On June 6, 2023, National Association of Software and Services Companies (“Nasscom”), a non-profit trade body and chamber of commerce of the Tech industry in India with over 3000 member-companies released a self governance framework in relation to generative artificial intelligence (“GenAI”). The set of guidelines (“Guidelines”) aims to act as common standards and best practices for researching, using, developing GenAI, to foster trust and mitigate the potential harms from the adoption of GenAI across the industries.
As per the Guidelines, GenAI has been defined as a type of artificial intelligence which,”…can create artefacts such as image, text, audio, video, and various forms of multi-modal content” based on a variety of inputs. The Guidelines have been divided into three mutually inclusive categories which means a stakeholder (including but not limited to tech companies, startups, open source developers and researchers) could fit into all the following three categories:-
-
- Research of GenAI: as the name suggests means fundamental and applied research on GenAI models, applications, tools and techniques. The stakeholders for this category would be researchers and their obligations are as follows:-
- Researchers shall systematically anticipate as well as evaluate both contingencies that might arise from the conduct of research;
- They shall release public disclosures containing values, goals, motivation for driving or funding a research project, methodologies, model training datasets and tools adopted for the conduct of such research;
- Researchers shall adhere to established privacy-preserving standards in research data collection, processing and usage and conducting safety testing GenAI models in regulated environments.
- Researchers shall account for the risk of harmful bias in research and deploy protocols and measures to mitigate it. They shall also (i) publish research findings in open source format (ii) foster collective inquiry into potential risks and benefits from the adoption of GenAI and (iii) develop prevalent societal values in GenAI.
- Researchers shall also prioritise research on GenAI applications, tools and techniques to improve technical AI safety to mitigate the potential harms associated with using, researching and developing GenAI.
- Development of GenAI: includes building and deploying GenAI models and sourcing training data for GenAI model design and development to design/power various commercial, non-commercial or personal applications. The developers of GenAI shall comply with the following obligations:-
- Developers shall abide by the guiding frameworks such as Naascom’s Responsible AI Governance Framework[1], Niti Aayog’s Approach for Operationalizing Principles for Responsible AI[2], UNESCO’s Recommendation on the Ethics of AI[3], OECD AI Principles[4].
- They shall retain qualified personnel such as ethics councils, risk and compliance officers as well as prescribe terms of service guidelines for the responsible use of GenAl.
- They shall publicly disclose the data and algorithm sources used for modelling and all other technical, non-proprietary information about GenAI solution’s development process, capabilities and limitations..
- Developers shall also practise due-diligence while developing and deploying GenAI solution. They shall strictly adhere to (a) applicable data protection and intellectual property laws in collection, processing and usage model training data (b) industry best practices for model design, development, deployment and monitoring (c) deployment of best-in-class cybersecurity infrastructure.
- Developers shall devise technical means to provide explanations for outputs generated by GenAI in high-stake contexts in an accessible and intelligible form. They shall also establish and adhere to fair and equitable schemes for (a) Liability apportionment (b) Mechanisms for grievance redressal in relation with mishaps caused by use, development and deployment of GenAI.
- Developers shall develop, deploy and use GenAI that prioritise energy efficiency in model training and development, disclose the source code, wherever possible, to reduce duplication of efforts.
- Use of GenAI:entails employing/modifying GenAI models for product or service development, enhancement of delivery, conducting scientific or academic research, writing computer code, generating creative outputs (e.g., artworks, musical compositions, product designs) for any commercial, personal or non-commercial purpose. Obligations for the users of GenAI are as follows:-
- In addition to the Developer’s obligations under point 2(a), the users of GenAI shall invest in workers upskilling and reskilling programs across the business roles and functions to balance out substitution, if any of the current workforce, induced by GenAI solution.
- Users of GenAI shall publicly disclose all technical, non-proprietary information about GenAI solution’s development process, capabilities and limitations of the downstream models and applications including but not limited to details of modifications performed on the models, corresponding data and algorithm sources.
- The Users shall also disclose use of GenAI for (i) product or service delivery/enhancement; (ii) deliverables generated within academic or commercial settings;(iii) produce the research outputs; (iv) claims in relation to intellectual property.
- Users shall use GenAI in compliance with the terms of service, prescribed guidance and applicable public regulations for safe and responsible use of such GenAI solutions.
- They shall also deploy and develop downstream models and applications in compliance with best industry practices such as feeding contextual awareness in model design, develop applications by diverse and multidisciplinary teams, phased deployment of large scale consumer facing applications, human in the loop design and deploy applications with guidance for responsible use, fair and lawful interpretation of results generated by GenAI.
- Users shall not use the GenAI solution to (a) infringe or misappropriate the rights of others; (b) to propagate disinformation or harmful social, economic or political biases.
- Users shall also exercise caution and due diligence in consuming/using content or media generated by GenAI by (a) refraining from sharing any sensitive, confidential or personally identifiable information in course of interaction with the GenAI solution; (b) instate corporate and institutional safeguards and codes of conduct to counter misuse or unauthorised use of GenAI solution.
- Akin to the Developers, the Users of GenAI shall also devise technical means to provide explanations for outputs generated by GenAI in high-stake contexts in an accessible and intelligible form. They shall also establish and adhere to fair and equitable schemes for (a) Liability apportionment (b) Mechanisms for grievance redressal in relation with mishaps caused by use, development and deployment of GenAI.
- Users shall use GenAI in alignment that prioritise energy efficiency in model redesigning and use, by disclosing the solution source code, wherever possible, to reduce duplication of efforts.
- Research of GenAI: as the name suggests means fundamental and applied research on GenAI models, applications, tools and techniques. The stakeholders for this category would be researchers and their obligations are as follows:-
In addition to the above, these Guidelines also sets out joint obligations of researchers, developers and users which entails conducting (a) universal AI literacy and awareness programs (b) technical AI safety research and (c) regulatory reform projects for trustworthy adoption of GenAI for benefit of all stakeholders, in collaboration with government and non-government organisations.
While these Guidelines broadly covers issues and potential harms related to GenAI it would be interesting to witness its applicability by Nasscom’s members, non-members and public at large, especially when the Guidelines are silent on monitoring mechanisms, penalty for non-compliance and/or legal effect, if any.
[1] https://indiaai.gov.in/responsible-ai/goverance-framework
[2] https://www.niti.gov.in/sites/default/files/2021-08/Part2-Responsible-AI-12082021.pdf
[3] https://unesdoc.unesco.org/ark:/48223/pf0000381137/PDF/381137eng.pdf.multi
[4] https://oecd.ai/en/ai-principles