To shine a light on the accomplishments of women in the AI domain, TechCrunch is initiating a series of interviews that spotlight exceptional women who have played a pivotal role in the AI revolution. These interviews will be featured throughout the year to coincide with the ongoing surge in AI advancements, aiming to showcase significant contributions that often go unnoticed. Explore more profiles here.
Emilia Gómez holds the position of principal investigator at the European Commission’s Joint Research Centre and serves as the scientific coordinator of AI Watch, an initiative by the EC aimed at monitoring the progress, adoption, and impact of AI in Europe. Her team provides valuable scientific and technical expertise to shape EC AI policies, including the recently introduced AI Act.
Grounded in the computational music domain, Gómez’s research delves into understanding how humans articulate music and the techniques used to digitally model it. Originating from the realm of music, Gómez explores the influence of AI on human behavior, particularly its effects on employment, decision-making, and the cognitive and socioemotional development of children.
Question and Answer Section
How did you venture into the field of AI, and what captivated you about it?
My journey into AI commenced with a focus on machine learning, specifically in crafting algorithms for the automated interpretation of music audio signals concerning melody, tonality, similarity, style, or emotion. These algorithms find applications across diverse domains, from music platforms to educational settings. My exploration revolved around designing innovative machine learning approaches for various computational tasks within the music sphere, emphasizing the significance of the data pipeline encompassing dataset creation and annotation. What intrigued me about machine learning was its modeling capabilities and the transition from knowledge-based to data-driven algorithm design. Instead of formulating descriptors based on our comprehension of acoustics and music, we began leveraging our expertise to devise datasets, architectures, and training and evaluation procedures.
Having witnessed the real-world impact of my machine learning research across different domains, from music platforms to symphonic music concerts, I recognized the profound influence these algorithms wield on individuals, such as listeners and musicians. This realization steered my focus towards AI evaluation rather than development, particularly examining the repercussions of AI on human behavior and exploring methodologies for evaluating systems in terms of aspects like fairness, human oversight, or transparency. This forms the current research focus of my team at the Joint Research Centre.
Which achievement in the AI domain are you particularly proud of?
From an academic and technical standpoint, I take pride in my contributions to music-specific machine learning architectures at the Music Technology Group in Barcelona, which have propelled advancements in the field, evident in my citation metrics. For instance, during my Ph.D., I introduced a data-driven algorithm for extracting tonality from audio signals (e.g., determining if a musical piece is in C major or D minor), which has emerged as a pivotal reference in the field. Subsequently, I collaborated on devising machine learning techniques for automatically describing music signals in terms of melody (e.g., facilitating song searches through humming), tempo, or emotional modeling in music. Many of these algorithms are now integrated into Essentia, an open-source library for audio and music analysis, description, and synthesis, and have been harnessed in numerous recommender systems.
One project that holds a special place is Banda Sonora Vital (LifeSoundTrack), honored with the Red Cross Award for Humanitarian Technologies, where we developed a personalized music recommender tailored to elderly Alzheimer’s patients. Additionally, the PHENICX project, a substantial European Union (EU)-funded initiative under my coordination, focused on leveraging music and AI to craft enriched symphonic music experiences.
I cherish the music computing community and am delighted to have served as the first female president of the International Society for Music Information Retrieval, where I have dedicated my career, with a keen interest in fostering diversity within the field.
In my current capacity at the Commission, where I assumed the role of lead scientist in 2018, I offer scientific and technical support for AI policies within the EU, notably the AI Act. Despite being less conspicuous in terms of publications, I am proud of my modest technical contributions to the AI Act. For instance, I played a role in harmonizing or translating legal and technical terminologies (e.g., proposing definitions grounded in existing literature) and evaluating the practical implementation of legal requirements, such as transparency or technical documentation for high-risk AI systems, general-purpose AI models, and generative AI.
I also take pride in my team’s efforts in bolstering the EU AI liability directive, where we delved into distinctive characteristics that render AI systems inherently risky, such as the absence of causality, opacity, unpredictability, or their self-learning and continuous-learning capabilities, and scrutinized the associated challenges in establishing causation.
How do you address the obstacles posed by the male-dominated tech and AI industries?
Navigating through not just the tech industry but also the male-dominated AI research and policy sphere poses its challenges. I lack a specific technique or strategy as this environment is all I have known. The prospect of working in a diverse or female-dominated setting seems appealing, echoing the sentiments of the Beach Boys’ song, “Wouldn’t it be nice?” I strive to sidestep frustration and relish the experience of collaborating in a milieu predominantly governed by assertive male counterparts while also enjoying partnerships with exceptional women in the field.
What guidance would you offer to women aspiring to enter the AI domain?
I would impart two pieces of advice:
-
Your presence is indispensable—step into our domain as there is a pressing need for a diverse array of perspectives, approaches, and ideas. For instance, according to the divinAI project, a venture I co-established to monitor diversity in the AI realm, only 23% of author names at the International Conference on Machine Learning and 29% at the International Joint Conference on AI in 2023 belonged to females, irrespective of their gender identity.
-
You are not alone—there exists a multitude of women, nonbinary colleagues, and male allies in the field, even if our visibility or recognition may be limited. Seek out these individuals for mentorship and support! The research arena hosts numerous affinity groups. For instance, during my tenure as president of the International Society for Music Information Retrieval, I actively participated in the Women in Music Information Retrieval initiative, a pioneering effort in promoting diversity in music computing featuring a highly successful mentoring program.
What are the most critical challenges confronting the evolution of AI?
In my view, researchers should devote equal attention to AI evaluation as they do to AI development, as there currently exists an imbalance. The research community is engrossed in advancing the state of the art concerning AI capabilities and performance, often overlooking the necessity for comprehensive evaluations, impact assessments, and external audits. The more sophisticated AI systems become, the more sophisticated their evaluations must be. The realm of AI evaluation remains underexplored, leading to numerous incidents that tarnish AI’s reputation, such as gender or racial biases embedded in datasets or algorithms.
What should AI users be mindful of?
Individuals utilizing AI-driven tools like chatbots should recognize that AI is not synonymous with magic; rather, it is a product of human intelligence. It is imperative for users to acquaint themselves with the operational principles and limitations of AI algorithms to effectively scrutinize and responsibly utilize them. Moreover, users should stay informed about the quality of AI products, their evaluation or certification processes, enabling them to discern which products merit their trust.
What is the optimal approach to ethically develop AI?
In my perspective, the optimal strategy for crafting AI products with a positive social and environmental impact, in an ethical manner, involves allocating resources towards evaluation, assessing social impact, and mitigating risks—particularly those pertaining to fundamental rights—prior to introducing an AI system into the market. This approach not only benefits businesses and instills trust in products but also serves the interests of society.
Responsible AI or trustworthy AI entails constructing algorithms that prioritize aspects like transparency, fairness, human oversight, and social and environmental well-being right from the initial stages of AI design. The AI Act not only establishes benchmarks for regulating artificial intelligence globally but also mirrors Europe’s emphasis on trustworthiness and transparency, fostering innovation while safeguarding citizens’ rights. This, I believe, will bolster public trust in both the product and the technology.
