OVERCOMING THE AI RESISTANCE
The hype around artificial intelligence (AI) has picked up steam in recent years. The introduction of easily-accessible generative AI tools has further fuelled this hype. Many organisations are now investing in AI initiatives aimed at increasing internal efficiencies, market competitiveness, and revenue. For many of these organisations however, the hype is not matching the reality on the ground, with many initiatives stuck at the experimentation stage, and very little real-world deployment of the identified use cases.
A crucial necessity for the practical implementation of these AI use cases is the willingness by end users to adopt the solutions that have been developed. Unfortunately, many organisation do not pay enough attention to this requirement when developing these solutions, leading to unrealised gains and frustrations all round.
There are a number of reasons why end users do not readily adopt these solutions, and they can broadly be placed within three themes:
- I do not understand this
- How is this any better?
- What’s in it for me?
I do not understand this
- Limited awareness and understanding - End users may not be aware of the potential benefits of AI systems, how these systems work, and how they can personally contribute to the development of such systems. If the technology is seen as complex, mysterious, or not easily explainable, users may default to more familiar alternatives and be unenthusiastic about the new solutions.
- Lack of involvement - A lack of input by end users in the development of AI systems spells trouble down the line. Typically, end users are relegated to the most mundane aspects of AI development such as the cleansing and classification or labelling of historical data. By not involving end users in the design aspects of what the system is meant to achieve, those developing such systems forego an early opportunity for buy in. Furthermore, this approach may result in a solution not centred around users’ needs, and potentially one with poor user experience.
How is this any better
- Lack of trust - A key factor that underpins the adoption of an AI solution is trust. Trust can mean many things, however in this context it is referring to trusting that the system is doing the right thing, or making the right decision. If an AI system provides incorrect or unreliable outputs, users are likely to lose confidence in its capabilities. Errors in predictions, recommendations, or decision making can lead to scepticism and rejection. Importantly, those developing such systems should keep in mind that users often expect a level of accuracy that no human being can match.
- Poor user experience - As mentioned earlier, a lack of involvement by end users in the design of the solution can lead to poor user experience. If the AI system is difficult to use, unintuitive, or fails to meet the users' needs effectively, they may reject it. A cumbersome interface, unclear instructions, or a lack of personalisation can also contribute to a negative user experience, leading to rejection.
What's in it for me
- Not direct beneficiaries - It is quite common that ideas for new AI systems originate from outsiders. In this context, outsiders refers to anyone that is not directly involved in the day-to-day process such as executives, business process analysts, or external consultants. Ideas originating from these “outsiders” are often focused on benefits that accrue to the organisation such as greater efficiency or higher profitability, and less so on benefits that accrue directly to the end users that are expected to use the system. In such a scenario, these systems are not readily adopted by end users, leading executives to take the stick approach to enforce usage.
- Fear of job displacement - Perhaps the most important barrier to adoption is the fear that AI and automation technologies have the potential to make end users redundant. Those touting the benefits of AI often state that it will do away with mundane tasks and free employees to do more engaging work. Ask what this more engaging work is however, and you are likely to get blank stares. It is then no wonder that end users are reluctant to fully adopt these systems, as they perceive this as contributing to their own potential job loss.
Approaches to overcome the resistance
There are a number of approaches that business leaders and those developing AI systems can adopt to overcome the end-user resistance. These include, increasing awareness; designing AI systems that are user-centred; reducing the development burden; integrating the AI systems into processes and workflows; and co-designing the new roles.
Increase awareness – The starting point of acceptance is demystifying AI. This does not mean turning front-line employees into data scientists, but rather giving them an appreciation of what AI can and cannot do. Importantly, this should not be a limited view based on the business processes in which the users are involved. Instead, a cross-sectional view should be sought, which also empowers the users to start imagining how AI could potentially solve their day-to-day challenges. A useful approach in this regard would be making users aware of how consumer apps they are already using employ artificial intelligence concepts such as machine learning, computer vision, and natural language processing.
User-centred design – While user-centricity is critical for the take-up of any solution, it is even more important for AI-driven solutions. End users typically have specific challenges with the execution of day-to-day activities. On the other hand, the development of new AI solutions needs additional data attributes to be captured or sourced which adds to the existing burden on users. If this additional data capture is not directly linked to solving the existing user pain points, then it is seen as a peripheral, non-value-adding task. The consequence of this is poor data quality which will affect the AI models, as well as lower productivity by end users who see the additional work as an unnecessary distraction.
It is thus important to deliberately design the new solutions with a focus on how they solve the users current challenges, and not only focus on organisational issues, or those of downstream beneficiaries. Needless to say, it is requirement to involve users in the design of these solutions, and stress how the solutions are meant to make their lives easier.
Reduce the development burden – As already mentioned, new AI solutions may require additional data to be captured. However, this is not the only activity that adds to the burden on end users. Others may include data cleansing activities to ensure that AI models are more accurate, as well as manual labelling of data that is required in the model training process. As much as possible, the development burden should be restricted to activities that are unavoidable.
Data cleansing and broader data management are topics on their own and at this point it suffices to say that manual cleansing by the same users should be avoided where possible. On the issue of data labelling, this should be restricted to instances where specialist knowledge is required. Where possible, externally-sourced and pre-labelled data should be used to develop the base models, then employing internal data to improve model performance over time. Similarly, where pre-trained models exist, they should be used to reduce the initial burden.
Integration into processes and workflows – Underpinning both user-centred design and the reduction of the development burden on end users, is the integration of AI solutions into business processes and workflows. A process that requires a user to log into a separate system to source information or perform part of the process introduces inefficiency and potential non-compliance. As far as is practically possible, the development of these AI solutions should be used as an opportunity to rethink and redesign existing processes and systems for better flow. In scenarios where legacy systems cannot be easily modified, the design should consider the use of new front-ends and data integration.
Co-design new roles – Very rarely do AI driven solutions completely replace jobs. More prevalent, is the automation and replacement of specific tasks within a job specification. This can introduce new problems as machines start to execute tasks much faster and pile up work in front of users, in effect turning end users into process constraints. The knee-jerk reaction may to automate further, resulting in even more resistance. A better approach would be to involve the end users in the design of the new roles, incorporating their input into which tasks should be automated. As mentioned earlier, end users experience specific pain points in their daily activities, and these represent the clearest starting points for automation.
In sync with the redesign of the users’ roles, leaders should rethink the incentive structure. It is likely that the old incentives will not be aligned with the new roles. The new incentives should not only consider how end users will thrive in their new roles, but should also encourage end users to seek out more AI for joint benefit.
Conclusion
In conclusion, overcoming end-user resistance to AI adoption requires a multi-faceted approach. This approach demands that organisations prioritise user understanding, trust, and personal benefits. As leaders and AI developers conceptualise and build new solutions, they need to consider that end users may not understand these systems; may not see how the systems improve on the status quo; or may perceive them as a threat. Ultimately, organisations that address these challenges effectively are better positioned to harness the benefits of AI while fostering a collaborative environment where AI augments human capabilities.