The discussion was attended by a member of the Parliament of South Africa, representatives of UNESCO (moderator), OpenAI, EY, Kaspersky, the African Institute of Science and Technology of Tanzania.
Digital technologies have revolutionized communication in the online space. However, the volume of misinformation on the web has increased dramatically, calling into question the authenticity and reliability of the data consumed. The latest technologies, such as artificial intelligence (AI), also increase the threat of misinformation and false information, allowing for the scaling up and widespread dissemination of unverified or false information. Today, governments may face difficulties in combating disinformation on the Internet if cooperation is not ensured at all levels between various stakeholders, including parliamentarians.
The following issues were discussed at the session:
-
How can the political, regulatory and legislative frameworks promote cooperation in the field of AI development based on trust, while continuing to stimulate innovation?
-
How to ensure that the development and use of AI technologies are safe, fair, reliable, transparent, explicable and trustworthy?
-
How to strengthen cooperation between governments, industry, experts and civil society to develop and implement norms and guidelines on the responsible use of AI?
-
How to ensure that AI complies with universal human rights and values with the widespread dissemination of technology in society, in economics and politics?
The participants noted that it is impossible to deprive any regions of the world or any sectors of the economy of the opportunity to use various tools and values arising from technology.
It was also noted that governance should not create artificial obstacles for AI developers, since it is necessary to continue to encourage a creative approach to business. However, regulatory practices should create additional incentives for companies involved in the development and implementation of AI. Governance should take into account the total amount of data in the processing loop and, in particular, cover multi-purpose AI systems trained on a large amount of data from open sources, as well as systems that are used in areas where risks are high and a high level of responsibility is required. In addition, statutory instruments should reflect industry requirements for AI systems, since each application area (use environment) has its own characteristics. The opinion was expressed that it is values that will lead us to a balanced governance of AI.
The representative of South Africa said that in his country, parliamentarians rely primarily on the Constitution and the law on data protection, which protects the personal information of citizens.
The speaker from EY company noted that when we approach governance, we do not always understand what people really want. So, for example, we can consider the level of privacy that we provide to our mobile phone. The same data sets can be used in an artificial intelligence system and we won't even notice it. He suggested that an understanding of technology should underlie governance as such.
The session participants answered questions from the audience - members of parliaments from various countries. In particular, a question was raised from the audience about how an electronic agricultural tool with AI support can benefit a farmer in Ghana if the databases used to train this AI tool are adapted for Texas? Answering the question, the representative of OpenAI said that large language models are trained on the part of the Internet that is mainly English-speaking, on which the highest quality initial training was conducted. They do not reflect the languages of the whole world to the extent that they should. And there are really significant performance gaps, but models are evolving and this is what developers are striving for. An example of the Pilate Bronco project for the Government of Iceland was given. The Icelandic language and its representation on the Internet are very small, the number of speakers is very small, and to make sure that the language can be used by AI-based tools, to find out which research methods will overcome some of these gaps is a big research task.
The representative of the EY also explained that the goal is to develop tools for humanity, and not for a specific region of the world. This is what we should strive for - to turn AI tools into a common depository of human knowledge.
