This month’s blog theme is ‘Artificial Intelligence.’ I am delighted to see management textbooks are now coming out on Agentic AI ( i.e. Think Rosie the robot maid from The Jestons or Autonomous vehicles). Two years ago you did not see that term in book titles at all. My blog input is to encourage you to read and use those books, but with a certain mindset. I will risk you thinking my blog input is silly – but I mean it sincerely! I have read nearly 30 books on AI, many of them twice, by the most credentialled of scholars and practitioners. But ahead of them all, in my opinion the single most important book for AI users and developers to read and heed is one I read to my children at bedtime.
Parish, Peggy. Amelia Bedelia. Greenwillow Books, (2013). That is just the current publishing date. Ms. Parish’s comical work is over 60 years old.
Yes, it is a children’s book. It will take you 10 minutes or less to read, and the storyline is a hardworking, completely devoted housekeeper, who has skills beyond the imagination, in her case baking pies! But she takes a literal interpretation of instructions to the extreme! Tell her to dust the drapes, and she will throw dust on the drapes! I mean no disrespect or stereotyping – I simply see her as the current level artificial intelligence.
So I want you hardworking, sleep-deprived, superstar AI users and practitioners to think of your AI as ‘Amelia Bedelia.’ Does it want to please you, aka give you an answer? Sure it does, and that is part of the reason we get AI hallucinations. What exactly did you ask it to do? And – drum roll please, what was the training data the AI could have used? Could the training data have biased the output?
Most ‘AI problems’ presentations these days incorporate the paperclip problem. I first read about the maniacal paperclip building AI example in Dr. Nick Bostrom’s Superintelligence. That example is solid, but it is getting to be a bit of cliché. Carefully thinking through what you are asking AI to do, and especially what could possibly go wrong, is never out-of-date. Apply your cybersecurity red team or penetration testing skills, just as you would in looking for weaknesses in your IT operations.
There are many top-rated universities selling online ‘AI Prompting’ programs that will help you become what I call an ‘AI whisperer’ for your organization. Good for them, and quite possibly good for you. But if you are short on the time or the $000s of dollars in tuition, consider at least an ‘Amelia Bedelia’ attitude!
*According to AWS, Agentic AI systems are designed to operate independently, setting goals and executing tasks without constant human input (AWS, 2024). Salesforce further notes that such systems can adapt and learn from their environment, enabling them to handle multi-step, complex workflows (Salesforce, 2024). AP News describes Agentic AI as the next frontier, transforming digital assistants into true collaborators (AP News, 2024).
References:
AWS. (2024). What Is Agentic AI? https://aws.amazon.com/what-is/agentic-ai/
Salesforce. (2024). What Is Agentic AI? https://www.salesforce.com/agentforce/what-is-agentic-ai/
AP News. (2024, March 18). What is agentic AI? The next frontier of artificial intelligence is autonomy. https://apnews.com/article/agentic-ai-autonomous-artificial-intelligence-518d6ae159d1f4d3343e98a456cb5221