As a mission driven digital research library, Zendy is committed to helping reduce inequality in access to academic literature for audiences worldwide.

By working closely in partnership with publishers and data providers to create an affordable and accessible route to quality, peer reviewed content, we aim to even the playing field of global research, enabling students and researchers from underserved markets to participate in the global academic and policy dialogue and to contribute to identifying solutions to our planet’s major challenges.

We believe that AI has an important role to play in fostering a fairer research and publishing ecosystem. By leveraging AI, we can offer a range of innovative solutions to facilitate access to content and to make the search journey quicker and more efficient, thereby supporting researchers’ needs irrespective of funding or location.

Whilst AI has the potential to transform research and education, Zendy acknowledges that it also raises ethical concerns around issues such as bias, privacy, and accountability. In response to these concerns and to the growing discussion around AI, we have formulated a list of AI imperatives to guide Zendy’s strategy for the development and integration of AI technologies into our products.

Zendy’s AI Imperatives:

1. AI in service of a better world

In line with our mission to reduce inequality in research and publishing, we believe AI can help to positively impact society by facilitating access to and use of academic content. Our portfolio of AI solutions and our ongoing development work reflects this ethos, with a focus on solutions designed to support our users in their search for quality, reliable data. As a signatory to the UN’s Global Publishing Compact, and working hand in hand with our partners, we aim to offer a more enriched, personalised learning and discovery environment for all users globally, with AI very much in service of a better world.

2. Openness, trust and transparency

We explain our use of AI tools, defining them in terms of their abilities and limitations. We clarify how they fit in to our platform’s ecosystem and how users can make the most of them. If the tool makes use of partner copyrighted data, as per licensed agreements, we put in place limitations and clear safeguards on that use. We recognise the need for openness, trust and transparency across all our activities, and we are open to feedback from all of our stakeholders.

3. Personal privacy and data governance

Zendy is a committed advocate of personal privacy. As we continue to develop new solutions and acquire larger data sets, we safeguard the personal information which is shared with us through security policies and procedures to secure our systems. We also ensure that data is collected, reproduced, and protected in a compliant and appropriate manner in accordance with applicable privacy laws and regulations.

4. Equity and unfair bias

Fairness and equity are at the heart of our mission to help create a better world. As we develop and implement new AI tools and solutions, we put in place procedures to ensure reliability, and carry out extensive review. We recognise that biases and hallucinations present challenges in machine learning models, similar to their presence in human cognition. Completely eliminating these issues may compromise the reasoning abilities of our models. Therefore, to minimise these challenges, we have implemented state-of-the-art techniques derived from recent research. These include prevention
and mitigation mechanisms like “chain-of-verification” and “counterfactual reasoning”, which are integrated into our models.

5. Human oversight and accountability

We believe human oversight is key to the successful development and deployment of a useful and reliable AI-powered solution. We maintain human oversight of the development of our AI tools and their output from design to deployment to ensure reliability and quality. We routinely conduct evaluation benchmarks on our models to
assess their tendency to reproduce falsehoods and biases, as well as to measure their accuracy. This ongoing evaluation process enables us to continually enhance and refine our models.