AI Ethics Brief #108: BigScience Model Governance & Responsible Use, Fair Pricing, System Cards, Sharing Space in ConvAI, and more ...

How do toolkits envision the work of AI Ethics?

Jun 29, 2022

Welcome to another edition of the Montreal AI Ethics Institute’s weekly AI Ethics Brief that will help you keep up with the fast-changing world of AI Ethics! Every week we summarize the best of AI Ethics research and reporting, along with some commentary. More about us at montrealethics.ai/about.

⏰ This week’s Brief is a ~46-minute read.

Support our work through Substack

💖 To keep our content free for everyone, we ask those who can, to support us: become a paying subscriber for as little as $5/month. If you’d prefer to make a one-time donation, visit our donation page. We use this Wikipedia-style tipping model to support the democratization of AI ethics literacy and to ensure we can continue to serve our community.

This week’s overview:

✍️ What we’re thinking:

Social Context of LLMs – the BigScience Approach, Part 4:
Model Governance and Responsible Use
Sharing Space in Conversational AI
Masa’s Review of “Human-Algorithm Collaboration: Achieving Complementarity and Avoiding Unfairness”

🔬 Research summaries:

Understanding technology-induced value change: a pragmatist proposal
A fair pricing model via adversarial learning
System Cards for AI-Based Decision-Making for Public Policy
Seeing Like a Toolkit: How Toolkits Envision the Work of AI Ethics
Human-Algorithm Collaboration: Achieving Complementarity and Avoiding Unfairness
Rewiring What-to-Watch-Next Recommendations to Reduce Radicalization Pathways
Ethical concerns with replacing human relations with humanoid robots: an Ubuntu perspective

📰 Article summaries:

Defending Ukraine: Early Lessons from the Cyber War
Yann LeCun has a bold new vision for the future of AI
Open-source language AI challenges big tech's models

📖 Living Dictionary:

What is the relevance of automation bias to AI ethics?

🌐 From elsewhere on the web:

Microsoft’s new moves in responsible AI
Yann LeCun has a bold new vision for the future of AI
La Habitación CrítIcA Podcast

💡 ICYMI

Energy and Policy Considerations in Deep Learning for NLP

But first, our call-to-action this week:

We’d love to hear from you, our readers, on what recent research papers caught your attention. We’re looking for ones that have been published in journals or as a part of conference proceedings.

Click to share your suggestions.

✍️ What we’re thinking:

Social Context of LLMs – the BigScience Approach, Part 4: Model Governance and Responsible Use

Who needs to be able to examine new technology? And how may we enable accountability of new technical systems along with responsible use?

The BigScience workshop was structured around the development and release of a large multilingual language model. As described in our previous blog post, its driving values include inclusivity, openness, and reproducibility – i.e. we want language communities to be able to examine and explore the model’s behaviors after the end of the workshop without distinction of location or affiliation. It also prioritizes the value of responsibility and building mechanisms to minimize misuses of the models, whether they come from inherently harmful applications or from a misunderstanding of the model capabilities. We built our release strategy around these two pillars.

People: the work produced in this domain is focused on model developers, direct and indirect (or active and passive) users of the trained models, and regulators developing new legislation based on documentation of the model performance and uses.
Ethical focus: The different values that we are aiming to uphold with our release strategy have traditionally been associated with strongly contrasted approaches at either end of the open/controlled release spectrum. While we do not think these values are in opposition to each other, jointly operationalizing them requires further ethical work and grounding choices and mechanisms in specific values.
Legal focus: Similarly to the data aspect, the legal work surrounding the model release has a dual focus. First, we need to be aware of emerging regulations on the use of AI or ML systems that may be relevant to our model. Second, we identify Responsible AI Licensing (RAIL) as a promising approach to govern uses of the model in a way that upholds all of our driving values, and design such a license as a legal tool to assert community control over potential model misuses.
Governance: Finally, collaborative governance of the model depends on having a sufficient understanding of its behavior, capabilities, and failure modes supported by transparent evaluation and extensive documentation. We outline our efforts on both of these aspects in the remainder of this post.

Sharing Space in Conversational AI

Conversations are collaborative. Sharing information from one party to another is a fundamental practice in communication. When it comes to conversational AI, however, people are often limited to a single voice assistant, or agent, for completing a task. As smart as some agents are, it is rare that multiple agents can join forces to help a person during a single conversation.

Interoperability, or creating voice services that are compatible with those of another technological parent, is a significant conversational AI challenge. Imagine planning a vacation and using a voice assistant or bot to complete logistical tasks, but each bot is specialized in something else. Perhaps one can book a flight, but another is needed for reserving a dinner and yet another for travel insurance. Rather than repeating the same credentials to every bot, maybe certain information can be shared with each one during the same conversation. This saves you time and allows each bot to specialize in what it does best.

