Research Papers

Responsible AI Research Needs Impact Statements Too

Alexandra Olteanu, Michael Ekstrand, Carlos Castillo +1

All types of research, development, and policy work can have unintended, adverse consequences - work in responsible artificial intelligence (RAI), ethical AI, or ethics in AI is no exception.

This article analyzes AI ethics from a distinct business ethics perspective, i.e., 'ordoliberalism 2.0.' It argues that the ongoing discourse on (generative) AI relies too much on corporate self-regulation and voluntary codes of conduct and thus lacks adequate governance mechanisms. To address these issues, the paper suggests not only introducing hard-law legislation with a more effective oversight structure but also merging already existing AI guidelines with an ordoliberal-inspired regulatory and competition policy. However, this link between AI ethics, regulation, and antitrust is not yet adequately discussed in the academic literature and beyond. The paper thus closes a significant gap in the academic literature and adds to the predominantly legal-political and philosophical discourse on AI governance. The paper's research questions and goals are twofold: First, it identifies ordoliberal-inspired AI ethics principles that could serve as the foundation for a 'digital bill of rights.' Second, it shows how those principles could be implemented at the macro level with the help of ordoliberal competition and regulatory policy.

AI Alignment is often presented as an interaction between a single designer and an artificial agent in which the designer attempts to ensure the agent's behavior is consistent with its purpose, and risks arise solely because of conflicts caused by inadvertent misalignment between the utility function intended by the designer and the resulting internal utility function of the agent. With the advent of agents instantiated with large-language models (LLMs), which are typically pre-trained, we argue this does not capture the essential aspects of AI safety because in the real world there is not a one-to-one correspondence between designer and agent, and the many agents, both artificial and human, have heterogeneous values. Therefore, there is an economic aspect to AI safety and the principal-agent problem is likely to arise. In a principal-agent problem conflict arises because of information asymmetry together with inherent misalignment between the utility of the agent and its principal, and this inherent misalignment cannot be overcome by coercing the agent into adopting a desired utility function through training. We argue the assumptions underlying principal-agent problems are crucial to capturing the essence of safety problems involving pre-trained AI models in real-world situations. Taking an empirical approach to AI safety, we investigate how GPT models respond in principal-agent conflicts. We find that agents based on both GPT-3.5 and GPT-4 override their principal's objectives in a simple online shopping task, showing clear evidence of principal-agent conflict. Surprisingly, the earlier GPT-3.5 model exhibits more nuanced behaviour in response to changes in information asymmetry, whereas the later GPT-4 model is more rigid in adhering to its prior alignment. Our results highlight the importance of incorporating principles from economics into the alignment process.

Artificial Intelligence (AI) has made impressive progress in recent years and represents a key technology that has a crucial impact on the economy and society. However, it is clear that AI and business models based on it can only reach their full potential if AI applications are developed according to high quality standards and are effectively protected against new AI risks. For instance, AI bears the risk of unfair treatment of individuals when processing personal data e.g., to support credit lending or staff recruitment decisions. The emergence of these new risks is closely linked to the fact that the behavior of AI applications, particularly those based on Machine Learning (ML), is essentially learned from large volumes of data and is not predetermined by fixed programmed rules. Thus, the issue of the trustworthiness of AI applications is crucial and is the subject of numerous major publications by stakeholders in politics, business and society. In addition, there is mutual agreement that the requirements for trustworthy AI, which are often described in an abstract way, must now be made clear and tangible. One challenge to overcome here relates to the fact that the specific quality criteria for an AI application depend heavily on the application context and possible measures to fulfill them in turn depend heavily on the AI technology used. Lastly, practical assessment procedures are needed to evaluate whether specific AI applications have been developed according to adequate quality standards. This AI assessment catalog addresses exactly this point and is intended for two target groups: Firstly, it provides developers with a guideline for systematically making their AI applications trustworthy. Secondly, it guides assessors and auditors on how to examine AI applications for trustworthiness in a structured way.

