Last year, politicians and policymakers around the world also grappled to make sense of how AI will fit into society. President Biden issued an executive order saying AI was the "most consequential technology of our time." The United Kingdom convened a global AI forum where Prime Minister Rishi Sunak warned that "humanity could lose control of AI completely." The concerns include the risk that "generative" AI -- which can create text, video, images and audio -- can be used to create misinformation, displace jobs or even help people create dangerous bioweapons.

When it comes to the California DMV, is this a case of brand new year, same old tune? It's a positive sign that the massive bureaucracy's director has been checking out reader complaints about the license renewal process for drivers after age 70, and here's a news bulletin: He's even tossing out some of the crazy test questions that many of you have been griping about. I'll get to that in a moment, but first let's dip into the mail bag, which continues to overflow with tales from the Department of Motor Vehicles. Dave Warburton, 76, of Santa Clarita went to renew his license the first week of January and was told there was no record of his pre-registration in the computer system. "Not off to a good start," he wrote in an email.

Just a few days ago, OpenAI's usage policies page explicitly states that the company prohibits the use of its technology for "military and warfare" purposes. That line has since been deleted. As first noticed by The Intercept, the company updated the page on January 10 "to be clearer and provide more service-specific guidance," as the changelog states. It still prohibits the use of its large language models (LLMs) for anything that can cause harm, and it warns people against using its services to "develop or use weapons." However, the company has removed language pertaining to "military and warfare."

In an era characterized by the pervasive integration of artificial intelligence into decision-making processes across diverse industries, the demand for trust has never been more pronounced. This thesis embarks on a comprehensive exploration of bias and fairness, with a particular emphasis on their ramifications within the banking sector, where AI-driven decisions bear substantial societal consequences. In this context, the seamless integration of fairness, explainability, and human oversight is of utmost importance, culminating in the establishment of what is commonly referred to as "Responsible AI". This emphasizes the critical nature of addressing biases within the development of a corporate culture that aligns seamlessly with both AI regulations and universal human rights standards, particularly in the realm of automated decision-making systems. Nowadays, embedding ethical principles into the development, training, and deployment of AI models is crucial for compliance with forthcoming European regulations and for promoting societal good. This thesis is structured around three fundamental pillars: understanding bias, mitigating bias, and accounting for bias. These contributions are validated through their practical application in real-world scenarios, in collaboration with Intesa Sanpaolo. This collaborative effort not only contributes to our understanding of fairness but also provides practical tools for the responsible implementation of AI-based decision-making systems. In line with open-source principles, we have released Bias On Demand and FairView as accessible Python packages, further promoting progress in the field of AI fairness.

Business and consumer surveys are an essential tool used by policy-makers and practitioners to monitor and forecast the economy. Their most valuable feature is to provide timely information about the current and expected state of economic activity that is relevant to integrate the sluggish release of macroeconomic indicators. Interestingly, surveys are often interpreted as measures of economic sentiment in the sense of providing the pulse of different aspects of the economy, such as the consumers' attitude toward spending or the expectation of purchasing managers about inflation. Some prominent examples are represented by the Survey of Consumers of the University of Michigan (MCS) for the United States (Curtin and Dechaux, 2015) and the Business and Consumer Survey (BCS) for the European Union (European Commission, 2016). Although surveys are very valuable and accurate proxies of economic activity, they are typically released at the monthly frequency which might limit their usefulness in high-frequency nowcasting of economic variables (Aguilar et al., 2021; Algaba et al., 2023).

Command and control (C2) paths for issuing commands to malware are sometimes the only indicators of its existence within networks. Identifying potential C2 channels is often a manually driven process that involves a deep understanding of cyber tradecraft. Efforts to improve discovery of these channels through using a reinforcement learning (RL) based approach that learns to automatically carry out C2 attack campaigns on large networks, where multiple defense layers are in place serves to drive efficiency for network operators. In this paper, we model C2 traffic flow as a three-stage process and formulate it as a Markov decision process (MDP) with the objective to maximize the number of valuable hosts whose data is exfiltrated. The approach also specifically models payload and defense mechanisms such as firewalls which is a novel contribution. The attack paths learned by the RL agent can in turn help the blue team identify high-priority vulnerabilities and develop improved defense strategies. The method is evaluated on a large network with more than a thousand hosts and the results demonstrate that the agent can effectively learn attack paths while avoiding firewalls.

