ai

The challenge of developing Artificial Intelligence for your organization

By Dr. Amit Bhadra, Dean – WSB

Are you waiting for Artificial Intelligence to mature so that you can sprint your way to leadership in AI as ‘Fast Followers’?

This is a strategy adopted by many Information Technology users.

This may not be a good approach when it comes to AI.

AI requires much more customization than most IT applications. If your AI application is very generic, it may add little value to your business. It could be reduced to some basic automation system.

If your AI application is going to use language it can be even more complex. The learning systems required to use language in AI applications will have to be very sophisticated given that the responses need to be tailored for a very large set of human verbal stimuli.

In order to implement AI the company needs to master the body of knowledge which is capable of making sense of ambiguity.

Hospitals have been working with IBM’s Watson platform to treat certain forms of illnesses. Despite having a vast database to work on in terms of diagnostics and potential prognostic options even hospitals with a vast set of resources in knowledge capital seem to have made little progress.

After AI applications are built it could take a considerable amount of time and effort to integrate it into the larger system.

The fit of the AI application with the IT processes of the organization would require substantial customization. If AI Systems have to be embedded into an existing CRM Application the interfaces would have to be developed if it has to work seamlessly with the larger application.

Leading CRM software Salesforce uses an embedded AI system called Salesforce Einstein.

Einstein learns from data available within the system to deliver predictions and recommendations based on your unique business processes. Paired with automation it delivers insights to truly connect with customers as a knowledge management system working on AI.

While pilots and prototypes can be built relatively quickly, implementation requires re-engineering systems and processes to benefit from AI. For example if AI is created for a customer response system, the organization structure and capabilities of employees would need to be redesigned in order to fully benefit from the AI application.

There are multiple tasks that can be performed by AI Applications in marketing, sales and service relationships. These would have to be conceptualized and prioritized. Human interface systems have to be developed.

Few AI systems are fully automated. Most require human intervention. Determining when the AI application must hand over the task to the human is a big challenge. What stimuli must be assessed in order to make this transition? These decisions are very specific to businesses and it is impossible to bring in a generic AI application and use it in the plug and play mode.

The humans who work with the AI systems would need to develop different skills and capabilities which would have to be of a considerably higher order than the knowledge systems being used by the AI process.

Machine learning occurs through the interactions between a human decision maker and the AI system. For Machine Learning to occur effectively, the interactions need to take place over a sustained period of time. Also called Interaction Learning, the organization must understand how the system needs to interact with its ecosystem.

If the human decision making process has to be baked into a repeatable algorithm, the time taken could be months and years. AI systems using algorithms based on dated human interactions could become obsolete quickly as decision making logic changes with time.

The interaction learning method must be refreshed periodically in order to ensure that AI is making decisions in a way humans would do at a given point of time. Not doing this would lead to AI Knowledge Obsolescence.

AI development and implementation must be a process of building one application at a time, learning from it and moving on.

There are no short cuts. Since it is highly customized it is likely to take time unlike ready to use applications.

Given the fact that learning curves are long, early adopters are likely to build a considerable lead over late adopters.

Early adopters would have built a considerable competitive advantage not to mention cost advantages over followers. As the benefits from AI materialize the followers will find it harder and harder to play catchup.

Companies would do well to look for standard applications ranging from natural language processing to computer vision which can be integrated into their systems and processes with relative ease.

The time to start learning what could be done is now rather than later.