Select a result to preview

Rossi2018BuildingTrustinAI

Francesca Rossi, "Building trust in artificial intelligence"

Bibliographic info

Rossi, Francesca. "Building trust in artificial intelligence." Journal of international affairs 72.1 (2018): 127-134.

Commentary

This paper focuses on the importance of trust in AI systems to facilitate widespread acceptance regarding their deployment. More specifically, the paper proposes several high-level guiding principles as well as possible, practical implementations to stimulate trust in AI systems.

In her paper Francesca Rossi argues that AI and humans have very complementary capabilities. Even though AI is likely to become pervasive in our everyday life, the rise of such powerful technologies also raise concerns such as their ability to make fair decisions and explain its reasoning and decision-making. As a result of the ongoing discussion between the potential benefits and dangers of AI technologies, Rossi discusses several high-minded guiding principles about AI design, development and usage. Despite such principles providing a useful first step, the paper proceeds to discuss concrete actions that should be taken to implement viable solutions. Such actions include explainability, bias detection, awareness and mitigation, and trusting AI producers. The latter action is acknowledged to be the most debatable one, given that trust can only be gained if companies are transparent about their data usage policies, which is generally a time-consuming process that can quickly lose its progress in case of missteps.

I do not find Rossi's insights to be very ground-breaking, which could be considered a weakness of the paper. Despite this, her plead for strong collaboration between AI developers on the one hand and policy makers and regulators on the other is very sensible. She states that bringing together experts from various fields within the domain of AI, which is already happening in Europe, could also be implemented in the USA for example to ensure a beneficial impact of AI technologies. She concludes by arguing that a holistic, multi-disciplinary and multi-stakeholder approach is the best way to create a comprehensive environment for trustworthy AI.

Excerpts & Key Quotes

Given that Rossi doesn't provide many controversial opinions that are likely to spark serious debate, I have opted for some passages that are informative and capture the essence of her message.

AI Factsheet

'To this end, IBM has proposed the idea of an “AI factsheet”, where developers should record all design decisions and performance properties of the developed AI system, from the bias handling algorithms, to the training datasets, to the explainability tools, etc.'

Comment:

This quote accurately captures the point Rossi is trying to make in the article. It brings together the high-level principles of IBM regarding trust and transparency, and some more concrete implementations. By demanding developers to record all choices made in the design process, explainability rises significantly and a high standard is set for compliance and accountability.

Principles of AI

'The World Economic Forum’s principles for ethical AI: Five principles that cover the purpose of AI, its fairness and intelligibility, data protection, the right for all to exploit AI for their wellbeing, as well as the opposition to autonomous weapons.'

Comment:

What stands out with this quote are two things. First of all, the widespread need for policies regarding ethical AI becomes evident by this quote. The fact that the World Economic Forum, an institute with seemingly little common ground with artificial intelligence, has come up with principles for ethical AI illustrates how important it is. Secondly, the principals are very comprehensive, covering many aspects of the spectrum and highlighting the many dimensions there are to this topic.

Establishing beneficial impact

'Besides hard laws, there are many ways a powerful technology such as AI can be directed toward beneficial impacts, such as standards, the proliferation of best practices and guidelines, and incentives'

Comment:

I want to highlight this quote as it sheds light on an aspect that is often overlooked when talking about ethical AI. Even though hard, concrete laws are necessary, Rossi states that significant change could already be achieved by following these directions. Even though I doubt whether large tech companies will follow guidelines if they're not officially documented, on a smaller scale it may be a good start towards improving ethical implementation of AI technologies.