The UNESCO (United Nations Educational, Scientific and Cultural Organization) is a United Nations agency whose purpose is to contribute to peace and security in the world through education, science, culture and communication. In order to achieve its objective, this organisation usually establishes guidelines and recommendations such as the one published on 5 July 2023 entitled 'Open data for AI: what now?'

In the aftermath of the COVID-19 pandemic the UNESCO highlights a number of lessons learned:  

1.  Policy frameworks and data governance models must be developed, supported by sufficient infrastructure, human resources and institutional capacities to address open data challenges, in order to be better prepared for pandemics and other global challenges.
2. The relationship between open data and AI needs to be further specified, including what characteristics of open data are necessary to make it "AI-Ready".
3. A data management, collaboration and sharing policy should be established for research, as well as for government institutions that hold or process health-related data, while ensuring data privacy through anonymisation and anonymisation data privacy should be ensured through anonymisation and anonymisation.
4. Government officials who handle data that are or may become relevant to pandemics may need training to recognise the importance of such data, as well as the imperative to share them.
5. As much high quality data as possible should be collected and collated. The data needs to come from a variety of credible sources, which, however, must also be ethical, i.e. it must not include data sets with biases and harmful content, and it must be collected only with consent and not in a privacy-invasive manner. In addition, pandemics are often rapidly evolving processes, so continuous updating of data is essential.
6. These data characteristics are especially mandatory for improving inadequate AI diagnostic and predictive tools in the future. Efforts are needed to convert the relevant data into a machine-readable format, which implies the preservation of the collected data, i.e. cleaning and labelling.
7. A wide range of pandemic-related data should be opened up,  adhering to the FAIR principles.
8. The target audience for pandemic-related open data includes research and academia, decision-makers in governments, the private sector for the development of relevant products, but also the public, all of whom should be informed about the available data.​
9. Pandemic-related open data initiatives should be institutionalised rather than ad hoc, and should therefore be put in place for future pandemic preparedness. These initiatives should also be inclusive and bring together different types of data producers and users.
10. The beneficial use of pandemic-related data for AI machine learning techniques should also be regulated to prevent misuse for the development of artificial pandemics, i.e. biological weapons, with the help of AI systems.

The UNESCO builds on these lessons learned to establish Recommendations on Open Science by facilitating data sharing, improving reproducibility and transparency, promoting data interoperability and standards, supporting data preservation and long-term access.

As we increasingly recognise the role of Artificial Intelligence (AI), the availability and accessibility of data is more crucial than ever, which is why UNESCO is conducting research in the field of AI to provide knowledge and practical solutions to foster digital transformation and build inclusive knowledge societies.

Open data is the main focus of these recommendations, as it is seen as a prerequisite for planning, decision-making and informed interventions. The report therefore argues that Member States must share data and information, ensuring transparency and accountability, as well as opportunities for anyone to make use of the data.

UNESCO provides a guide that aims to raise awareness of the value of open data and specifies concrete steps that Member States can take to open their data. These are practical, but high-level steps on how to open data, based on existing guidelines. Three phases are distinguished: preparation, data opening and follow-up for re-use and sustainability, and four steps are presented for each phase. 

It is important to note that several of the steps can be carried out simultaneously, i.e. not necessarily consecutively.

1. Develop a data management and sharing policy: A data management and sharing policy is an important prerequisite for opening up data, as such a policy defines the governments' commitment to share data. The Open Data Institute suggests the following elements of an open data policy: 

• A definition of open data, a general statement of principles, an outline of the types of data and references to any relevant legislation, policy or other guidance.
• Governments are encouraged to adhere to the principle "as open as possible, as closed as necessary". If data cannot be opened for legal, privacy or other reasons, e.g. personal or sensitive data, this should be clearly explained. 

In addition, governments should also encourage researchers and the private sector in their countries to develop data management and sharing policies that adhere to the same principles.

2. Collect and collate high quality data: Existing data should be collected and stored in the same repository, e.g. from various government departments where it may have been stored in silos. Data must be accurate and not out of date. Furthermore, data should be comprehensive and should not, for example, neglect minorities or the informal economy. Data on individuals should be disaggregated where relevant, including by income, sex, age, race, ethnicity, migration status, disability and geographic location. 
3. Develop open data capabilities:  These capacities address two groups: 

• For civil servants, it includes understanding the benefits of open data by empowering and enabling the work that comes with open data.
• For potential users, it includes demonstrating the opportunities of open data, such as its re-use, and how to make informed decisions.

