@ChatGPT, who should be responsible to train you for First Nations languages?
Dr Mahendra Samarawickrama (GAICD, MBA, SMIEEE, ACS(CP))
?? ICT Professional of the Year 2022 | IEEE AI Standards Committee Member | Emerging Technology | AI Governance | AI Strategy and Risk | Technology Foresight | Ethical & Sustainable Technology Advocate | Keynote Speaker
If you are interested in research, development and collaboration on First Nations language and cultural preservation initiatives, please reach out to Dr Mahendra Samarawickrama (GAICD, MBA, SMIEEE, ACS(CP)) ([email protected]). Some of the initiatives and social impact projects focusing on bringing technology equity to the first nations people can be referred here.
Introduction
NLP modes such as ChatGPT do not have any personal opinions and are developed to provide neutral responses based on the data they were trained on. These AI models or numerical representations of language may, however, be subject to human biases. The lack of?diversity, equity and inclusion?in the NLP training data is one such bias in language modelling. In this article, we will discuss?this bias from?First Nations Peoples'?perspective.
We will start with the title question,?who should be responsible for training AI for the First Nations language? Where?the following is ChatGPT's response itself.
"The responsibility for training language models like me to understand and generate text in First Nations languages can be shared by a number of different actors, including:
Ultimately, it is important that there is collaboration and partnership between these various actors to ensure that language models like me are developed and used in a way that is respectful, equitable, and benefits First Nations communities."
Research and Development towards bringing NLP equity to First Nations People
Although natural language processing (NLP) is used in critical real-world applications, they are not standardised or regulated. Underrepresented data is used by technology companies to develop cutting-edge AI that impacts both society and technology. In order to train AI with First Nations Peoples' data, significant research and collaboration are needed.
The loss of language can lead to the loss of cultural heritage, traditions, and values, causing cultural erosion and loss of identity. The use of the native language can also improve mental health and well-being by fostering a sense of belonging and connection to one's cultural heritage. Additionally, the preservation and revitalisation of native languages can improve physical health by passing on traditional knowledge about medicinal plants and treatments. Therefore, the preservation of First Nations languages is essential for maintaining the health, cultural identity, and psychological well-being of the First Nations community.
Training language models with First Nations language data is crucial for making AI capabilities more equitable for First Nations peoples. The uniqueness and diversity of different languages mean that training AI models on a subset of languages do not accurately represent all language groups. By training language models with First Nations language data, the model can better understand and recognise these languages' specific syntax, grammar, and vocabulary, leading to improved accuracy and understanding when processing text in these languages. Suppose the language models do not have enough training data or exposure to the language's specific linguistic features and patterns. In that case, they struggle to interpret and generate text in these languages, leading to errors, inaccuracies and ethical concerns.
Role of the First Nations communities and organisations to integrate the language into AI
Training AI models with First Nations language data requires the development of First Nations language dictionaries, translation tools, and educational resources while preserving the language and its cultural heritage when perpetuated through technology. From the First Nations people's perspective, that must be a collaborative effort between the SME groups, social-justice organisations, technology communities and most importantly, the First Nations people. Practically, this collaboration can be achieved by involving First Nations people in the following steps:
领英推荐
Role of the Government and regulatory bodies
The United Nations has passed several special resolutions recognising the rights of First Nations peoples, including the declaration on the rights of Indigenous Peoples (UNDRIP) in 2007. UNDRIP outlines the rights of indigenous peoples, such as the right to self-determination, the right to their lands, territories, and resources, and cultural identity. These special resolutions provide a framework for governments and corporations to protect the First Nations' rights and culture. UNDRIP recognises that indigenous peoples have the right to participate in decision-making that affects their lives, including designing, developing and deploying AI and emerging technologies. This includes First Nations peoples' rights to shape their digital and AI futures with national interests promoting cultural identity and embracing emerging tech.
Human ethics can support bringing AI equity to First Nations peoples by promoting principles of fairness, respect, and justice in developing and deploying AI and emerging technologies. That is widely discussed in Responsible AI. Ethical considerations can ensure that these technologies are designed and used in ways that respect the rights and dignity of First Nations peoples and promote their well-being. That includes avoiding algorithmic biases, respecting cultural diversity and the right to self-determination, and giving First Nations communities a voice in developing and deploying AI in their territories. Adherence to ethical principles can build trust and foster positive relationships between First Nations communities and the technology industry, leading to more equitable outcomes for First Nations peoples.
