Updates from the HLG-MOS Projects

Generative AI

  • During its first plenary meeting in February, the project AI agreed to create several subtask teams to span across six use cases (co-pilot assistants, Retrieval-Augmented Generation, Information extraction, Generation and editing of metadata, dissemination using AI and communication) and four aspects (project management and development journey; prompt engineering; architecture, systems and applications stack; governance, compliance and ethics)
  • Six sub-task teams kicked off their work in March and April to collect use cases across the members, define the agenda and clarify organizational aspects
  • In order better promote knowledge-sharing, chairs proposed to consolidate the sub-task teams into two bigger task teams, focusing the workplan respectively on technical aspects (prompt sharing, architecture, systems and applications stack) and project management (including governance, risks and ethical aspects)
  • The project is also setting up its collaboration space to gather resources, minutes and other useful resources for the members

The next steps will include the discussion of use cases by members and external presenters.

Statistical Open-Source

  • Inserted Project Manager Carlo Vaccari (April)
  • Added new countries: Netherlands, Ireland, Eurostat, Serbia, USA (Postman), Italy (CREA)
  • Meeting on April 17th: decided organizational stuff (periodicity, sub-teams), analyzed priorities, started discussion on "Awesome list" 
  • Two sub-teams defined:
    • Governance: maintenance, licensing, open source culture, ...
    • Repository: existing repositories, tools choice, recommendations
  • Proposal for joint work with the "Generative AI" project on open tools and open training data

Updates from the Modernization Groups

Blue-skies Thinking




Identifying Topics/Opportunities

CONTINIOUS

Topics that have been discussed during BSTN this year have included the following:

  • Quantum Computing (QC):
    • The discussion primarily revolved around threats that QC may pose to society at large and NSOs due to the potential power of QC to break current encryption methods.
    • The potential for QC to completely disrupt the internet and the sharing of data was noted.
    • The arrival of the QC age could occur within a matter of years.
  • Data Spaces:
    • These are platforms which use cloud infrastructure to share data between data providers and the potential users of their data.
    • Opportunities were noted, such as easier data sharing.
    • Concerns were noted, such as the potential rise of monopolies, and whether the administrators of such spaces are taking into account data interoperability and other issues.
Future of NSIs


  • Several meetings have occurred, including an in person meeting in NY
  • A Strategic Paper will be produced, which will cover the following topics:
    • The history and role of NSOs
    • External challenges such as the data landscape, and structural changes
    • Data stewardship
    • Forecasting the future of NSIs
    • Standards, transparency and trust
  • A statswiki space has been established, and a Google Doc created so that work can begin on the paper.
Digital Twins

IN PROGRESS

No update at the current time.

Applying Data Science and Modern Methods





General Overview

IN PROGRESS

The ADSaMM work-stream has two task teams underway, and both are making good progress and a high level of participation / engagement. Consideration is being given to establishing a third task team.

A 3rd task team is being considered as a consequence of a suggestion from Osama Rahman (UK Data Science Campus) around the practical implementation of data science techniques in official statistics. Following 2 or 3 rounds of meetings, the group of interested NSO are focussed on three Use Cases: PETs, EO / mobile data / Geospatial and Nowcasting. Stats Japan have offered to host a workshop in October on the agreed topic.

Task Team #1: Uncertainty Quantification

IN PROGRESS

The Uncertainty Quantification (UQ) task team are underway and I can report that:

  • The team have met twice, comprises 27 participants, and led by Mohammed Haddou from Statistics Canada.
  • At the kick-off meeting (February), following general introductions etc., the team agreed the scope, deliverables and outline of a work plan. The group also formed smaller groups to tackle subtasks & chapters. In addition, following a task team member’s suggestion the group decided to begin by reviewing and discussing a particular paper on conformal prediction (CP) while simultaneously progressing with our tasks and deliverables.
  • At the team’s second meeting (April), Mohammed gave a presentation on CP, which sparked productive discussions, including the potential application of uncertainty quantification in official statistics.
  • At the forthcoming (May) meetings Mohammed will complete his CP presentation, and Siu-Ming Tam will present a bootstrap method he developed for quantifying uncertainty.
  • The task team is exploring the establishment of a Teams channel to enhance their communication and collaboration.

Task Team #2: Advancing Responsible AI.

IN PROGRESS

The Advancing Responsible AI task team is also underway and I can report that:

  • The team have met twice, comprises 13 participants, and led by Riitta Piela from Statistics Finland.
  • Similar to UQ task team, the team have got into their work with establishing a work program, identified work packages and subtasks that various team members can work on.
  • At forthcoming meetings, various team members will present some ideation thoughts.
  • The task team is also keen to identify a suitable platform so the task team can communicate & collaborate.

Capabilities and Communication


Work and Job of the Future - Extended work on Generic Growth Model

IN PROGRESS

All Task Teams have regular monthly meetings.

Message to the countries to request wider distribution/use of the Generic Growth Model will be sent to the countries together with the surveys prepared by it's sub-teams (employer branding and evaluation of blended working).

Data Analytics

IN PROGRESS

Team had a couple of successful meetings, and prepared presentation detailing Statistics Canada's journey with respect to maturing their HR business intelligence/people analytics function. The plan is to prepare a paper on how to help countries to start on HR business intelligence journey, and lessons learned. 