Such seamless transfer of information is much easier said than done. As voice becomes a primary mode of interaction across devices, it raises the question of how to make conversations more inclusive while maintaining user privacy.

Masa’s Review of “Human-Algorithm Collaboration: Achieving Complementarity and Avoiding Unfairness”

Read the full summary here.

🔬 Research summaries:

Understanding technology-induced value change: a pragmatist proposal

The introduction of new technologies into society may sometimes lead to changes in social and moral values. For example, explainability has been articulated as a new value in response to the opaqueness of machine learning. The article offers a new theoretical account of value change based on philosophical pragmatism.

A fair pricing model via adversarial learning

Sacrificing predictive performance is often viewed as an unacceptable option in machine learning. However, we note that to satisfy a fairness objective, the predictive performance can be reduced too much, especially for generic fair algorithms. Therefore, we have developed a more suitable and practical framework by using autoencoders techniques.

System Cards for AI-Based Decision-Making for Public Policy

AI systems have been increasingly employed to make or assist critical decisions that impact human lives. Minimizing the risks and harms of an AI system requires a careful assessment of its socio-technical context, interpretability and explainability, transparency, nondiscrimination, robustness, and privacy and security. This paper proposes a System Accountability Benchmark, a criteria framework for auditing machine learning-based automated decision systems, and System Cards that visually present the outcomes of such audits.

Seeing Like a Toolkit: How Toolkits Envision the Work of AI Ethics

Numerous toolkits have been developed to support ethical AI development. However, ethical AI toolkits, like all tools, encode assumptions in their design about what the work of “doing ethics” looks like—what work should be done, how, and by whom. We conduct a qualitative analysis of AI ethics toolkits to examine what their creators imagine to be the work of doing ethics, and the gaps that exist between the types of work that the toolkits imagine and support, and the way that the work of ethical AI actually occurs within technology companies and organizations.

Human-Algorithm Collaboration: Achieving Complementarity and Avoiding Unfairness

In many real-world scenarios, humans use AI tools as assistants, while ultimately making the final decision themselves. In this paper, we build a theoretical framework to analyze human-algorithm collaboration, showing when combined systems can have lower error and be more fair – and when they can’t.

Rewiring What-to-Watch-Next Recommendations to Reduce Radicalization Pathways

Recommender systems play a key role on Online Social Platforms (OSP), for enhancing the connectivity between users and available contents. This paper focuses on mitigating the “radicalization pathway” affecting the recommendation output of “what-to-watch-next” algorithms. We define the problem of reducing the prevalence of radicalization pathways by selecting a small number of recommendations to change, in order to minimize the segregation among all radicalized videos.

Ethical concerns with replacing human relations with humanoid robots: an Ubuntu perspective

Would you be comfortable replacing some of your human connections with a robot one? With a focus on humanoid robots, the Ubuntu perspective is harnessed to argue that your answer should be ‘no’.

📰 Article summaries:

Defending Ukraine: Early Lessons from the Cyber War

What happened: An interesting pattern can be noticed when looking back at the wars that have occurred during the last two centuries. Countries wage wars using the latest technology, and the wars themselves accelerate technological change. The war in Ukraine is no different. The Russian invasion includes a cyber strategy that is made up of at least three distinct efforts: (1) Destructive cyberattacks within Ukraine (2) Network penetration and espionage outside Ukraine (3) Cyber influence operations targeting people around the world

Why it matters: As we assess the early strengths and weaknesses of offensive and defensive cyber operations, this report provides a handful of conclusions from the war’s first four months. For example, most countries should be able to disburse and distribute digital operations and data assets across borders in order to defend themselves against a military invasion. Moreover, Ukraine has been able to withstand a high percentage of destructive Russian cyberattacks due to recent advances in cyber threat intelligence and end-point protection, which include AI.
Between the lines: The war in Ukraine highlights several lessons that have a common theme of calling for a coordinated strategy to strengthen defenses against the full range of cyber destructive operations. Four common tenets have been mentioned: (1) Advances in digital technology and AI will be needed to counter cyber threats that are advanced by actors inside and outside the government. (2) Unlike the traditional threats of the past, cyber responses must rely on greater public and private collaboration. (3) There is a need for close and common multilateral collaboration among governments to protect open and democratic societies. (4) Free expression should be upheld and censorship should be avoided in democratic societies.