The emergence of unveiling human-like behaviors in Large Language Models (LLMs) has led to a closer connection between NLP and human psychology, leading to a proliferation of computational agents. Scholars have been studying the inherent personalities displayed by LLM agents and attempting to incorporate human traits and behaviors into them. However, these efforts have primarily focused on commercially-licensed LLMs, neglecting the widespread use and notable advancements seen in Open LLMs. This work aims to address this gap by conducting a comprehensive examination of the ability of agents to emulate human personalities using Open LLMs. To achieve this, we generate a set of ten LLM Agents based on the most representative Open models and subject them to a series of assessments concerning the Myers-Briggs Type Indicator (MBTI) test. Our approach involves evaluating the intrinsic personality traits of Open LLM agents and determining the extent to which these agents can mimic human personalities when conditioned by specific personalities and roles. Our findings unveil that: $(i)$ each Open LLM agent showcases distinct human personalities; $(ii)$ personality-conditioned prompting produces varying effects on the agents, with only few successfully mirroring the imposed personality, while most of them being ``closed-minded'' (i.e., they retain their intrinsic traits); $(iii)$ combining role and personality conditioning can enhance the agents' ability to mimic human personalities; and $(iv)$ personalities typically associated with the role of teacher tend to be emulated with greater accuracy. Our work represents a step up in understanding the dense relationship between NLP and human psychology through the lens of Open LLMs.

Recently, many studies utilized adversarial examples (AEs) to raise the cost of malicious image editing and copyright violation powered by latent diffusion models (LDMs). Despite their successes, a few have studied the surrogate model they used to generate AEs. In this paper, from the perspective of adversarial transferability, we investigate how the surrogate model's property influences the performance of AEs for LDMs. Specifically, we view the time-step sampling in the Monte-Carlo-based (MC-based) adversarial attack as selecting surrogate models. We find that the smoothness of surrogate models at different time steps differs, and we substantially improve the performance of the MC-based AEs by selecting smoother surrogate models. In the light of the theoretical framework on adversarial transferability in image classification, we also conduct a theoretical analysis to explain why smooth surrogate models can also boost AEs for LDMs.

This paper introduces and studies online conversion with switching costs (OCS), a novel class of online problems motivated by emerging control problems in the design of sustainable systems. We consider both minimization (OCS-min) and maximization (OCS-max) variants of the problem. In OCS-min, an online player aims to purchase one item over a sequence of time-varying cost functions and decides the fractional amount of item to purchase in each round. The player must purchase the entire item before a deadline, and they incur a movement cost whenever their decision changes, i.e., whenever they purchase different amounts of the item in consecutive time steps. From the player's perspective, the goal is to minimize their total cost, including the total purchasing cost and any movement cost incurred over the time horizon. In OCS-max, the setting is almost the same, except the player sells an item fractionally according to time-varying price functions, so the goal is to maximize their total profit, and any movement costs are subtracted from the revenue. In both settings, the cost/price functions are revealed one by one in an online manner, and the player makes an irrevocable decision at each time step without the knowledge of future cost/price functions. Our motivation behind introducing OCS is an emerging class of carbon-aware problems such as carbon-aware electric vehicle (EV) charging [12] and carbon-aware compute shifting [1, 3, 22, 23, 46, 57], which have attracted significant attention in recent years.

Recent years have seen rapid development of descriptor generation based on representation learning of extremely diverse molecules, especially those that apply natural language processing (NLP) models to SMILES, a literal representation of molecular structure. However, little research has been done on how these models understand chemical structure. To address this black box, we investigated the relationship between the learning progress of SMILES and chemical structure using a representative NLP model, the Transformer. We show that while the Transformer learns partial structures of molecules quickly, it requires extended training to understand overall structures. Consistently, the accuracy of molecular property predictions using descriptors generated from models at different learning steps was similar from the beginning to the end of training. Furthermore, we found that the Transformer requires particularly long training to learn chirality and sometimes stagnates with low performance due to misunderstanding of enantiomers. These findings are expected to deepen the understanding of NLP models in chemistry.