What an old AI researcher thinks after watching the movie "Green Book" - about Racism, Discrimination, and AI (Artificial Intelligence)

(Diary of an Old AI Researcher who is still Programming)

18 March 2019

Bravo, the cinema "Green Book" !!
It is natural that the movie has won the Oscar best picture.

Perhaps, people are moved by the movie not because it is a moving story of the past but because it digs the problems out of our current world.

The racism and the discrimination have never disappeared, and it is a sad fact that AI technologies are said to have potential to amplify the racism and the discrimination.

As an AI researcher, I would like to say that it is not the fault of AI but the fact is that the big data accumulated from the internet have begun to reveal the hidden existence of racism and discrimination driven by the latest statistical methods which are often called AI technologies.

We, the old AI researchers, have continued to study what intelligence is, and we know that intelligence is, of course, far more than only the statistical analysis of data.

The statistical methods developed in the current third AI boom have achieved great progress in filling the gap between data and information.
As a result, unfortunately, we have got the information that the racism and the discrimination are embedded universally in the big data.
However, we know that data and information are only small parts of intelligence.
We, the old AI researchers, have struggled to deal with knowledge and wisdom beyond data and information.

The fact that most of the people who watch the movie "Green Book" are moved shows that people have natural feeling that fighting against the racism and the discrimination is the right way.
If this natural feeling is not well embedded in the big data, we, the AI researchers, should develop AI technologies that can gather this natural feeling.

I dare say that this is not so difficult problem.
I do not think that we, the AI researchers, should directly embed any moral judgement into AI by design, but I believe that we can build AI systems to incorporate the democratic assembly of human knowledge and wisdom.

The AI systems may not behave in the way we have designed, because AI can change its behavior based on results of learning.
If the racism and the discrimination are hidden in big data, what the people should do is to feed continuously our knowledge and wisdom into AI systems to upset the big data.
We can do it by verbalizing our knowledge and wisdom.
AI systems will be able to learn how to make correct decisions by applying mostly the same methods of learning from data to the verbalized information of our knowledge and wisdom.

To sum up, whether AI can be moral or not is not the problem of AI, but it is the problem of our society. We should verbalize our knowledge and wisdom.
The verbalized knowledge and wisdom need not be compactly structured.
The current AI technologies are good at dealing with big data of large dimension.
Hence, we should just accumulate enormous amount of fragments of our knowledge and wisdom and feed them continuously into AI systems.
I believe that AI systems that will be embedded in our society should enhance the democratic aggregation of our knowledge and wisdom.
In other words, we, AI researchers, should build AI systems to solve the problems caused by AI systems, as I wrote in Toward AI-embedded Society where AI is Not Recognized as AI.

The system we have built, which has the knowledge base of ethical discourse, is a first small step in this direction.


© 2019 Koichi Hori