It is predicted that just as computers and networks have received wide adoption over a period of time, from the days when they were used by only experts, so also will AI spread its wings and become integral to the way we conduct business or lead our lives. With enhancements rapidly taking place in the development of tools available for AI, no longer would AI require high calibre expertise to conceptualise and deploy. Google’s Cloud AutoML or Data Robot are examples of automated tools that have the potential to minimise the dependence on experts and empower the users. As a result of this process, the costs are dropping significantly, making AI’s rapid adoption feasible.
The starting point for AI to be efficient and useful is availability of data sets that could be accessed from large data warehouses and data lakes. A variety of data explorers are now available to help put to use data from open sources. Massive varieties of algorithms available through repositories such as GitHub could be tapped instead of building algorithms from scratch although this still requires expertise in statistics or mathematics in addition to computer science. Avoidance of bias creep into the models and interpretation of the models developed are other areas that will require considerable attention.
Therefore, the success of AI rests not just on the simplification of methods and tools to make it work, but is significantly connected with the availability of reliable data access and analytical models used for interpretation of data. In the years ahead it is anticipated that democratisation of AI will not only be restricted to data, algorithms, storage and models but it will also be important to address the overall value it can create for business or society. Hence parameters that impact the value proposition would also have to be carefully looked into while building AI tools for adoption by wider audiences.
In the process of democratisation of AI, we should also be cautious about the potential pitfalls. AI systems may be built with a lot of sophistication by experts but without suitable training or controls, it would be difficult to pass them on to end users. Even if they are being implemented initially, sometimes the problems may emerge only after they are put to use and by then the damage would already have been done. An example of such a problem has been experienced in some AI supported recruitment tools which discriminate against gender or certain communities. Therefore both the tool designers as well as those implementing them should take adequate care to ensure user profile matches tool functionality and have methods to constantly train and coach the tool as well as the users.
It would become feasible to include subject matter experts in the AI development process which will speed up AI development and also improve accuracy levels. Democratisation of AI will also help address larger societal needs such as healthcare, climate action, disaster predictions and management, global security, agriculture output and personalised learning. In this context, it is also important to rethink the organisation culture. Airbnb is a good example of an enterprise empowering its employees with decision- making authority at multiple levels by making all data accessible to them. Instead of the top-down approach of taking decisions, it is possible to empower the employees with the required intelligence to make decisions. Thus democratisation of decision-making needs to go hand in hand with democratisation of AI if it is to eventually facilitate innovation.
The writer is chairperson, Global Talent Track, a corporate training solutions company