top of page

When To AI

Writer's picture: Amelia WatersAmelia Waters

Updated: Jan 27

Recently a friend from my BCG days, currently a professor of innovation at Stanford University, made a compelling case for why people should be using AI whenever and wherever they can. In his perspective, AI augments human capabilities.


In theory, I agree. However, in practice, AI performance is jagged at best. Fantastic in some things; definitively below average in others.


As a management consultant whose days are full, I'm always looking for ways to augment what I can do and increase my output. Over time, I've developed a 5-question framework for "When to AI" when delivering outcomes (e.g., reports, analyses, presentations, etc)


Important note: This is not a framework for using AI to help you learn. I've a different framework for that.


The WHEN-TO-AI Framework




How does this work in practice?


I don't use AI to write any communications whose audience matters. I'm a good writer. The words come easily, and in my unique tone. Even when I train AI to mimic my style, I find myself having to edit extensively--which takes far more time than just writing it correctly in the first place.


I don't use AI to run analyses on client data. Leaving aside any possible data leak or confidentiality issues, I've not yet found an AI tool with the ability to critically scrutinize not-clean data, make executive decisions on how to clean and segment the data, and then run the analysis to draw out the insights. The challenge with insight-seeking data analysis is that it's more akin to playing detective. The steps are sequential, but they're also iterative, and demands intuition that comes from experience. I need an AI tool that is capable of thinking, "This doesn't pass the sniff test...." and I haven't found that tool yet.

On the other hand, I use AI for image creation, e.g. for the avatars that I use in my online gaming. Only my alliance sees it, and no one objects to cute kitten pictures that aren't real-life photography.


I use AI for web-based research, and even then, I lean on tools like Perplexity that will provide me with the sources rather than the black hole of other LLMs. When I use Perplexity, I check all its sources and curate it extensively. I know what truth and quality look like, and I can validate Perplexity's answers quickly.


In short, "When-To-AI" is highly contextual and based as much on your personal capabilities as on AI's capabilities. It's important to know when AI can augment your skills vs when it's just a time-suck (for now, until it gets better.) And always, keep experimenting and testing for the best AI tools.


My Favorite AI Hack


One of my favorite hacks is to use the exact same prompt on multiple LLMs and to compare the results. It's an excellent test of which LLM is better able to deliver on the "truth" and "quality" elements, and in a tone that minimizes editing should I choose to use it.



Let me know what else is worth checking out! I'm always on the lookout for new tools to augment my research and thinking.

16 views0 comments

Recent Posts

See All

Comments


© 2023 by EDSO Edge, LLC. Powered and secured by Wix

bottom of page