that there will be superintelligence, because next to the risks of a superintelligent AI there would be many chances, like huge progress in science which would lead to more automated work which would lead to an overall wealthier population. I know, that sounds very utopian, but I am general a very optimistic person.

Nevertheless I do not think there will be superintelligence any time soon. There are still many questions left. Can we reconstruct intelligence by understanding how our brain works? Is this the right way to reach superintelligence? Or are there other paths which we do not know yet which led to superintelligence? Many essential question are still not answered. And the fact that todays AIs are still expert systems which can only learn about (more or less) one discipline but can not do anything else makes me skeptical. Therefore I do not see superintelligence in the next 50 years. There will be probably a next AI winter before we can take the first step in the direction of superintelligence. Because todays technology is not ready for superintelligence. There has to been made major breakthroughs in either neuroscience or computer science before we can go into the direction of a superintelligent AI.

Tweet me @LSchultebraucks if you like this post.