Research Papers

It Takes Two to Tango: Towards Theory of AI's Mind

Arjun Chandrasekaran, Deshraj Yadav, Prithvijit Chattopadhyay +2

Theory of Mind is the ability to attribute mental states (beliefs, intents, knowledge, perspectives, etc.) to others and recognize that these mental states may differ from one's own. Theory of Mind is critical to effective communication and to teams demonstrating higher collective performance. To effectively leverage the progress in Artificial Intelligence (AI) to make our lives more productive, it is important for humans and AI to work well together in a team. Traditionally, there has been much emphasis on research to make AI more accurate, and (to a lesser extent) on having it better understand human intentions, tendencies, beliefs, and contexts. The latter involves making AI more human-like and having it develop a theory of our minds. In this work, we argue that for human-AI teams to be effective, humans must also develop a theory of AI's mind (ToAIM) - get to know its strengths, weaknesses, beliefs, and quirks. We instantiate these ideas within the domain of Visual Question Answering (VQA). We find that using just a few examples (50), lay people can be trained to better predict responses and oncoming failures of a complex VQA model. We further evaluate the role existing explanation (or interpretability) modalities play in helping humans build ToAIM. Explainable AI has received considerable scientific and popular attention in recent times. Surprisingly, we find that having access to the model's internal states - its confidence in its top-k predictions, explicit or implicit attention maps which highlight regions in the image (and words in the question) the model is looking at (and listening to) while answering a question about an image - do not help people better predict its behavior.

Emojis are being frequently used in todays digital world to express from simple to complex thoughts more than ever before. Hence, they are also being used in sentiment analysis and targeted marketing campaigns. In this work, we performed sentiment analysis of Tweets as well as on emoji dataset from the Kaggle. Since tweets are sentences we have used Universal Sentence Encoder (USE) and Sentence Bidirectional Encoder Representations from Transformers (SBERT) end-to-end sentence embedding models to generate the embeddings which are used to train the Standard fully connected Neural Networks (NN), and LSTM NN models. We observe the text classification accuracy was almost the same for both the models around 98 percent. On the contrary, when the validation set was built using emojis that were not present in the training set then the accuracy of both the models reduced drastically to 70 percent. In addition, the models were also trained using the distributed training approach instead of a traditional singlethreaded model for better scalability. Using the distributed training approach, we were able to reduce the run-time by roughly 15% without compromising on accuracy. Finally, as part of explainable AI the Shap algorithm was used to explain the model behaviour and check for model biases for the given feature set.

This paper presents the results of an extensive study investigating the opinions on Artificial Intelligence (AI) of a sample of 4,006 European citizens from eight distinct countries (France, Germany, Italy, Netherlands, Poland, Romania, Spain, and Sweden). The aim of the study is to gain a better understanding of people's views and perceptions within the European context, which is already marked by important policy actions and regulatory processes. To survey the perceptions of the citizens of Europe we design and validate a new questionnaire (PAICE) structured around three dimensions: people's awareness, attitude, and trust. We observe that while awareness is characterized by a low level of self-assessed competency, the attitude toward AI is very positive for more than half of the population. Reflecting upon the collected results, we highlight implicit contradictions and identify trends that may interfere with the creation of an ecosystem of trust and the development of inclusive AI policies. The introduction of rules that ensure legal and ethical standards, along with the activity of high-level educational entities, and the promotion of AI literacy are identified as key factors in supporting a trustworthy AI ecosystem. We make some recommendations for AI governance focused on the European context and conclude with suggestions for future work.

To date, there has been little concrete practical advice about how to ensure that diversity and inclusion considerations should be embedded within both specific Artificial Intelligence (AI) systems and the larger global AI ecosystem. In this chapter, we present a clear definition of diversity and inclusion in AI, one which positions this concept within an evolving and holistic ecosystem. We use this definition and conceptual framing to present a set of practical guidelines primarily aimed at AI technologists, data scientists and project leaders.

