We are excited to bring Transform 2022 back in-person July 19 and virtually July 20 - 28. Join AI and data leaders for insightful talks and exciting networking opportunities. Register today!
Women in the AI field are making research breakthroughs, spearheading vital ethical discussions, and inspiring the next generation of AI professionals. We created the VentureBeat Women in AI Awards to emphasize the importance of their voices, work, and experience and to shine a light on some of these leaders. In this series, publishing Fridays, we’re diving deeper into conversations with this year’s winners, whom we honored recently at Transform 2021. Check out last week’s interview with a winner of our AI responsibility and ethics award.
For many people, nothing sounds more tech-y than AI and machine learning. But the field is of rising interest in social sciences, too. Arezou Soltani Panah, a computer scientist and the winner of VentureBeat’s Women in AI rising star award, has made strides using AI technologies to tackle complex social problems such as loneliness, family violence, and social stigma. She’s even created novel machine learning techniques for this work, specifically.
“Traditional social science is a territory mostly occupied with qualitative researchers and empirical scientists, and perhaps they’re less aware of the benefits of AI for their field,” she told VentureBeat. “But the marriage between AI and social science has already been established and is evolving.”
An immigrant to Australia from Iran, Soltani Panah’s work focuses on social inequality and disempowerment. And it’s cross-discipline in every way, often requiring collaboration with government policy advisors and subject matter experts like social scientists.
We’re pleased to present Soltani Panah with this much-deserved award. We recently caught up with her to chat more about how social science and AI come together, her favorite projects, and future hope for the field of AI.
This interview has been edited for brevity and clarity.
VentureBeat: How do you define your work?
Arezou Soltani Panah: Over the past decade, I’ve worked on interdisciplinary projects and have developed various research interests along with a technical skill set. My PhD in computer science focused on the application of digital watermarking technology for secure data provenance transmission, mostly for IoT big data. I later joined Swinburne Social Innovation Research Institute, where my research focused on solving wicked social problems such as loneliness, family violence, and social stigma. I created a range of novel machine learning solutions that spanned across those disciplines to create responsible AI research. And last year, I joined Deakin’s Applied Artificial Intelligence Institute as part of the Defence Applied AI Experiential CoLab project. The goal was to enhance search and rescue missions in Australia, mainly with the aid of natural language processing techniques.
Prior to my academic career, I worked as a software engineer in industry for four years. And as is evident from my career trajectory, when it comes to learning opportunities and problem solving, I never confine myself to one particular area. In my opinion, innovation requires both reaching across fields and, often, acquiring more than a surface-level understanding of those fields.
VentureBeat: How did you first get into AI? And what interests you most about it?
Soltani Panah: My first pragmatic introduction to AI in terms of using algorithms at a tremendous speed dates back to 2010, where I was completing my bachelor honors degree at Tehran Polytechnic. For my research project, I used a suite of statistical machine learning methods for classifying speech acts as a part of a synthesized text-to-speech system for Persian language. Back then, the number of open source libraries for both speech processing and machine learning was very limited, and a lot of my time was spent on boilerplate coding and mathematical modeling.
Fast forward to 10 years later, I was exposed to a similar project generating transcription data for a hypothetical search and search exercise. It was really fascinating to see the stage we’re in now in terms of availability of machine learning frameworks and the abundance of pretrained models. Now that machine learning models have advanced to a certain point of intelligence, I think there’s more opportunities to focus on the human reasoning aspect of AI.
VentureBeat: People typically think of AI and machine learning in terms of the tech industry and very tech-y use cases. But what role can it play in social sciences?
Soltani Panah: Traditional social science is a territory mostly occupied with qualitative researchers and empirical scientists, and perhaps they’re less aware of the benefits of AI for their field. But the marriage between AI and social science has already been established and is evolving. Having said that, there is still a long way to go to customize AI solutions for social studies. For instance, esteem and personal recognition need has benefited from advancements in AI, where work around searching for, finding, and identifying information has been made easier through natural language processing (NLP) techniques such as sentiment analysis, stance detection, citation analysis, and so on and so forth. Additionally, historical data and long-term causal effects are in the center of attention for many social science projects, and using AI to answer casual questions is very complex.
VentureBeat: And what kinds of challenges are there when using AI for this type of work? Is it welcomed by others or misunderstood? What about dealing with data and algorithmic bias?
Soltani Panah: There is an ongoing chaos around algorithmic bias and wrestling with some ethical questions in relation to data privacy. In my humble opinion, ethics and privacy are orthogonal issues. In particular, there’re many privacy-enhancing techniques with a strong privacy guarantee (such as k-anonymity, l-diversity, t-similarity, and other blockchain-based approaches) that can be helpful in reconciling the use of data for research and privacy. However, many of these techniques do not come as off-the-shelf open-source solutions and therefore do not get the recognition they deserve. In relation to algorithmic bias, I must mention, the computer picked the bias from humans in the first place. If you’re adding, say, race as a column to a dataset and training model, you’re already assuming there is a correlation between your predictive variable and race — and your AI model will continue to do so.
VentureBeat: Of all your research projects so far, is there one that sticks out to you? And is there an area of research you have your eye on to tackle with AI next?
Soltani Panah: I have two favorite projects in the area of policy informatics. One was measuring outcomes of family violence policy interventions. This project was commissioned by the Victoria Department of Premier and Cabinet, and we applied text mining and topic modeling techniques across four data sources to understand the point of contrast between government policies and public perceptions of family violence, and how the discussion has changed over time. My second favorite was related to tracking changes in news media framing of obesity over three decades and how this relates to public health policy developments. We teamed up with Victorian Health Promotion Foundation and developed word embedding models to measure the degree of weight stigmatization on four key individual or structural dimensions including gender, healthiness, socioeconomic status, and stereotypes.
Next, I’m transitioning to industry, for good reasons. As recently stated by the Federal Education Minister Alan Tudge, research commercialization will play a strong role in post-COVID recovery for the Australian community. In my opinion, universities are not training the required taskforce for this to happen right now, and most of the time, the AI-related research projects are not production-ready by industry standard. I most recently joined a Melbourne-based startup, Little Birdie, to extend my technical experience in operationalizing AI systems. Hopefully, this is not an end to my research career, but instead makes me prepared for the nascent field of MLOps.
VentureBeat: What is your hope for the future of AI as a field and as a tool?
Soltani Panah: One thing I’m particularly interested in and hopeful about is developing artificial general intelligence. Right now, people are trying to use one-size-fits-all technologies based on deep neural networks to build intelligence for tasks that are fundamentally different. We have observed recent breakthroughs in computer vision and natural language processing fields with AlphaZero and GPT-3, but there are still serious shortcomings for in-context learning. While some people have predicted that AGI won’t arrive any time sooner than the year 2300, whenever it happens, it will be a profound scientific achievement.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Learn more about membership.