The United Nations Sustainable Development Goals (UN SDGs) can be adopted as a comprehensive framework for promoting equity and inclusivity in the development and deployment of AI, including for First Nations people. Since it is the 2030 agenda of United Nations towards sustainability, the goal of bringing tech equity to First Nations people can be easily aligned with the organisational strategy or social responsibility.
In particular, UN SDGs 3 (Good Health and Well-being), 4 (Quality Education), 10 (Reduced Inequalities), 11 (Sustainable Cities and Communities), and 17 (Partnerships for the Goals) all can play important roles in bringing AI equity to First Nations people. UN SDG 3 highlights the importance of mental health and well-being of First Nations people, their risks due to lost of cultural heritage, and how AI and digital transformation can preserve their language, arts and culture to mitigate that risk. UN SDG 4 emphasises the importance of quality education, which is crucial in closing the education gap and promoting cultural preservation among First Nations peoples. UN SDG 10 addresses rectifying inequalities and disparities of First Nations communities by promoting equitable AI and industry 4.0 transformations by addressing the disproportionate data and underrepresented tech-governance leadership and emerging-tech workforce. UN SDG 11 focuses on sustainable AI in cities and communities, which is important in ensuring that First Nation communities have access to and can benefit from AI. Finally, UN SDG 17 emphasises the importance of collaboration and partnership in achieving those SDGs, which is crucial in ensuring that First Nations communities have a voice in and can benefit from the development and use of AI.
Adopting an?integrative framework?of?Responsible AI?can orchestrate the relevant UN SDGs and stakeholders towards equitable AI for First Nations peoples and beyond. That encompasses diversity, equity and inclusion for identifying and mitigating any biases in AI before deploying and for improving data representation. Furthermore, organisational initiatives focusing on UN SDGs can contribute to the value-sensitive design of AI applications.
In such a Responsible AI framework, leaders can focus the key dimensions of
to understand the stakeholders, strategy, social justice and sustainable impact. As shown in the figure, KITE abstraction framework analyses the synergy and social impact of AI from organisational, social and sustainability perspectives. The interdependent perspectives enable the evaluation of motivations for AI ethics and good governance, AI for good, AI for sustainability, and social diversity and inclusion in AI strategies and initiatives. In our experience, this framework enables organisations to systematically engage with the community, volunteers and partners to collaborate towards ethical and sustainable AI for social justice. It hides the application-specific complexities in AI and generalizes the key success factors (KSF) of AI initiatives where stakeholders can easily understand their responsibilities for sustainability and social justice. These key success factors include but are not limited to social DEI (Diversity, Equity and Inclusion), SDGs (sustainable development goals), strategy, ethics and governance in AI. Moreover, this framework supports mitigating AI risks related to biases in various aspects, including bias in data, algorithms, and the people involved in AI. For the complete strategy, please refer to my submission?bit.ly/AIESG?in response to the Australian Government AI regulation consultation process.
Conclusion
Incorporating First Nations people's languages into emerging NLP models and AI applications is crucial for promoting diversity, equity, and inclusion in AI and addressing Responsible AI. In that way, AI-driven NLP can better interpret the unique linguistic features and patterns, leading to improved accuracy in processing and generating text in these languages. This helps to ensure that the AI is not biased towards dominant languages and instead recognises the value and importance of all languages, including those of First Nations people. Additionally, incorporating First Nations languages into NLP models helps to preserve and promote these languages, which are often underrepresented and at risk of being lost. By taking this step towards Responsible AI, leaders, policymakers, regulators, organisations, and communities can help to create a more inclusive and equitable NLP language models where all languages and cultures are valued and respected.
FOptica, FSPIE, CMatP, MAICD, FGIA
1 年A lot of assumptions and assertions are made here to appease popular expression. But in the first instance have you asked any indigenous person (first nation or intermediate) whether they want their language potentially usurped by AI in the same way English, for example, is being usurped? A constant criticism in Australia is the repeated request for involvement, not an imposition. The Tasmanian aborigines, for example, have likely had to put up with more than 60 000 yrs of repeated bullying and displacement that saw them pushed into one of the most remote corners of the world so it's understandable they are a tad sick of it.