9 September 2019 In Portraits By URelles

We met the specialist in artificial intelligence ethics, and we’re pretty reassured about the future.

Creating technology is great, but knowing how it will be used is even better! URelles met Dr. Rumman Chowdhury, THE specialist in artificial intelligence (AI) ethics, at the Mutek_IMG forum in Montreal. She was presenting a conference on “algorithmic colonialism”. Quite an encounter! Interview with a passionate professor.

Émilie Vion: Why is it important for you to work on the ethics of artificial intelligence?

Rumman Chowdhury: It’s very important because these technologies are used in everyday life, all the time. Most people don’t have the knowledge required to understand or recognize the technical logic of AI. In my background, I felt it was important to use it for the good of all and thus help people understand AI.

Émilie Vion: You say that the most important thing is to make people aware of the impact of these technologies. How can non-specialists be helped to better understand AI?

Rumman Chowdhury: In the context of training or awareness-raising, people don’t need to know how to program. Above all, they need to be able to better understand and measure the effects of these technologies on our daily lives. The important thing is to understand how and why technologies are implemented in our societies. We also need to ask ourselves what impact technologies have on human lives.

Émilie Vion: How can we better distinguish between what is AI and what isn’t?

Rumman Chowdhury: The term AI is becoming increasingly overused. To be more precise, we could say that true artificial intelligence is what is known as deep learning, when an artificial intelligence selects information from real-time data, which is then modeled like a human brain by analysts, in order to propose a usable result. This is real learning, in real time. On the other hand, we have machine learning, which is different from deep learning. This is a theoretical explanation from technologists. From an academic or social point of view, the definition is almost anecdotal.

Émilie Vion: The notion of impartiality is important to you. What components does artificial intelligence need to be more ethical?

Rumman Chowdhury: To be fairer, we’d have to talk about fairness, responsibility, transparency and definition. There are two ways of looking at it:

1. How to set up an algorithm in such a way as to present a more transparent and fair result. The algorithm doesn’t exist in a bubble cut off from the world, it belongs to a technical ecosystem. Simply calibrating the algorithm in the right way is not enough.

2. It’s also important to know how the algorithm will be used and interpreted in the future. Even if the data is accurate and correct, this doesn’t necessarily guarantee that the algorithm’s output will be used appropriately.

Émilie Vion: This brings us to the question of bias in AI. How could they be reduced, and what are your thoughts on this, particularly as the daughter of an immigrant?

Rumman Chowdhury: It’s in everything that defines me, whether as the daughter of immigrants, as a queer woman, and moreover, as someone who works in technology. Bias is deeply rooted in our society. Interestingly, the principle of bias is understood in completely different ways in the data sciences and the social sciences. From a purely technical point of view, bias is a view held by the majority and therefore quantifiable, implying that there is an absolute truth. If we use the right data, in the right way, there shouldn’t be any biases or difficulties of interpretation.

From the point of view of the social sciences, even if the data are accurate, there can be bias, in the sense of partiality. For example, in a tool for selecting candidates for a job in IT, even if the data is as accurate as possible, the result will be subjective because of the societal biases that humans have with regard to the place of women in technology.

We mustn’t think about bias exclusively from the point of view of the data, but also from that of how humans use the result.

Émilie Vion: Would one of the solutions be to have more diversity in programming and development?

Rumman Chowdhury: There’s an example I like to give in relation to this question. I recently went to Norway during the winter, and found that at a certain temperature, my smartphone would switch off. The only reason the manufacturer hasn’t thought about this is because they developed the tool in California and didn’t take this issue into account. A team able to develop a smartwatch capable of recognizing the type of breaststroke performed in a swimming pool can certainly make a cell phone resistant to the cold. Due to a lack of geographic diversity, this element was not considered. So yes, any form of diversity brings a solution.

Émilie Vion: You became a data processing scientist in 2014, before it became the hottest job in tech. What attracted you to this specialty?

Rumman Chowdhury: I was often told that I had a unique background with an outstanding pedigree. There was also a shortage of social scientists. I first understood and imagined the position as a social science job where you have to use and analyze data about humans and represent them in statistical models. It’s not just a job for a programmer. Programmers are not trained to understand massively localized human data. We need to understand that the use of AI goes beyond pure code. From the point of view of the developer, who will be entirely focused on writing the code, external interpretation by users will not be anticipated. For example, in a tool helping to anticipate crime rates, even if the data is correct and the algorithm has been developed in the best possible way, the result will be heavily based on physical aspects. This is biased and unfair. So it’s important to think beyond the development and programming process, and focus more on the problem-solving process and how it will be used.

About
Latest Posts

URelles

Auteur de l'article.