Data science is expanding rapidly but most practitioners skip the basics.
Cassie Kozyrkov (formerly Google’s Chief Decision Scientist) previously shared 32 foundational questions that every data scientist should answer.
I got bored and distilled this into the 10 most helpful questions.
Each question offers multiple valid perspectives ⬇️ **
1/ Start with epistemology.
“What does it mean to know something?”
In data science, do we “know” something when our model predicts it? When p < 0.05? When we have correlations? **
2/ Next: “How do you know that you know?”
For example, if a model has 95% accuracy, what does this really mean? **
3/ Look into critical distinctions:
“What’s the difference between knowledge, truth, assumptions, opinions, and beliefs?” **
4/ The trust equatio…
Data science is expanding rapidly but most practitioners skip the basics.
Cassie Kozyrkov (formerly Google’s Chief Decision Scientist) previously shared 32 foundational questions that every data scientist should answer.
I got bored and distilled this into the 10 most helpful questions.
Each question offers multiple valid perspectives ⬇️ **
1/ Start with epistemology.
“What does it mean to know something?”
In data science, do we “know” something when our model predicts it? When p < 0.05? When we have correlations? **
2/ Next: “How do you know that you know?”
For example, if a model has 95% accuracy, what does this really mean? **
3/ Look into critical distinctions:
“What’s the difference between knowledge, truth, assumptions, opinions, and beliefs?” **
4/ The trust equation:
“What makes evidence trustworthy?” **
5/ The daily dilemma:
“If logic and data point opposite directions, which way do you go?”
I try to avoid data paralysis where possible. **
6/ The measurement trap:
“Does every quantity you’ve measured really exist?” **
7/ The ML paradox:
“Can you make reliable predictions without understanding how something works?” **
8/ The humility check:
“Are analytical insights facts?” **
9/ The ethics question nobody asks:
“Is it ethical to use statistics for persuasion?” **
10/ The builder’s dilemma:
“Is there a difference between knowing something is true and acting as if it’s true? **
Healthcare is a segment where these questions meet reality.
In a way, there is an abundance of data but is largely underutilized (especially due to HIPAA compliance).
For us, @IntegralPrivacy focuses on “how do we unlock what we what already know” instead of just “what can we know?” **
• • •
Missing some Tweet in this thread? You can try to force a refresh