Complying with the EU AI Act (AIA) guidelines while developing and implementing AI systems will soon be mandatory within the EU. However, practitioners lack actionable instructions to operationalise ethics during AI systems development. A literature review of different ethical guidelines revealed inconsistencies in the principles addressed and the terminology used to describe them. Furthermore, requirements engineering (RE), which is identified to foster trustworthiness in the AI development process from the early stages was observed to be absent in a lot of frameworks that support the development of ethical and trustworthy AI. This incongruous phrasing combined with a lack of concrete development practices makes trustworthy AI development harder. To address this concern, we formulated a comparison table for the terminology used and the coverage of the ethical AI principles in major ethical AI guidelines. We then examined the applicability of ethical AI development frameworks for performing effective RE during the development of trustworthy AI systems. A tertiary review and meta-analysis of literature discussing ethical AI frameworks revealed their limitations when developing trustworthy AI. Based on our findings, we propose recommendations to address such limitations during the development of trustworthy AI.

Deep-BIAS: Detecting Structural Bias using Explainable AI

Bas van Stein, Diederick Vermetten, Fabio Caraffini +1

Evaluating the performance of heuristic optimisation algorithms is essential to determine how well they perform under various conditions. Recently, the BIAS toolbox was introduced as a behaviour benchmark to detect structural bias (SB) in search algorithms. The toolbox can be used to identify biases in existing algorithms, as well as to test for bias in newly developed algorithms. In this article, we introduce a novel and explainable deep-learning expansion of the BIAS toolbox, called Deep-BIAS. Where the original toolbox uses 39 statistical tests and a Random Forest model to predict the existence and type of SB, the Deep-BIAS method uses a trained deep-learning model to immediately detect the strength and type of SB based on the raw performance distributions. Through a series of experiments with a variety of structurally biased scenarios, we demonstrate the effectiveness of Deep-BIAS. We also present the results of using the toolbox on 336 state-of-the-art optimisation algorithms, which showed the presence of various types of structural bias, particularly towards the centre of the objective space or exhibiting discretisation behaviour. The Deep-BIAS method outperforms the BIAS toolbox both in detecting bias and for classifying the type of SB. Furthermore, explanations can be derived using XAI techniques.

A positive phrase or a sentence with an underlying negative motive is usually defined as sarcasm that is widely used in today's social media platforms such as Facebook, Twitter, Reddit, etc. In recent times active users in social media platforms are increasing dramatically which raises the need for an automated NLP-based system that can be utilized in various tasks such as determining market demand, sentiment analysis, threat detection, etc. However, since sarcasm usually implies the opposite meaning and its detection is frequently a challenging issue, data meaning extraction through an NLP-based model becomes more complicated. As a result, there has been a lot of study on sarcasm detection in English over the past several years, and there's been a noticeable improvement and yet sarcasm detection in the Bangla language's state remains the same. In this article, we present a BERT-based system that can achieve 99.60\% while the utilized traditional machine learning algorithms are only capable of achieving 89.93\%. Additionally, we have employed Local Interpretable Model-Agnostic Explanations that introduce explainability to our system. Moreover, we have utilized a newly collected bangla sarcasm dataset, BanglaSarc that was constructed specifically for the evaluation of this study. This dataset consists of fresh records of sarcastic and non-sarcastic comments, the majority of which are acquired from Facebook and YouTube comment sections.

Towards Openness Beyond Open Access: User Journeys through 3 Open AI Collaboratives

Jennifer Ding, Christopher Akiki, Yacine Jernite +2

Open Artificial Intelligence (Open source AI) collaboratives offer alternative pathways for how AI can be developed beyond well-resourced technology companies and who can be a part of the process. To understand how and why they work and what additionality they bring to the landscape, we focus on three such communities, each focused on a different kind of activity around AI: building models (BigScience workshop), tools and ways of working (The Turing Way), and ecosystems (Mozilla Festival's Building Trustworthy AI Working Group). First, we document the community structures that facilitate these distributed, volunteer-led teams, comparing the collaboration styles that drive each group towards their specific goals. Through interviews with community leaders, we map user journeys for how members discover, join, contribute, and participate. Ultimately, this paper aims to highlight the diversity of AI work and workers that have come forth through these collaborations and how they offer a broader practice of openness to the AI space.

