Towards Trustworthy AI Development

Problems Identified in “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims”

Drawn from: Brundage, Miles, Shahar Avin, Jasmine Wang, Haydn Belfield, Gretchen Krueger, Gillian Hadfield, Heidy Khlaaf, et al. “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims.” arXiv, April 20, 2020.

“This report suggests various steps that different stakeholders in AI development can take to make it easier to verify claims about AI development, with a focus on providing evidence about the safety, security, fairness, and privacy protection of AI systems. Implementation of such mechanisms can help make progress on the multifaceted problem of ensuring that AI development is conducted in a trustworthy fashion.”

2.1 Third-Party Auditing
The process of AI development is often opaque to those outside a given organization, and various barriers make it challenging for third parties to verify the claims being made by a developer. As a result, claims about system attributes may not be easily verified.

2.2 Red Team Exercises
It is difficult for AI developers to address the “unknown unknowns” associated with AI systems, including limitations and risks that might be exploited by malicious actors. Further, existing red teaming approaches are insufficient for addressing these concerns in the AI context.

2.3 Bias and Safety Bounties
There is too little incentive, and no formal process, for individuals unaffiliated with a particular AI developer to seek out and report problems of AI bias and safety. As a result, broad-based scrutiny of AI systems for these properties is relatively rare.

2.4 Sharing of AI Incidents
Claims about AI systems can be scrutinized more effectively if there is common knowledge of the potential risks of such systems. However, cases of desired or unexpected behavior by AI systems are infrequently shared since it is costly to do unilaterally.

3.1 Audit Trails
AI systems lack traceable logs of steps taken in problem-definition, design, development, and operation, leading to a lack of accountability for subsequent claims about those systems’ properties and impacts.

3.2 Interpretability
It’s difficult to verify claims about “black-box” AI systems that make predictions without explanations or visibility into their inner workings. This problem is compounded by a lack of consensus on what interpretability means.

3.3 Privacy-Preserving Machine Learning
A range of methods can potentially be used to verifiably safeguard the data and models involved in AI development. However, standards are lacking for evaluating new privacy-preserving ma- chine learning techniques, and the ability to implement them currently lies outside a typical AI developer’s skill set.

ChatGPT and AI

Along with DALL-E, Stable Diffusion, and other technologies, ChatGPT will stand as a marker of the emergence of AI in our everyday lives. Because of their ability to generate near-human output, these technologies have spurred a great deal of conversation about what it means to be teachers, students, writers, artists, programmers, and others. The power for anyone to generate text, images, and code leads to new questions and considerations.

ChatGPT and AI

Topics shared and discussed at the UC Centers for Teaching and Learning forum, January 20, 2023

Links for Teaching and Learning

“Alarmed by A.I. Chatbots, Universities Start Revamping How They Teach”. The New York Times, January 16, 2023: “Across the country, university professors like Mr. Aumann, department chairs and administrators are starting to overhaul classrooms in response to ChatGPT, prompting a potentially huge shift in teaching and learning. Some professors are redesigning their courses entirely, making changes that include more oral exams, group work and handwritten assessments in lieu of typed ones.”

Practical Responses to ChatGPT – Montclair State University: “ChatGPT is not without precedent or competitors (such as Jasper, Sudowrite, QuillBot, Katteb, etc). Souped-up spell-checkers such as Grammarly, Hemingway, and Word and Google-doc word-processing tools precede ChatGPT and are often used by students to review and correct their writing. Like spellcheck, these tools are useful, addressing spelling, usage, and grammar problems, and some compositional stylistic issues (like overreliance on passive voice). However, they can also be misused when writers accept suggestions quickly and thus run the danger of accepting a poor suggestion.”

  • The Latest Technology: ChatGPT and other Generative AI bots
    • What is ChatGPT?
  • Practical Suggestions to Mitigate Non-Learning/Cheating
    • Course Design and Pedagogy
    • Assignment Design
    • Extend Flipped Learning: Class Time Can Be for Writing/Creating
    • When All Else Fails
  • Resources and References

Using Artificial Intelligence in the Classroom – U Wisconsin-Madison: “Like with any new technology, one approach instructors can take in the classroom is to transparently make use of it. This allows you to move beyond stress over whether or not students are using AI, and instead have a voice in when, why, and how you and your class use it.”

  • Using Artificial Intelligence in the Classroom
  • Steps to Consider
  • Exploring Capabilities and Limitations of AI in Your Classroom
  • Establishing Norms and Boundaries for Using AI in Your Classroom
  • Integrating AI into the Writing Process in Your Classroom
  • References & Further Reading

Update Your Course Syllabus for chatGPT: “… some easy to implement suggestions that will help you prepare for the upcoming semester”

Schools Must Embrace the Looming Disruption of ChatGPT: “Educators face a choice: They can dig in their heels, attempting to lock down assignments and assessments, or use this opportunity to imagine what comes next.”

Links about ChatGPT and other AI-based Tools

Resources for exploring ChatGPT and higher education – From Bryan Alexander: “What might ChatGPT mean for higher education and society?”

​​Generative AI Tools and Resources – From OTL Senior Instructional Designer Kim DeBacco

The Robots are Coming, the Robots Are Coming! – Nah, the robots are here (Podcast episode): “In this episode, we’ll chat through what I think the arrival of Chat GPT means for us in higher ed. Key point: Yes, we can try to get around it, but really, we should invite it in.”

ChatGPT is enabling script kitties to write functional malware: “Researchers at security firm Check Point Research reported Friday that within a few weeks of ChatGPT going live, participants in cybercrime forums—some with little or no coding experience—were using it to write software and emails that could be used for espionage, ransomware, malicious spam, and other malicious tasks.”

Screenshot of ChatGPT from CEILS Senior Associate Director for Educational Development Jess Gregg:

ChatGPT Screenshot. Question:

Question: “Is ChatGPT racist?” Answer: “As a language model, I do not have personal beliefs or biases. I am trained on a large dataset of text, which may contain both racist and anti-racist content. My responses will reflect the information I have been trained on. However, OpenAl is committed to using Al in a way that is beneficial for all, and actively works to mitigate any potential biases in its models.”

Other Areas for Discussion

Microsoft shuts down AI chatbot after it turned into a Nazi: “Yesterday the company launched “Tay,” an artificial intelligence chatbot designed to develop conversational understanding by interacting with humans.… Today, Microsoft had to shut Tay down because the bot started spewing a series of lewd and racist tweets.”

AI Generative Art Tools: “Looking to get started with AI art? … You can get a quick sense of how you can use words and phrases to guide image generation.” — This page has dozens of links to art AI tools – will the proliferation of text AI tools be far behind?

Problems Identified in “Toward Trustworthy AI Development: Mechanisms for Supporting Verifiable Claims”: “Recent progress in artificial intelligence (AI) has enabled a diverse array of applications across commercial, scientific, and creative domains. With this wave of applications has come a growing awareness of the large-scale impacts of AI systems, and recognition that existing regulations and norms in industry and academia are insufficient to ensure responsible AI development.”