The creator of artificial intelligence wants to end the AI races and develop features that help human society.
On November 17th, Yoshua Bengio, one of the founding fathers of AI technology, attended the MIT Forum. Yoshua told Will Knight, editor of the Technology Review page about his fears that the original idea was slowly developing in the direction of control.
– What do you think about the idea of an AI race among big nations?
– Yoshua Bengio: I do not like that idea. It is completely wrong. Competition to build AI is good, but they have to be put in science. I think the best thing to do is to build AI technology that benefits as many people as possible.
– Is there a way to promote technological cooperation among countries?
– We have been able to create more learning opportunities for people from developing countries. In Europe, America or Canada, it is very difficult for an African student to obtain a visa. Applying for a visa is like playing a lottery and they have the right to refuse at any time. This is not fair at all.
Under their conditions, research resources are already very limited nowadays even a community of researchers can not be built. We plan to address this issue at the Global AI Conference (ICLR) held in Africa in 2020.
– Do you worry that many technology companies in China are dominating AI?
– Yes. That is why we need to have more democracy in AI research and development. The essence of artificial intelligence research is making a lot of money, gaining power.
The bigger the company, the more income it gets and the more qualified people. I think that is wrong. Even though AI research is still in a democratic state, power is focused on some very dangerous people.
– There has been much debate over the use of AI for military purposes. What’s your opinion?
– I strongly oppose using AI for war purposes.
– Even the use of AI without damaging anything?
– I do not want to stop it. But, integrating AI into combat robots is inhumane. We need to change the culture of using AI, which means changing the law and conventions. A long way to go.
Of course. No one can stop it completely and everyone will again speculate that “no country is another country, especially some countries that are terrorizing”. I emphasize that we have to make them feel guilty about it, but building the defense system is obvious. The biggest difference in this matter is that the defense system targets robots and AI weapons target humans, though both use artificial intelligence.
– Should AI experts work for the military to ensure everything is legal?
– They should be like that if they maintain professional ethics.
– What is your greatest interest in emerging AI studies?
– I think we should focus on the challenges that AI will face rather than be satisfied with the former. I’m not saying that I would skip deep learning. On the other hand, I want to take it as a foundation for development.
But, we have to teach it more and at the same time motivate AI to learn naturally and to explore the world, to build up causal behavior. Letting people think like humans requires a long-term investment and scientists, and no one else has to carry on that task.
– He emphasized the word “cause and effect,” why is it important and why is it so difficult to achieve?
– If we build a causal paradigm of the present, AI can adapt itself to the most unexpected situations. We have the special ability to adjust ourselves to the experience we have. Machines are not so because they do not have a causal pattern.
We can outline but they are still very sketchy. We need machines that can build a causal model. Up to a limit can not be perfect. The reason is that because the perfect causal model does not exist, we always make mistakes. But, we know how to improve ourselves better than many other animals.
We do not yet have an algorithm for that, but if people focus on work and study, we will make significant progress.