A lot of folks I work with / talk to, struggle with an aspect of GenAI that to me comes down to a metal model shift that is needed. We need to rethink our expectations from computers/systems.
There are 3 things we expect from computers -
Consistency: Traditional programs give the same output for the same input every time.
Predictability: The behavior is predictable and reliable.
Correctness: The output is usually correct, assuming no bugs or errors.
When we look at our bank balance on an ATM machine or multiply 2 numbers or place an order on Amazon, there is no variability from one execution of the task to the other. When we do a Google search twice we get the same search results.
While with Generative AI we get -
Variability: GenAI can give different outputs for the same input. This is because it generates responses based on probabilities and patterns learned from vast amounts of data.
Contextual Understanding: The responses depend on the context and vary so as to provide a more human-like interaction.
Creative and Adaptive: GenAI can produce creative and adaptive responses, making it useful for tasks requiring nuance and flexibility.
For example, I asked GenAI to 'Write a haiku about variability, contextual understanding and creativity'. I got this output.
Patterns shift and change,
Context blooms, insights emerge,
Creativity.
Then I asked the same question again and I got -
Change flows like a stream,
Context breathes new life and form,
Creativity.
What this means is that when working with GenAI we need to understand that variability, contextual understanding and creativity are core to the design of GenAI.
This in turn has practical implications that we need to be aware of when we work with GenAI or use GenAI agents. Any task where you are ok with slight differences in the response from one time to the next, is good for GenAI. Any task where there is only 1 answer, GenAI is probably not the solution.
So what do we do when we want to use GenAI and we want the same answer, tools will need to be developed that handle this variability but it is not native to GenAI.