2 Key things that AI Can’t Do Yet for Your Business

By : |December 24, 2018 0

AI has been adopted by various industries to boost insights, decision making and increase efficiencies. While machine learning and its subset, deep learning have evolved over the years, they still have significant challenges to overcome. A recent survey found that 45 per cent of executives who have yet to invest in AI have a fear of falling behind competitively. Yet, there are multiple challenges and constraints with AI adoption, as a result of which there is expected to be a big divide between industries who are high on the AI adoption maturity versus those that are slow to adopt.

AI needs more labelled data

Srividya Kannan, Founder , Director - Avaali Solutions Pvt Ltd.

Srividya Kannan, Founder , Director – Avaali Solutions Pvt Ltd.

AI needs hundreds of thousands of more labelled data than humans to understand and perform. In many cases, the data is not be available or may require a lot of time and effort for labelling. Where the stakes are high (for example legal repercussions or key revenue related decisions to be taken), users may be keen to know how and why the AI worked to reach its recommendations. The training datasets need to be large and comprehensive for effective machine learning.

Deep learning requires several thousands of records to become good for building models to perform at the level of humans. There may be numerous variables to the associated tasks and for each of these variables, another large data set may be required to be obtained.  There are various developments in this technology that is still work in progress.

                                 

___________________________________________________________________________________________________________

Semi supervised or unsupervised approaches as an alternate to large labelled sets. One shot learning could reduce the need for large data sets with the AI model being able to learn from a relatively smaller number of real-world variations. The various technology advancements are also a part of the reason for slower adoption in some enterprises, since they’re avoiding being early adopters and would prefer to take advantage of such advancements.

Can’t fully trust the recommendations for critical illness

While robots and AI may open a wealth of opportunities to perform our jobs better, it is premature to imagine a world where leadership, decision making, creativity and problem solving may be completely replaced by machines. For instance, in the medical industry, AI is still being met with some level of skepticism given that humans cannot fully trust the recommendations made by a machine (especially for critical illness) coupled with the fact that there is lack of patient data in digital format (patient data is still stored in paper formats). Likewise, in industries where the data infrastructure may still be in physical formats or may need special permissions to access, AI may not work effectively. Again, lack of transparency about what goes into the algorithms and who to hold responsible in case there are mistakes, make it further difficult to trust.

AI is not intended just to replace humans. Viewing AI that way is dangerous. Rather people would use AI to deliver their jobs far more effectively. By offering new tools, it could potentially build new lines of business for entrepreneurs. By 2022, it is predicted that decision making will be the responsibility of hybrid teams comprising of humans and AI systems. Enterprises are still taking a conservative approach to AI given that they’re keen to identify clear use cases where they also have adequate data sets. While they may be experimenting with small projects or POC’s, they recognize that there is no room for complacency. With the right planning and technology, AI could be that golden goose that could bring in immense productivity and prosperity for enterprises.

No Comments so fars

Jump into a conversation

No Comments Yet!

You can be the one to start a conversation.