I think that Responsible AI is one of the most important topics covered during data science day at the iSchool. Of course, there were many other important topics also discussed but now, whenever I think of negative outcomes of unresponsible AI, it reminds me Hollywood movie of “Terminator” in which “Skynet” AI system seized control of weapon system and initiated nuclear war.

In short, one of the important insights gained or solidified during the lecture “Exploring Responsible AI” presented by Jasmina Tacheva, was that AI has greater power, and the greater the potential benefits, the greater the for potential damages. I think that this insight is very important for data scientists to consider.

The Power of AI is Increasing

According to Accenture and Frontier Economics, by 2035, it is predicted that the use of AI will increase dramatically (in some industries including education, construction, healthcare, and wholesale/retail, the figures are between 55%-84%!). Therefore, as data scientists, we should pay special attention to ethics and biases within the scope of the responsible AI.

The Importance of Ethics When Using AI

Regarding ethics, as data scientists, we should be careful not to create ethical issues such as using personal data without consent because AI systems are socio-technical as Professor Jasmina Tacheva taught us during her data science day discussion. This point was also re-enforced later, when Professor Jeffrey Saltz discussed these issues in my Introduction to Data Science class (IST687). Therefore, it is necessary to make AI tools responsible and core of it is transparency and data protection.

We Need to Understand the Data Used to Build the Model

As a data scientist, we should be careful and know about potential harmful impact from our work.

A key part of minimizing ethical issues is understanding the data used to build a model. Biased dataset can lead to biased model and as a result, all our outcomes and predictions can be biased. Hence, we must avoid making the mistake on collecting historical data that is biased. This is of utmost importance for data scientists. This was covered by both Prof. Tacheva as well as Prof. Saltz in IST687.

Some Biases are more Challenging to Identify and Mitigate

In terms of data biases, there are several biases such as non-response bias, sampling bias, and historical bias. It can sometimes be easy to detect and correct data-related and/or computational biases. However, other biases, such as human emotional bias can be difficult to eliminate as humans often rely on mental shortcuts. This insight was from Professor Jaime Banks talk – Our Minds and the Machines: Exploring Mental Shortcuts in Human-AI Interactions.

Furthermore, as assistant professor Jaime Banks stressed, there is threat that people easily rely on social machines and allowing those machines run heuristics. Based on the above-mentioned reasons, AI system are socio-technical, we must consider AI from many angles.

(As aside, another interesting topic discussed by Assistant Professor Jaime Banks was how pollination problem can be addressed using new bees, cyborg bees, or robot bees. This topic was very interesting and pollination problem is one of the most challenging issues as bee population is freefall worldwide according to UN, and without bee pollinators, one-thirds of our food source could disappear according to some analysis. Specifically, in a study published in Insect Conservation and Diversity, Rutgers scientists confirmed that pollination of watermelon flowers by wild bees decreased by more than half between 2005 and 2012.)

Key Take-Aways

Generally, the insights gained from the session helped me realize that I should try to ensure the responsible application of AI and explore potential issues before starting AI projects.

I now better understand the importance responsible AI and what issues we might encounter during AI-related analysis. I also understand the social aspect of AI, as it is used to address many issues around us and makes our lives much simpler.

Finally, I liked Data Science Day as professors shared their research areas and some findings about where AI and human beings are now and in the future. It was interesting to obtain knowledge by attending the session and learning a lot from the professors.