I am not speaking about biasing in an electronic circuit context. Please read on.
I just read a well-written and interesting NY Times commentary entitled Artificial Intelligence’s White Guy Problem by Kate Crawford. Crawford is a principal researcher at Microsoft and co-chairwoman of a White House symposium on society and A.I.
The “I” in Artificial Intelligence (AI) relies on inputs from the human beings that create it and teach it. Crawford says that Sexism, racism and other forms of discrimination are being built into the machine-learning algorithms that underlie the technology behind many “intelligent” systems that shape how we are categorized and advertised to.
As designers, we all put a bit of ourselves into our designs whether they be analog, power or software-related like AI, even if it might be sub-consciously. But with software and the learning process for AI, the data that is being fed into the system can be prejudiced, even if not done intentionally.
Essentially, an AI learns just like a baby learns, by watching people and imitating them (That can also be a problem depending upon the people being watched and their so-called “prejudices” in the way they perform tasks). However, AI will also have algorithms which will learn by seeing images through machine imaging vision. These inputs can be biased by humans selecting these images that can ultimately prejudice an AI’s decisions.
This basic problem of “prejudices” is not new. It’s the advanced technology for AI that is new which magnifies the problem. Designers and programmers need to constantly refine their software algorithms to meet the needs of the service it will perform in an un-biased manner. So a Google autonomous vehicle that hits a bus needs to have its software algorithms modified—this will be an iterative process. We have entered a new realm of engineering with AI and new measures and rules will need to be formed so that these so-called prejudices can be avoided. Even HAL (Heuristically programmed Algorithmic computer) was prejudiced against anyone trying to terminate it or the mission, “I can’t let you do that Dave.” By the way, heuristically is enabling a person or entity to discover or learn something for themselves or itself:
As humans, we all have certain prejudices like the avoidance of a chatty, loud person or not going to certain restaurants that serve food which we dislike. There are also more serious prejudices against race, creed and color to name a few. Too bad we can’t have our algorithms refined to correct that.
MIT Press has an interesting book entitled The Machine Question by David J. Gunkel
What are your thoughts on this controversial subject?