You don’t need to code every method you use, but you do need to understand what happens under the hood and know when your data is violating assumptions or visuals are creating inaccurate portrayals.

Otherwise, you risk creating interesting looking results that turn out to be entirely fictional.

AI has made it incredibly easy to throw fancy methods at data without understanding them. We can discover a new model, apply it to data, and generate a pretty looking visualization in a single prompt— but that convenience can hide errors, violations of assumptions, and create false confidence.

It wasn’t about being a programmer. It was about understanding what the method is actually doing — the assumptions it makes, the ways it can break, and what its output really means.

That is what we are seeking to build at #EmpathixAI. We like to say that we aren't really an AI company - we're a science company that happens to have found ways to use AI to do better research, and our goal is to continually build in a way that unlocks new ways of understanding human behavior.

The full potential for #AI in research will not come from automating research tasks - summarization, data tagging, etc. Instead, it will come from applications that use AI to fundamentally rethink what is possible in human research.

Griffith's manifesto was written before advancements in #GenerativeAI led to the explosion of technological innovation of the last few years, but this core argument seems custom built for it.

Instead, we should be using technological advances to fundamentally rethink the ways that we do research to access its full potential.

One of his core arguments was that despite dramatic technological advancements, research methods have changed very little. Instead of finding new methods for understanding human behavior, we've just moved them to different platforms and, maybe, increased the scale a bit.

Very few #AI applications are truly revolutionizing industries or fields. The most exciting applications are.

Online samples performed poorly until they didn’t. Conversely phone surveys performed great until they didn’t. Never get so stuck on a perceived “gold standard” that you’re unwilling to see what’s started working (or what is starting to fail).

The right move isn’t to defend one method as pure—it’s to understand its limits, use it where it makes sense, and, when the stakes are high, see whether other methods confirm the finding.

The sampling space is a great example - researchers get caught up in their "gold standard" method, whether it be in-person, text-to-web, phone, etc. The reality is no sampling method performs well enough to be a gold standard - they're all biased, just in different ways.

When people treat a particular method as the only right way to answer a question they shut down vital parts of inquiry: testing whether findings hold up across approaches and deeply understanding the limitations of any given method.