Neelima Misra

AI, Analytics and RPA Lead

How to teach your AI to handle crisis?

2020-apr-14 05:30:13

It is terrifying and exciting at the same time to see how we are dealing the current crisis. An article in WIRED highlights how lessons learned from SARS and H1N1 helped Hong Kong, Taiwan, Japan and South Korea respond to COVID-19 with agility.

No matter how each country has dealt with it or planning to deal with this going forward, few things that are evident from our behavior.

    • We always have more than one option

      • Many countries have chosen an approach to limit the spread by lock downs and avoiding any kind of public gathering, while few other are considering herd immunity.
      •  
    • Our response to crisis is very aligned to our viewpoints, culture and core values

      • To tackle COVID-19, countries and communities have made these choices based many parameters like population, readiness to handle crisis and specific socioeconomic conditions.
      •  

If every country and community has different ways to respond to crisis, it is interesting to analyze how will we to teach AI to handle crisis like this? To answer the same, this article tries understanding human behavior under crisis and explores how that can be applied to AI.

Human brains are designed to process uncertainty based on certain basic principles. A typical classification of crisis is based on the level of knowledge about a crisis event's occurrence (either known or unknown) and the level of knowledge about its impact (either known or unknown).

      • Known–knowns (knowledge)
        • Example- We are aware how COVID-19 transmission happens.  
        •  
      • Unknown–knowns (impact is unknown but existence is known, i.e., untapped knowledge)
        • Example- COVID-19 patients are treated with malaria oral drug called Hydroxychloroquine.
        •  
      • Known–unknowns (risks)
        • Example- More than 60% of the worlds’ population should be infected to develop natural immunity against COVID-19.
        •  
      • Unknown–unknowns (unfathomable uncertainty)
        • Example- COVID-19 accidentally released from a test lab.

 

While known- known and unknown-unknown are less useful, known-unknown can help to make better decisions in uncertain times (in the above example,  we may choose to get infected and develop natural immunity) and unknown-knowns can surprise and delight you with things you’re discovering ( in the above example, we may realize the Hydroxychloroquine administrated with the right amount at the right stage of COVID-19 is able to treat is successfully all the time).

However, while training your AI a right balance is needed while designing uncertainty models.

In future government officials and health care professionals will seek more assistance from AI to deal with this kind of crisis. However, it is important that those AI are designed to operate on dual uncertainty mode. This AI should not only be able handle risks (known-unknowns) and take decisions based on all available data, but also is should be able let humans override/break system flow sometimes learn new things ( unknown- known).  

Tags: AI Big_Data