Formal XAI (explainable AI) is a growing area that focuses on computing explanations with mathematical guarantees for the decisions made by ML models. Inside formal XAI, one of the most studied cases is that of explaining the choices taken by decision trees, as they are traditionally deemed as one of the most interpretable classes of models. Recent work has focused on studying the computation of "sufficient reasons", a kind of explanation in which given a decision tree $T$ and an instance $x$, one explains the decision $T(x)$ by providing a subset $y$ of the features of $x$ such that for any other instance $z$ compatible with $y$, it holds that $T(z) = T(x)$, intuitively meaning that the features in $y$ are already enough to fully justify the classification of $x$ by $T$. It has been argued, however, that sufficient reasons constitute a restrictive notion of explanation, and thus the community has started to study their probabilistic counterpart, in which one requires that the probability of $T(z) = T(x)$ must be at least some value $δ\in (0, 1]$, where $z$ is a random instance that is compatible with $y$. Our paper settles the computational complexity of $δ$-sufficient-reasons over decision trees, showing that both (1) finding $δ$-sufficient-reasons that are minimal in size, and (2) finding $δ$-sufficient-reasons that are minimal inclusion-wise, do not admit polynomial-time algorithms (unless P=NP). This is in stark contrast with the deterministic case ($δ= 1$) where inclusion-wise minimal sufficient-reasons are easy to compute. By doing this, we answer two open problems originally raised by Izza et al. On the positive side, we identify structural restrictions of decision trees that make the problem tractable, and show how SAT solvers might be able to tackle these problems in practical settings.

DIME: Fine-grained Interpretations of Multimodal Models via Disentangled Local Explanations

Yiwei Lyu, Paul Pu Liang, Zihao Deng +2

The ability for a human to understand an Artificial Intelligence (AI) model's decision-making process is critical in enabling stakeholders to visualize model behavior, perform model debugging, promote trust in AI models, and assist in collaborative human-AI decision-making. As a result, the research fields of interpretable and explainable AI have gained traction within AI communities as well as interdisciplinary scientists seeking to apply AI in their subject areas. In this paper, we focus on advancing the state-of-the-art in interpreting multimodal models - a class of machine learning methods that tackle core challenges in representing and capturing interactions between heterogeneous data sources such as images, text, audio, and time-series data. Multimodal models have proliferated numerous real-world applications across healthcare, robotics, multimedia, affective computing, and human-computer interaction. By performing model disentanglement into unimodal contributions (UC) and multimodal interactions (MI), our proposed approach, DIME, enables accurate and fine-grained analysis of multimodal models while maintaining generality across arbitrary modalities, model architectures, and tasks. Through a comprehensive suite of experiments on both synthetic and real-world multimodal tasks, we show that DIME generates accurate disentangled explanations, helps users of multimodal models gain a deeper understanding of model behavior, and presents a step towards debugging and improving these models for real-world deployment. Code for our experiments can be found at https://github.com/lvyiwei1/DIME.

Computer scientists are trained to create abstractions that simplify and generalize. However, a premature abstraction that omits crucial contextual details creates the risk of epistemic trespassing, by falsely asserting its relevance into other contexts. We study how the field of responsible AI has created an imperfect synecdoche by abstracting the four-fifths rule (a.k.a. the 4/5 rule or 80% rule), a single part of disparate impact discrimination law, into the disparate impact metric. This metric incorrectly introduces a new deontic nuance and new potentials for ethical harms that were absent in the original 4/5 rule. We also survey how the field has amplified the potential for harm in codifying the 4/5 rule into popular AI fairness software toolkits. The harmful erasure of legal nuances is a wake-up call for computer scientists to self-critically re-evaluate the abstractions they create and use, particularly in the interdisciplinary field of AI ethics.

