Narrow AI or Generic AI : Which one to build.

Most important quality that helped humans to flourish was the ability to think about future. Other species only think about the present and take a decision based on current environment and circumstances. But a human has the quality of learning from their past, assess the current environment and plan for future based on the information it gathers from the environment and past experiences.

But now scientist are working in the direction to bring same capability in machines. A large number of enterprises and researchers are working to build machines which can think and act like humans. These machines can learn from the past and current knowledge they have and can make decisions based on that knowledge. We every day we hear about advancements in Artificial Intelligence and Cognitive computing. People are curious about Artificial Intelligence and its capabilities but at the same time also afraid of it. News about job losses, generation of machines which can control humans and similar such news are creating panic in the society.

But Artificial Intelligence is a reality now and sooner or later will be an integral part of our daily life (Google maps, Alexa, Google home are live examples). It’s also apparent that AI and other digital technologies like IOT, block-chain, and others will disrupt the way we live and work today. Institutions are also feeling the heat and trying to find ways to accommodate these changes. Leaders in different organizations are quite conscious about rapid development in AI and are keen to adopt these technologies.

Though organizations are keen to adopt Artificial intelligence, lot of preparation and assessment is required before adoption. It requires changing the current way of working, adjusting operations to accommodate AI solutions, understanding and identifying the processes to be replaced by AI agents, understanding how AI will impact the customer experience and most important is to understand the cost of implementing AI and how to monetize it.

Considering the cost and complexity involved in AI implementation, in my opinion, bottom up approach is the right way to start AI journey within any organization. By bottom up, I mean to implement small and independent AI solutions targeting specific problems, instead of creating a full-fledged AI platform to cater all the areas identified for AI implementation.

To build an Intelligent Agent, we need huge data to train Intelligent Agent and Domain knowledge of the specific area which agent will serve. Next step is to develop and use a sophisticated algorithm to create a decent model, using domain knowledge and data. To build a full-fledged Generic AI platform require data from different use cases and very complex algorithm to build and train model for such a diverse data. It’s not an easy task and required a lot of cost and time to design such a system. Even after putting huge cost and effort performance of such generic solutions are mostly not as per expectations. Even after initial training, these systems need lot more time and data to keep on refining the model to bring it close to expectations. 

Instead for a Narrow Agents, a comparatively smaller set of data, and a less complex algorithm is required. It’s easy for organizations to identify a specific area, and arrange resources like data, and domain expertise to build a Narrow Intelligent agent capable only to solve that specific problem. Multiple such systems can be built to work independently and solve specific problems using.

Next step can be to merge some identical systems and create an intelligent agent which is capable to solve a broader problem. And in last after many such iterations, a generic AI system capable of solving all/most of organization’s challenges using can be built. 

These days almost every cloud vendor like AWS, AZURE, and Google are providing services to build such Narrow Intelligent Agents. Organizations can leverage these platforms and services to build their own AI solutions.