AI/ML Study Group Meeting

Meeting Agenda

Date: 6 March, 2024 at 2.30 pm (ET)

Meeting Chairs: Ma'n H Zawati, Rosalyn Ryan (regrets)

Zoom meeting ID: 980 1743 4429

Zoom link: https://zoom.us/j/98017434429

Passcode: GA4GH

Zoom Recording: zoom

Passcode: 3H4E=wwz

Attendees:

1. Ma'n H Zawati
2. Diya Uberoi
3. Marc Fiume
4. Gemma Brown
5. Jessica Seegobin
6. Donrich Thaldar
7. Dianne Nicol
8. Emma Kondrup
9. Francis P Crawley
10. Serena Kher
11. Sandrine Richard
12. Jordan Lerner Ellis

Minutes:

Welcome & Introductions

MZ  welcomed everyone, provided a brief history of the workings behind this group and invited all attendees to introduce themselves and their interest in AI. Everyone went around the room and introduced themselves.

Group Description & Aims

MZ set out the aims of the group. He spoke about the way some groups tend to discuss AI generally and how literature continues to focus on the ethical and legal implications of AI. The question therefore, is how to now maximize on what has been done, to develop a practical tool that could be useful for the GA4GH community. One possible application that would benefit from guidance of this working group is the development of a conversational AI/generative AI tool, which if queried in English could access multiple data sets and get answers. MZ explained how these applications are going to become more popular with time and that given the ethical challenges they pose, it will become important that developers have guidelines that they can use to develop more trustworthy tools. Any guidance point however, should still be implementable by developers and show to work in practice. This group will thus aim to put this forward by 1. Developing guidance points following literature review, normative document analysis and discussions, and  2. using a case study to expand these points to create a tool that will be beneficial to the GA4GH community.

MZ then turned the discussion over to MF, to give a description of the practical applications of conversational AI. MF spoke about initiatives that DNA Stack has been working on to facilitate the development of federated GA4GH compliant networks. In short, this allows organizations around the world to co-analyze data without having to move the data from one place. He discussed many of the projects that DNA stack has been involved in. These include Viral AI and Neuroscience AI. IPCHiP is another driver project, connecting pediatric hospitals around the world. He highlighted the potential of conversational AI tools to query across federated networks through human language instead of machine language or other APIs and the way they work technically.

This, however, presents a slough of ethical and legal implications. Technology can be built, but guidance on the way these technologies can better account for security, and privacy (among other issues) will be necessary.

Discussion

MZ  therefore opened up the discussion to understand the scope of such a tool and how we could ensure more trustworthy AI. DN suggested that scope and limits should guide this development. Should we think about liability issues? Should we think about accountability? Should we think about privacy concerns and misuse? A range of factors can influence the impact that AI will have.

DT  stated one way to define the scope of the guidance document and the tool is by focusing on privacy and other harms caused by the implementation of AI.  Lack of Transparency drives many of these concerns, so transparency will be important to the development of more ethical AI tools.  DT, FC, SR, & JLE highlighted the importance of developing standards on transparency and disclosure which could help set limits on the data that can be accessed and shared by a potential AI tool.

Liability could also be a factor that can guide the development of guidance points. DT & MZ highlighted the need to address liability issues for developers and users.

Lack of trust however, can be a barrier to data sharing and it is important for AI tools to be trustworthy. Sectors use these tools differently and the impact they can have on practices will differ, so another possibility would be to consider the use of the tool by sectors.

Regardless, an understanding of the law and its impact on AI will be necessary. MZ therefore highlighted the importance of understanding how this topic is being addressed in different jurisdictions and the need to understand the state of legislation in countries.

Beyond law, collaborations will also be crucial to taking this work forward. FC suggested reaching out to WGs such as BGI, Schengen National Gene Bank, Life of Sciences, and Pistoia Alliance to expand the reach of the groups work.

MZ highlighted the importance of ensuring that the output is feasible, practical and of direct importance with the mandate of GA4GH and in service of its larger community.

The meeting ended on a very high note, with everyone commending the importance of this project and its novelty.

Key Takeaways:

• In a world where AI is expanding and science has revolutionized policies, the time has come for guidelines to work in tandem with the development of more trustworthy AI tools.
• This group offers the opportunity to close the gap, using guidance points and a pilot study to guide the development of a conversational AI tool.

Next Steps:

• Literature review and qualitative document analysis on policies, frameworks, code of conduct etc. in different jurisdictions.
• Development of an excel sheet to track  policies and legislation across jurisdictions.
• Next meeting will take place in 2 months approx. - details to be sent..