4. Prepare data for AI: If data is not only to be used by humans, but can also feed AI systems, it must meet a few more criteria to be AI-ready. 

• The first step in this regard is to prepare the data in a machine-readable format.
• Some formats are more conducive to readability by artificial intelligence systems than others.
• Data must also be cleaned and labelled, which is often time-consuming and therefore costly. 

The success of an AI system depends on the quality of the training data, including its consistency and relevance. The required amount of training data is difficult to know in advance and must be controlled by performance checks. The data should cover all scenarios for which the AI system has been created.

1. Select the datasets to be opened: The first step in opening the data is to decide which datasets are to be opened. The criteria in favour of openness are: 

• If there have been previous requests to open these data
• Whether other governments have opened up this data and whether this has led to beneficial uses of the data. 

Openness of data must not violate national laws, such as data privacy laws. 

2. Open the datasets legally: Before opening the datasets, the relevant government has to specify exactly under which conditions, if any, the data can be used. In publishing the data, governments may choose the license that best suits their objectives, such as the creative Commons and Open. To support the licence selection the European Commission makes available JLA - Compatibility Checkera tool that supports this decision
3. Open the datasets technically: The most common way to open the data is to publish it in electronic format for download on a website, and APIs must be in place for the consumption of this data, either by the government itself or by a third party.

Data should be presented in a format that allows for localisation, accessibility, interoperability and re-use, thus complying with the FAIR principles.

In addition, the data could also be published in a data archive or repository, which should be, according to the UNESCO Recommendation, supported and maintained by a well-established academic institution, learned society, government agency or other non-profit organisation dedicated to the common good that allows for open access, unrestricted distribution, interoperability and long-term digital archiving and preservation. 

4. Create a culture driven by open data: Experience has shown that, in addition to legal and technical openness of data, at least two other things need to be achieved to achieve an open data culture: 

• Government departments are often not used to sharing data and it has been necessary to create a mindset and educate them to this end. 
• Furthermore, data should, if possible, become the exclusive basis for decision-making; in other words, decisions should be based on data. 
• In addition, cultural changes are required on the part of all staff involved, encouraging proactive disclosure of data, which can ensure that data is available even before it is requested. 

1. Support citizen participation: Once the data is open, it must be discoverable by potential users. This requires the development of an advocacy strategy, which may include announcing the openness of the data in open data communities and relevant social media channels. 

Another important activity is early consultation and engagement with potential users, who, in addition to being informed about open data, should be encouraged to use and re-use it and to stay involved.  

2. Supporting international engagement: International partnerships would further enhance the benefits of open data, for example through south-south and north-south collaboration. Particularly important are partnerships that support and build capacity for data reuse, whether using AI or not. 
3. Support beneficial AI participation: Open data offers many opportunities for AI systems. To realise the full potential of data, developers need to be empowered to make use of it and develop AI systems accordingly.  At the same time, the abuse of open data for irresponsible and harmful AI applications must be avoided. A best practice is to keep a public record of what data AI systems have used and how they have used it.
4. Maintain high quality data: A lot of data quickly becomes obsolete. Therefore, datasets need to be updated on a regular basis. The step "Maintain high quality data" turns this guideline into a loop, as it links to the step "Collect and collate high quality data".

These guidelines serve as a call to action by UNESCO on the ethics of artificial intelligence.  Open data is a necessary prerequisite for monitoring and achieving sustainable development monitoring and achieving sustainable development. 

Due to the magnitude of the tasks, governments must not only embrace open data, but also create favourable conditions for beneficial AI engagement that creates new insights from open data for evidence-based decision-making. 

If UNESCO Member States follow these guidelines and open their data in a sustainable way, build capacity, as well as a culture driven by open data, we can achieve a world where data is not only more ethical, but where applications on this data are more accurate and beneficial to humanity.

References

https://www.unesco.org/en/articles/open-data-ai-what-now

Author : Ziesche, Soenke , ISBN : 978-92-3-100600-5

_____________________________________________________________________________________

Content prepared by Mayte Toscano, Senior Consultant in Data Economy Technologies.

The contents and points of view reflected in this publication are the sole responsibility of its author.