Use The “Small Tasks” Approach to Get More Out of AI Tools
Small tasks are easier to supervise than large tasks because humans have a limited amount of attention with which to process sensory inputs. We’ve adapted to tune out things that aren’t important, and when we’re faced with lots of information, we’re more likely to — subconsciously — decide that most of it is not important. When faced with an AI output that is large or verbose, this adaptive disinterest can make it easy for errors to slip by unnoticed.
We can ameliorate this risk of unnoticed errors in AI practice by reducing the scope of a task delegated to an AI.
For example, consider a recurring update describing progress on work tasks. It is possible, certainly, to prompt the AI to describe the full scope of the update, and to then edit or review the output to ensure that it’s aligned with the expected content and that it is accurate. This requires writing a prompt, corralling all the relevant information to be included, and finally editing the output: three distinct activities. While this might be easier in some ways than writing out the update, it is unlikely to be faster, and it requires the additional efforts of prompting and review.
What if, instead, the prompt was reduced to ask only for a template for the message? If the AI produces a valid template, which seems likely given the prevalence of status update examples throughout the internet, then the only task remaining for the human is filling out the template. This is much easier than creating it, and the data source — the human — is already known to be likely to be reliable. This means that the review process can be much shorter and the task is complete that much faster. And, if the template produced works well, it can be used again for the next instance of this particular recurring update message.
Reducing the scope of the task given to the AI makes it easier to complete the task faster with a high degree of confidence. It’s easier for a human to catch errors in a tightly scoped task than in a larger task, and it’s faster to combine information from a trusted data source with questions produced by an AI than it is to verify all the information an AI offers in response to questions. Using an AI to produce small, incremental gains like this creates compound acceleration and can increase work quality. If you’d like to learn more about adopting AI, visit https://mockingbird.cc.
