The future of AI is human-centred

We argue that the future of AI systems is deeply human centred. Rather than replacing human workers, our experience suggests that  AI systems are increasingly being deployed to work alongside people, allowing them to complete tasks more efficiently and effectively. A human centred approach helps increase adoption and impact of new AI tools, while reducing implementation risks and decreasing costs.

Placing humans, including their work practices, ways of understanding their work and their values at the centre of the process is essential for driving successful product adoption, impact, and effectiveness, while helping mitigate a raft of potential risks. By incorporating human-centred design (HCD) principles, organisations can create AI products that are intuitive, user-friendly, and deeply responsive to human requirements, maximising the benefits of those products while mitigating potential implementation challenges.

Recent research published by the Stanford University’s Human-Centred Artificial Intelligence group (Stanford HAI) argues that human-centred AI is essential for developing systems and products that consider broader stakeholder concerns, ethical issues, and border corporate social responsibility issues.  

Deeply understand user needs

A key aspect of an HCD approach is to deeply understand the needs and requirements of the users who will use the system as part of their work practice. This involves actively engaging with users throughout the development process, conducting user research, and collecting feedback to inform the design and functionalities of the application. By gaining insights into the preferences, pain points, and expectations of users, organisations can build AI products that better meet their needs and increase the chance of successful adoption.

Design products that match the way people think

HCD emphasises developing products - including interfaces, interactions and workflows - that reflect the ways users think about their work practice. This involves simplifying complex processes, providing clear and concise instructions, and reducing the amount of time it takes for a user to think about and then complete activities. By designing AI products that are genuinely that are easy to navigate and understand, organisations can ensure that users can effectively utilise the capabilities of AI products, without feeling overwhelmed or frustrated.

For example, for many situations a simple ‘chat’ interface represents a time-consuming and clunky user experience. The performance of many Gen AI systems is tied to the quality of prompting, and the knowledge and expertise required for many everyday users if prohibitive.  We argue that the best experiences of AI products often require little if any human intervention - with the outputs and results presented to a user for review based on automated inputs from a variety of other systems or processes. 

Design for diversity

An HCD approach takes into account the capabilities and limitations of the users interacting with the AI application. Not all users will have the same level of technical expertise or familiarity with AI technologies. Organisations should design AI products that represent a spectrum of user proficiency levels, providing appropriate guidance, support, and user- friendly explanations of the AI system’s functionalities. This approach ensures inclusivity and accessibility for all users.

Test, test, test

An essential aspect of HCD is conducting iterative testing and gathering feedback from users throughout the development process. By involving users in the testing phase, organisations can identify usability issues, understand user preferences, and refine the AI application accordingly. This iterative feedback loop facilitates continuous improvement and helps organisations optimise the application’s functionality, reliability, and user experience.

Bring stakeholders into the design process

HCD encourages multidisciplinary collaboration between various stakeholders, including designers, developers, AI experts, and end-users. This collaborative approach fosters knowledge-sharing, ensuring that diverse perspectives and expertise are considered in the design process. The collective input of different stakeholders helps uncover potential challenges, surface innovative ideas, and ensure that AI products, including those incorporating LLMs, are ethically sound and beneficial for the intended users.

Consider ethical issues during design

We argue for the importance of identifying ethical considerations during application development and deployment of AI products. Organisations must ensure that the AI system respects user privacy, safeguards against biases and discrimination, and adheres to ethical guidelines. By incorporating responsible AI practices, organisations can mitigate potential risks and build trust with users, reinforcing the positive impact of LLMs within a framework that respects human values and rights. HCD approaches to development explicitly aim to surface ethical considerations.

Design for training upfront

Incorporating the HCD approach also entails providing adequate training and support to users who interact with the AI application. Organisations should offer resources, tutorials, and assistance to help users understand and make the most of the AI system’s capabilities. By empowering users and building their confidence in utilising the application, organisations can facilitate user adoption and ensure that users derive maximum benefit from LLMs.

Summary

In summary, we argue that adopting a Human-Centred Design (HCD) approach is essential in building AI products, including those leveraging Large Language Models (LLMs). By deeply understanding user needs, designing intuitive interfaces, addressing user capabilities and limitations, incorporating iterative feedback, encouraging multidisciplinary collaboration, considering ethical considerations, and providing user training and support, organisations can ensure that AI products align with human requirements and maximise the benefits of LLMs. An HCD approach mitigates potential challenges associated with AI implementation, ensuring user satisfaction, acceptance, and effective utilisation of LLM-based AI systems.