Thank you for Subscribing to CIO Applications Weekly Brief
Human Intelligence before Artificial Intelligence
Luiz Wolf, IT Director, Kraft Heinz Company, Member of Artificial Intelligence, Deep Learning, Machine Learning group
Here at Kraft Heinz, we dare to do better every day. We are in the pursuit of leveraging new technologies and AI-based solutions to better the lives of our employees and consumers. Our purpose, “Let’s Make Life Delicious” is our north star, and through the power of technology, we’re helping feed the world–and we do it deliciously. We’re methodically implementing AI and BI solutions that will transform our company but are doing so in a human-centric approach with proper governance.
Take a simple question such as, “Is today a holiday?”This common question can result in multiple answers and when asking a human, it will be an interactive conversation that will eventually lead to the right answer.On the other hand, while chatting with an automated solution, the individual must seek a solution on their own.
Several scenarios can influence the automated solution’s, e.g., chatbot response:
Am I interested in holidays where I am based on the location of my work profile?
Am I using this information to buy local sales to see if I need to go to work?
Can I use internet-based information or an official company calendar?
Before we think about adding intelligence to a chatbot or other automation, we need to make sure that we know all the scenarios, including the data source itself, and who will keep the information up-to-date.
While chatting with an automated solution, the individual must seek a solution on their own
This “solid” foundation includes, but is not limited to:
• A process for managing organizational knowledge.
• Owners for each rule, knowledge, or article.
• Documentation in natural language and user-friendly diagrams.
• Governance in place to ensure proper end-to-end management.
Without these steps, the chances are that any new AI chatbot, RPA, or automation will give incorrect answers and will never learn from its errors. In addition, any business intelligence (BI) based on the wrong data will provide inaccurate information and therefore lead to poor quality information sharing resulting in potentially bad decisions.
Easy questions such as, "Is today a holiday?” must have a human analysis first to become a rule for the company, and all known scenarios must be documented and maintained by an owner. This owner will obtain AI and BI feedback, therefore approving new rules and improvements in the accuracy of responses and decisions.
AI always starts as a baby and grows with learning and observations, but it needs to have a subject matter expert e.g., owner to guide and govern it. Here is a very simple example:
Scenario: Today is July 4 in China. I ask my AI Chatbot: “Is today a holiday?”
Logic: Go to a holiday calendar and provide the answer.
Bot Answer: Today is a working day, not a holiday.
Question from bot: “Are you happy with my answer?” User: “No, I’m not happy” (I am American, and I expected: “YES, today is a holiday”)
This answer, “No, I'm not happy, ” needs to feed the logic review process to ensure continuous improvement and growth. AI and BI can help to improve the response, but someone overseeing this rule needs to investigate what went wrong and add and/or approve new logic. After, a governance process will ensure that the X percentage of “Yes, I am happy with the answer” is increasing over time.
In this particular example, there is an important note to take into consideration: If the company does not have someone and somewhere with the full company calendar by location and kept up-to-date there is no magic, and the chatbot and AI will give wrong answers and never improve. Again, this emphasizes the criticality of ensuring there is both governance and owner(s).
As a growing baby, Artificial Intelligence is a permanent learning process, so some mistakes will happen, but human owners, not robots, need to be there as parents to fix and improve the “Baby AI-Engine” during this journey.