What made you decide to pursue a career in AI research?
The general idea of doing something STEM was pretty obvious to me. My mom is a math teacher, so I was exposed to math since childhood and it led to my passion for it. Until high school I was mostly interested in mathematics and computer science existed at the edge of my interest area. However, when doing my undergraduate degree, I discovered that the most interesting subjects were theoretical computing. Because of my background, the most reasonable option to pursue that interest was research.
Briefly, what is your current research at IDEAS NCBR about?
In the “AI for Security” research team led by Tomasz Michalak, we are working on modeling patrolling scenarios for infrastructure protection and generating most efficient strategies. The problem itself has many exact solutions already, but they scale up very badly. Existing methods can solve problems with 5-10 potential attack targets. In real life, we want to protect infrastructure with the number of targets larger by at least an order of magnitude, i.e. 500 targets or more. Additionally, with size the connections between them get more complex. These aren’t problems solvable in a reasonable time through exact methods. We are aiming to create solutions that will be actually useful and work towards implementing existing ones in real-life applications.
Do you think there is a gender disproportion in the number of AI/ML experts?
If I was to reply based on a quick look around, it would be an astounding yes. I see it especially among senior researchers – among people around me, at my university or authors of papers I’m reading. Which is not to say that the situation is much better among my peers, although the imbalance is lower. A quick google search shows that this observation is not some local anomaly, but a global trend. In 2020 it has been estimated that only about 12% of leading ML researchers were women.
If yes, why? What could be the reason for so few women to decide to pursue a scientific career in AI/ML?
I think the reason is two-fold.
There are girls that could’ve developed interest and necessary skills, but never did. This happens in many ways. In many environments the belief that women are innately less skilled in STEM still persists. It is rare to be said directly, although I have encountered people (even teachers, even women) who did say that openly. Most importantly, however, it leads to differences in how boys’ and girls’ mistakes and successes are treated.
Successes in math or science, especially at early stages of education, tend to be presented as an effect of natural talent for boys and of hard work for girls (despite requiring both). This leads to girls underestimating their capabilities and withdrawing from STEM related activities. Additionally, a significant proportion of “talented” students have a variety of skills and interests and the choices they make about their education (choice of school or final examination subjects) are left up to a chance. Gender disproportion among adults they see is one of the factors that may sway the decision.
There are also women who already developed interest in ML, who studied computer science or math, who know how to program and have a solid mathematical background but decide to not do research. The question about whether to go into academia or industry bothers people regardless of their gender, but women have to consider many more potential cons of scientific career. A common one is the ease with which it can be reconciled with motherhood. If a woman wants to have a child, unless she decides to adopt, significant proportion of initial responsibilities falls on her. Pregnancy (and the recuperation after one), initial care requires a lot of time away from work, and even more so does further childcare that often is her responsibility because of the social approach to parenthood.
Then there is the need to deal with sexism that unfortunately comes with working in male-dominated domain. Even before entering academic institutions as their employees, women experience it from the same people they will be working among. The fears it prompts aren’t baseless: experienced female researchers do report both minor (impacting well-being) and major (impacting course of career) experiences of sexism.
What, in your opinion, could change that situation?
The reason is two-fold and such should be the response.
We should certainly make sure that teachers pay attention to their biases. It’s impossible to be fully unbiased, but there is still a lot that could be done in that matter. This applies equally to kindergarten teachers and university lecturers. Changing general perception of women in STEM, e.g. through providing examples of their successes, would also be helpful in lowering both children’s and their parents/guardians’ biases.
In case of STEM interested and skilled women, the priority should be creation of friendlier environments, such as more efficient and safer methods of reporting sexist behaviors and recruitment and evaluation processes reducing possibility of (conscious or unconscious) assessor bias by anonymizing personal data at stages where possible.
How can women contribute to the development of this area of science?
The first thing that comes to my mind is: the same as anyone of any other gender. Even if women are on average socialized differently to men, they tend to behave and think similarly to them in similar roles. There is a lot of potential benefit from including women in creating AI. As much as we would like to see science as objective and detached from our personal experiences, that’s not true. There are multiple examples of research and technology forgetting about differences between average man and average woman. A famous one is that car safety tests weren’t required to include mannequins with typical female body shape until this century. Therefore, having more diverse researchers can lead to inclusion of diverse experiences in modeling real life.
If you were to point out one character trait that describes a good scientist, what would it be?
Patience. Much of the work consists of meticulously analyzing existing research (for gaps, baselines and solutions), precisely checking your own ideas and conducting a lot of very similar experiments.