Evaluation of blended (hybrid working)

IN PROGRESS

Survey on the evaluation of blended working is on the final stages of revision. We plan to send survey to the countries in May.

Employer branding

IN PROGRESS

Task Team prepared draft survey on employer branding, that is under review now. We plan to send survey to the countries in May.

Ethical management (Data and Business)

IN PROGRESS

Work on the Reference Book is progressing. After the Workshop on Ethics four new members joined the Task Team, and we finally have confirmed co-chair of the Task Team (Martin Beaulieu from Statistics Canada)

Ethics Workshop 2024 (26-28 March 2024)

IN PROGRESS

Workshop on Ethics went exceptionally well, and we received 100% positive feedback from the participants. On the last day of the Workshop, discussion focused on the preparation of the Reference Book on Ethics, and we received valuable input form the participants. There is interest from various countries in the reference book and future workshops in this area. Suggestions for future work include: increasing staff engagement in changing organizational culture, social acceptability, statistical data science and responsible AI, ethics and new data sources, ethics in institutional context and ethics in the context of data stewardship.

Workshop on HRMT OC

IN PROGRESS

Information Notice 1 is prepared and circulated to the team members for  their review. Invitation will be sent to the countries at the beginning of May.

AI for official statistics - communication perspective

IN PROGRESS

The task team has regular monthly meetings sharing use cases and needs and challenges. 

    • An interesting aspect of the cases we discuss could be any ethical guidelines on HOW to use AI in the daily work and HOW to COMMUNCATE that we have used AI for a press release, monthly report, social media, etc?
    • Some potential solutions could come with the Artificial Intelligence Act (AIA) EU. At the moment, this is still in the drafting phase: EUR-Lex - 52021PC0206 - EN - EUR-Lex (europa.eu).
    • Researching and presenting if such exist, could be a component part of the cases when presented.
    • Generating images through AI can be challenging, both, from the perspective of writing descriptive prompts and specific requests but also from the perspective of getting the desired images: style, color scheme, organisation templates, etc.
    • BIS, BPS Indonesia, UNECE and Eurostat showcased the practical solutions on writing requests and generating text and images using various tools/combinations (ChatGPT, Adobe Firefly, Monika, Playground, etc.). The presentations will be made available on the  wiki.
    • Different platforms/versions of tools offer different design solutions and products, but also different services as referring to intellectual property and legal coverage/commitment services. Choosing one or another depends on the organisation policy, needs and requirements.
    • The European Commission provides guidelines for using AI under specific conditions, emphasizing caution with words and prompts advising to avoid using personal information and exact location/dates.
    • Elements like text, flags, and dates in AI images (which might be generated in an weird way) can be addressed by asking in the prompts to leave space for customization, and later adding the logos or text according to organizational templates.

Next meeting participants will discuss on the expected output, continuing to share case studies


Supporting Standards

 

CES endorsement of GSIM 2.0


Following discussion of GSIM 2.0 at the CES Bureau meeting in February, a paper for discussion at the CES plenary session (in June) has been drafted.

ModernStats World Workshop


The organization of the ModernStats Word Workshop in October is moving forward. The call for abstracts and information note has been published here. The workshop is going to focus on the use of standards and tools to improve interoperability, transparency and metadata-driven pipelines with the aim at sketching the future of statistical production beyond 2025.

Conference on Smart Metadata for Official Statistics (Paris, 11-12 April)


UNECE contributed to the COSMOS conference on 11-12 April, related to smart metadata. UNECE presented a poster, with InKyung being part of its scientific committee. The topics discussed provide a useful basis for further consideration at the ModernStats World Workshop

There was also a useful meeting on April 10th in Paris right before COSMOS. The morning discussion was focused on ways in which SDMX and DDI are complimentary with each other, and the afternoon was dedicated to learn about transformation and validations languages (VTL/SDTL/SDTH) and how they can be used in the context of automated pipelines. As well as exchanging insights about new developments (SDTL/SDTH and VTL-DDI interoperability), it was a good opportunity for DDI and SDMX experts to share ideas on this topic. There was an agreement to draft a brief note suggesting a favoured approach to take to address interoperability of SDMX and DDI.

Revision of GSBPM and GAMSO activity


Work on the revision of GSBPM is continuing as planned, currently examining feedback received on the Design phase of GSBPM.

SDMX-DDI-GSBPM activity


Work on finalizing the SDMX-DDI-GSBPM report is close to completion, but is proceeding more slowly than anticipated.

Common Statistical Data Architecture


We are still seeking a leader for the activity on Common Statistical Data Architecture (CSDA) – Suggestions are appreciated!

Core Ontology for Official Statistics


This work is not due to start until summer 2024.

65th ISI World Statistics Congress 2025


We submitted a proposal to the 65th ISI World Statistics Congress 2025 in The Hague. The proposed session will discuss how implementation standards can be used together with conceptual ModernStats models to improve interoperability at technical, semantic and organizational levels, and how they can be leveraged to build statistical production pipelines that are metadata-driven, semantically consistent and reusable.




  • No labels