As artificial intelligence is reshaping our future, the OECD (Organisation for Economic Co-operation and Development) invited different stakeholders to discuss the opportunities and challenges posed by AI on Day 1 of their 2021 International Conference on Artificial Intelligence in Work, Innovation, Productivity and Skills. With representatives from business, labour, civil society and the technical community, the roundtable discussion addressed insights on the gaps and priorities that must drive the debate around the future of AI. 
 

AI is at the core of every technology and offers innovative solutions in areas ranging from health care to business to the arts. The novelty of data driven innovation poses a number of threats that are only just unravelling in front of us. Hence stakeholders have expressed the importance of prioritising human rights, safety and security both in the design and implementation of AI, as well as multi-stakeholder collaboration to navigate bias and self-interest.

Human-centric AI from its design to implementation: 

The primary technical challenge for artificial intelligence is creating human centred AI in the design phase, before the system is set up, as Clara Neppel, representing IEEE and the technology sector, pointed out. While such a development process is complex, she concluded that crucially, the people who are affected by the technology should be consulted and taken into account in this process to find the acceptable trade-off for all parties. The key issue is privacy. AI runs on data; it creates patterns and “learns” based on pre-collected statistics and information. As this data is collected from people, they unavoidably feed and become part of the algorithm.  

In order for privacy to not be disregarded for performance’s sake, it must be implemented responsibly and securely. AI system’s behaviour and output must be continuously monitored and adjusted. Consequently, there is necessity for new metrics to measure the success of technology and AI: expanding beyond the six principles for success, the six P’s, such measures must be reassessed. Profit cannot be the primary indicator. New guidelines for public procurement and increased responsibility of public bodies are key factors to move forward.  

Marc Rosenberg of the Centre for AI and Digital Policy similarly presented the challenge of algorithmic transparency in this process of machine learning. The complexities of AI systems is puzzling even to the very programmers that create it, thus transparency becomes both a technological and democratic issue. He remarked that restrictions by governments are necessary and that AI policy must be addressed, in consideration of the values outlined in the AI principles stated by organisations and associations such as the OECD. Key for upholding democratic values with AI is evidence-based policy formulation. Public participation gives legitimacy to policies and will contribute to AI policies that are fair and sustainable.  

The intrinsic bias of human data directing the algorithm: 

Anna Byhovskaya, representing labour stakeholders and the TUAC, further stressed the challenge of retaining liability, accountability and transparency. Algorithms that are based on biased data will unavoidably reproduce that same bias, which can have dire consequences for example in hiring and firing processes. A company with a history of hiring predominantly white, male staff will create an algorithm that reproduces this reality in future selection processes, and amplify such discrimination. AI and data governance thus have to be brought together and be reassessed accordingly. In addition, firms carry responsibility to accurately inform workers on the systems that they deploy, the extent of their surveillance and the way their data is being used. The employer must also be open to discussing alternatives. Moreover, since the worker adds value to data by contributing to the algorithm, their wage should potentially be adjusted to compensate this contribution. Certification mechanisms with social partners can allow measurement and control of these applications in the working context. 

AI offers practical solutions across sectors that can enhance productivity and efficiency. Nicole Primmer of the OECD once more stressed that AI is a major opportunity in the field of business, boosting competitiveness and innovation. But rather than viewing it as replacing human capabilities, it should be complementing them. AI development must be grounded in AI ethics, positive perception and acceptance of the public. Popular support can be ensured through effective data production and data government frameworks to create trustworthy, human-centric AI.

Next steps to take:

AI has growth potential for businesses and economies, and will be crucial for the recovery after COVID; we have only just scratched the surface of its potential in the areas of health care, environmental issues, value and retail, and similar fields of human advancement. We are not yet aware of the full potential of AI or the relative strength of humans vs. AI, as Andreas Schleicher, Director of Education and Skills at the OECD, indicated. This is to be inquired more scientifically and psychologically in the near future, to ensure that AI will augment human capabilities and not replace them. Thus Schleicher offered the final impulse of the discussion: “if we want to keep the world human centred, we need to keep investing in humans as much as in technology”. Progress cannot be an end in itself.   

by Stina J. Nölken