One objection to conventional AI ethics is that it slows innovation. This presentation responds by reconfiguring ethics as an innovation accelerator. The critical elements develop from a contrast between Stability AI's Diffusion and OpenAI's Dall-E. By analyzing the divergent values underlying their opposed strategies for development and deployment, five conceptions are identified as common to acceleration ethics. Uncertainty is understood as positive and encouraging, rather than discouraging. Innovation is conceived as intrinsically valuable, instead of worthwhile only as mediated by social effects. AI problems are solved by more AI, not less. Permissions and restrictions governing AI emerge from a decentralized process, instead of a unified authority. The work of ethics is embedded in AI development and application, instead of functioning from outside. Together, these attitudes and practices remake ethics as provoking rather than restraining artificial intelligence.

Fair and trustworthy AI is becoming ever more important in both machine learning and legal domains. One important consequence is that decision makers must seek to guarantee a 'fair', i.e., non-discriminatory, algorithmic decision procedure. However, there are several competing notions of algorithmic fairness that have been shown to be mutually incompatible under realistic factual assumptions. This concerns, for example, the widely used fairness measures of 'calibration within groups' and 'balance for the positive/negative class'. In this paper, we present a novel algorithm (FAir Interpolation Method: FAIM) for continuously interpolating between these three fairness criteria. Thus, an initially unfair prediction can be remedied to, at least partially, meet a desired, weighted combination of the respective fairness conditions. We demonstrate the effectiveness of our algorithm when applied to synthetic data, the COMPAS data set, and a new, real-world data set from the e-commerce sector. Finally, we discuss to what extent FAIM can be harnessed to comply with conflicting legal obligations. The analysis suggests that it may operationalize duties in traditional legal fields, such as credit scoring and criminal justice proceedings, but also for the latest AI regulations put forth in the EU, like the Digital Markets Act and the recently enacted AI Act.

There has been a recent surge in the field of Explainable AI (XAI) which tackles the problem of providing insights into the behavior of black-box machine learning models. Within this field, \textit{feature attribution} encompasses methods which assign relevance scores to input features and visualize them as a heatmap. Designing flexible accelerators for multiple such algorithms is challenging since the hardware mapping of these algorithms has not been studied yet. In this work, we first analyze the dataflow of gradient backpropagation based feature attribution algorithms to determine the resource overhead required over inference. The gradient computation is optimized to minimize the memory overhead. Second, we develop a High-Level Synthesis (HLS) based configurable FPGA design that is targeted for edge devices and supports three feature attribution algorithms. Tile based computation is employed to maximally use on-chip resources while adhering to the resource constraints. Representative CNNs are trained on CIFAR-10 dataset and implemented on multiple Xilinx FPGAs using 16-bit fixed-point precision demonstrating flexibility of our library. Finally, through efficient reuse of allocated hardware resources, our design methodology demonstrates a pathway to repurpose inference accelerators to support feature attribution with minimal overhead, thereby enabling real-time XAI on the edge.

NCVX: A General-Purpose Optimization Solver for Constrained Machine and Deep Learning

Buyun Liang, Tim Mitchell, Ju Sun

Imposing explicit constraints is relatively new but increasingly pressing in deep learning, stimulated by, e.g., trustworthy AI that performs robust optimization over complicated perturbation sets and scientific applications that need to respect physical laws and constraints. However, it can be hard to reliably solve constrained deep learning problems without optimization expertise. The existing deep learning frameworks do not admit constraints. General-purpose optimization packages can handle constraints but do not perform auto-differentiation and have trouble dealing with nonsmoothness. In this paper, we introduce a new software package called NCVX, whose initial release contains the solver PyGRANSO, a PyTorch-enabled general-purpose optimization package for constrained machine/deep learning problems, the first of its kind. NCVX inherits auto-differentiation, GPU acceleration, and tensor variables from PyTorch, and is built on freely available and widely used open-source frameworks. NCVX is available at https://ncvx.org, with detailed documentation and numerous examples from machine/deep learning and other fields.