Yann LeCun has a bold new vision for the future of AI

What happened: Yann LeCun, who is chief scientist at Meta’s AI lab, has drafted an approach that he believes will one day give machines the common sense needed to navigate the world. This could significantly contribute to building artificial general intelligence (AGI), which would have the ability to reason and plan like humans. The main idea involves a neural network that can get rid of the need for pixel-perfect predictions in order to learn how to view the world at different levels of detail. This would allow it to simply focus only on those features in a scene that are relevant to the task at hand. The core network would be paired with the ‘configurator’, which determines what level of detail is required.
Why it matters: “Common sense tells us what events are possible and impossible, and which events are more likely than others.” Teaching common sense to machines is challenging, because neural networks today need to be shown thousands of examples before they start to spot patterns. LeCun has highlighted the benefit of training a neural network to focus only on the relevant aspects of the world. He has also expressed his skepticism around large language models and reinforcement learning being able to build AGI.
Between the lines: This vision is exciting, but there are still many unknowns. The biggest challenge LeCun is facing is that he “does not know how to build what he describes.” For example, one of the main “mystery ingredients” is how to train a neural network to be the configurator. Although there have been various reactions to his proposal, he has gotten praise for asking the right questions and releasing a document that acts as a research proposal with few clear answers.

Open-source language AI challenges big tech's models

What happened: BigScience, a collaboration involving around 1,000 primarily academic volunteers, aims to reduce the harmful outputs of AI language systems. Trained with US$7-million-worth of publicly funded computing time, the BLOOM language model is an attempt to break big tech’s stranglehold on natural-language processing. It has 176 billion parameters, which is on par with GPT-3, and it is also open-source and multilingual.
Why it matters: The researchers who built BLOOM include ethicists, legal scholars and philosophers and employees from Facebook and Google. This is in stark contrast to most natural-language models, which are typically built by small in-house teams. Another notable difference is that most major models use language directly from the web. Instead, the BigScience researchers hand-picked nearly two-thirds of their 341-billion-word data set from 500 diverse sources. Moreover, to train BLOOM, BigScience was granted free access to France’s national Jean Zay supercomputer facility outside Paris.
Between the lines: Although BLOOM is bound to have some biases, the use of multicultural and high-quality sources will certainly improve on existing models. Access to the model is a key step for responsible machine learning. In the future, BLOOM could be used in research outside AI, such as extracting information from collections of historical texts or making classifications in the field of biology.

📖 From our Living Dictionary:

What is the relevance of automation bias to AI ethics?

👇 Learn more about why it matters in AI Ethics via our Living Dictionary.

🌐 From elsewhere on the web:

Microsoft’s new moves in responsible AI

Our founder, Abhishek Gupta, was featured in VentureBeat with his comments on the latest Responsible AI Standard from Microsoft.

One factor that deserves more attention is accounting for the environmental impacts of AI systems, “especially given the work that Microsoft does towards large-scale models,” said Gupta. “My recommendation is to start thinking about environmental considerations as a first-class citizen alongside business and functional considerations in the design, development and deployment of AI systems,” he said.

Read the full article here.

Yann LeCun has a bold new vision for the future of AI

Our founder, Abhishek Gupta, was featured in MIT Technology Review with his comments on the latest paper from Yann LeCun.

There’s another issue, too. If they were to work, LeCun’s ideas would create a powerful technology that could be as transformative as the internet. And yet his proposal doesn’t discuss how his model’s behavior and motivations would be controlled, or who would control them. This is a weird omission, says Abhishek Gupta, the founder of the Montreal AI Ethics Institute and a responsible-AI expert at Boston Consulting Group.

“We should think more about what it takes for AI to function well in a society, and that requires thinking about ethical behavior, amongst other things,” says Gupta.

Read the full article here.

Our Partnerships Manager, Connor Wright, spoke about on the role of diversity and plurality in AI application development teams, among other issues.

Charlamos con Connor Wright, Manager de Relaciones del Instituto de Ética de la IA de Montreal (MAIEI), sobre su actividad profesional y la del MAIEI, el antropomorfismo, el ecosistema startup del espacio AI Ethics, o el papel de la diversidad y la pluralidad en los equipos de desarrollo de aplicaciones de IA, entre otros asuntos.

Spotify: Episodio 8 - Connor Wright: Manager de Relaciones del Instituto de Ética de la IA de Montreal

Ivoox: Episodio 8 - Connor Wright: Manager de Relaciones del Instituto de Ética de la IA de Montreal - La Habitación Crítica - Podcast en iVoox

Apple Podcasts: La Habitación Crítica: Episodio 8 - Connor Wright: Manager de Relaciones del Instituto de Ética de la IA de Montreal on Apple Podcasts

💡 In case you missed it:

Energy and Policy Considerations in Deep Learning for NLP

As we inch towards ever-larger AI models, we have entered an era where achieving state-of-the-art results has become a function of access to huge compute and data infrastructure in addition to fundamental research capabilities. This is leading to inequity and impacting the environment due to high energy consumption in the training of these systems. The paper provides recommendations for the NLP community to alter this antipattern by making energy and policy considerations central to the research process.

Take Action:

Click to share your suggestions.

The AI Ethics Brief

Discussion about this post