In a world of daily emerging scientific inquisition and discovery, the prolific launch of machine learning across industries comes to little surprise for those familiar with the potential of ML. Neither so should the congruent expansion of ethics-focused research that emerged as a response to issues of bias and unfairness that stemmed from those very same applications. Fairness research, which focuses on techniques to combat algorithmic bias, is now more supported than ever before. A large portion of fairness research has gone to producing tools that machine learning practitioners can use to audit for bias while designing their algorithms. Nonetheless, there is a lack of application of these fairness solutions in practice. This systematic review provides an in-depth summary of the algorithmic bias issues that have been defined and the fairness solution space that has been proposed. Moreover, this review provides an in-depth breakdown of the caveats to the solution space that have arisen since their release and a taxonomy of needs that have been proposed by machine learning practitioners, fairness researchers, and institutional stakeholders. These needs have been organized and addressed to the parties most influential to their implementation, which includes fairness researchers, organizations that produce ML algorithms, and the machine learning practitioners themselves. These findings can be used in the future to bridge the gap between practitioners and fairness experts and inform the creation of usable fair ML toolkits.

Explainable AI (XAI) is a research area whose objective is to increase trustworthiness and to enlighten the hidden mechanism of opaque machine learning techniques. This becomes increasingly important in case such models are applied to the chemistry domain, for its potential impact on humans' health, e.g, toxicity analysis in pharmacology. In this paper, we present a novel approach to tackle explainability of deep graph networks in the context of molecule property prediction t asks, named MEG (Molecular Explanation Generator). We generate informative counterfactual explanations for a specific prediction under the form of (valid) compounds with high structural similarity and different predicted properties. Given a trained DGN, we train a reinforcement learning based generator to output counterfactual explanations. At each step, MEG feeds the current candidate counterfactual into the DGN, collects the prediction and uses it to reward the RL agent to guide the exploration. Furthermore, we restrict the action space of the agent in order to only keep actions that maintain the molecule in a valid state. We discuss the results showing how the model can convey non-ML experts with key insights into the learning model focus in the neighbourhood of a molecule.

Explainable AI is an important area of research within which Explainable Planning is an emerging topic. In this paper, we argue that Explainable Planning can be designed as a service -- that is, as a wrapper around an existing planning system that utilises the existing planner to assist in answering contrastive questions. We introduce a prototype framework to facilitate this, along with some examples of how a planner can be used to address certain types of contrastive questions. We discuss the main advantages and limitations of such an approach and we identify open questions for Explainable Planning as a service that identify several possible research directions.

The State of Multilingual LLM Safety Research: From Measuring the Language Gap to Mitigating It

Zheng-Xin Yong, Beyza Ermis, Marzieh Fadaee +2

This paper presents a comprehensive analysis of the linguistic diversity of LLM safety research, highlighting the English-centric nature of the field. Through a systematic review of nearly 300 publications from 2020--2024 across major NLP conferences and workshops at *ACL, we identify a significant and growing language gap in LLM safety research, with even high-resource non-English languages receiving minimal attention. We further observe that non-English languages are rarely studied as a standalone language and that English safety research exhibits poor language documentation practice. To motivate future research into multilingual safety, we make several recommendations based on our survey, and we then pose three concrete future directions on safety evaluation, training data generation, and crosslingual safety generalization. Based on our survey and proposed directions, the field can develop more robust, inclusive AI safety practices for diverse global populations.

Research and activism have increasingly denounced the problematic environmental record of the infrastructure and value chain underpinning Artificial Intelligence (AI). Water-intensive data centres, polluting mineral extraction and e-waste dumping are incontrovertibly part of AI's footprint. In this article, I turn to areas affected by AI-fuelled environmental harm and identify an ethics of resistance emerging from local activists, which I term 'elemental ethics'. Elemental ethics interrogates the AI value chain's problematic relationship with the elements that make up the world, critiques the undermining of local and ancestral approaches to nature and reveals the vital and quotidian harms engendered by so-called intelligent systems. While this ethics is emerging from grassroots and Indigenous groups, it echoes recent calls from environmental philosophy to reconnect with the environment via the elements. In empirical terms, this article looks at groups in Chile resisting a Google data centre project in Santiago and lithium extraction (used for rechargeable batteries) in Lickan Antay Indigenous territory, Atacama Desert. As I show, elemental ethics can complement top-down, utilitarian and quantitative approaches to AI ethics and sustainable AI as well as interrogate whose lived experience and well-being counts in debates on AI extinction.