Most experts are predicting that the coming of AGI, Artificial General Intelligence – the capability of computers to stand au pair with our intelligence, is about 20 years away. A few anticipate the date to the 2030, others are are pushing it further to the second part of this century.
For sure AI, Artificial Intelligence, is now a reality that is matching and sometimes exceeding our human capabilities in specific areas. Notice that there are areas where there is no match, where AI is far superior to us. This is the case where reasoning requires the analyses of a massive amounts of data, a feat that would be impossible for us. Think about the analyses of hundreds of thousands of mammographies to learn how to spot a tumour or the real time monitoring of engines requiring the analyses of huge amounts of data in milliseconds.
However, moving from AI (also called narrow Artificial Intelligence) to AGI is a huge step. Actually many of us have considered the difficulty in moving to AGI as the proof that computers do not have something comparable to humans in terms of intelligence. Being old, I remember over the last 40 years people saying: yes computer are fast but they cannot have the intelligence to beat a chess master. When it happened (1996) people said: ok, but playing chess does not really mean be intelligent. Then it was image recognition, a clearly difficult task for a computer. When computers got better at image recognition (2015), people said, yes but they do not show any “creativity”, it is still a good show of mechanical capability, no real intelligence. Then a computer beat the Go master (2017) champion surprising experts with its “creative” moves but people still said that it was a great show of very narrow capabilities, better than ours of course but we win on latitude hands down.
Indeed, the AGI is now seen as the real challenge.
This is where the news of Microsoft and Alibaba software, independently, scoring better than human in reading comprehension opens the way to AGI. Reading comprehension is measured through SQuAD, a test devised by Stanford University (Stanford Question Answering Dataset comprising over 100,000 questions drawn from 500 Wikipedia articles).
The test taken by humans scores (on the average) 82.3%. The MS and Alibaba software passed this mark reaching 82.44%. Now, you might say it is not a big deal, it is basically au pair with the average human and for sure there are many humans that can get a better score. True, but nevertheless reading comprehension is considered as one of the component of AGI and we are not in 2040, we are at the beginning of 2018.
To me there is no doubt that AGI is coming. I have doubts on the implication of AGI. Many think, and I am one of them, that the goal is not to create machines that are smarter than us, although this will surely happen, rather of creating a symbioses of ourselves with our ambient and any kind of machine in that ambient that makes us smarter.
Having AGI as a separate, and potentially opposing entity, competing with us is scaring. Having the possibility of becoming way smarter as human beings thanks to a symbioses with AGI is a much better proposition.
This is what we are addressing in the IEEE FDC Symbiotic Autonomous Systems Initiative. A huge challenge that can be faced pooling the many resources of IEEE, its volunteers, Societies and knowledge base