Can you read minds with data science?🧠
Can you read minds with data science?🧠
Hello!
Last month, Wikipedia celebrated its 20th anniversary. I’m a huge fan of the site and am amazed at how the team and its vast army of contributors have kept it going. It’s funded completely by donations. If you find it useful, consider making a donation.
This newsletter will take you about 5 minutes to read.
I. Spotlight: Can you read minds with data science?🧠
“Our market share declined sharply by 10% last quarter.” screams the headlines of an executive briefing.
As the CMO scrambles to find answers to this challenge, she turns to you, a data leader, for support. What do you think is critical to help plan the CMO’s decisions?
No, it's not the data that you can collect about the market or your competitors.
It's not the analytics techniques you can apply to model the scenarios.
It's not even the storytelling techniques you can use to drive understanding.
Then, what’s most important? Let’s find out by talking about a fun game show.
Mind reading on prime time
Have you heard of the popular game show ‘Twenty Questions’? It was a hit when it was launched on Radio. It ran for over 20 years and sparked many variants across countries.
In this game, a panel of hosts was challenged to identify an object in the listener’s mind by asking 20 questions or less. Each of the questions was answered with a simple ‘yes’ or ‘no.’
It was a classic parlor game of deduction.
What made the show stand out? The panel shocked audiences by guessing people, places, or items, often with just 6 or 7 questions.
One time, the panel did the unthinkable by guessing ‘Brooklyn’ correctly before asking a single question! (They got this by studying the reactions of the studio audience😀)
(Photo by Alessandro Cerino on Unsplash)
The key to winning this game is to ask the most optimized set of questions. It’s an accelerated journey of discovery with deductive reasoning.
You navigate with precision using carefully chosen questions. As you narrow down from the general to the specific, you sharply converge on the answer.
Things aren’t very different when it comes to solving business problems using data science.
What’s the most important thing in data science
Deep at the root of every business problem lie hidden insights. You must pursue a similar journey of discovery to uncover these nuggets of insights.
The first and most crucial step is to ask the right questions to understand the problem. Then, you choose a series of probing questions to keep narrowing down at every step as you apply the data, analytics, and storytelling techniques.
Here’s how it plays out, in sequence:
The questions will determine the data you must collect.
The data will dictate the analytics techniques you must apply.
The analytics findings will influence the visual stories you must create.
The visual stories will, in turn, drive the business decisions you should make.
Ignore #1 above, and your teams will get onto a wild goose chase. They will spend lots of time and money coming up with predictive models and fancy visuals.
But these insights will be useless for decision-making.
Through 2022, only 20% of analytic insights will deliver business outcomes. - Gartner
So, that’s how we must help our CMO solve the challenge of the declining market share. Start by questioning what caused the decline. Was it a product failing in one geography? Follow-up with questions on which competitor took away the business.
Continue this line of questioning until you get to the treasure trove of insights. Then, the business decisions to be made will jump out at you.
II. Industry Roundup:
1. Deep Learning May Detect Breast Cancer Earlier than Radiologists
4 minutes | Health IT Analytics | Jessica Kent
Radiology and diagnosis of diseases is an area that deep learning (DL) continues to disrupt. In a paper published in Nature Medicine, DL has demonstrated the ability to detect breast cancer one to two years earlier. Apart from outperforming radiologists, the model generalized its learning, from Western population to Chinese patients.
2. Dall-E: AI can now ‘read & see’, and its creativity got a boost
5 minutes | Forbes | Rob Toews
OpenAI the organization behind GPT-3 released a new AI model, DALL-E. When given a text caption as input, it produces original images. While it shows high creativity, what’s remarkable is that this is an early example of ‘multi-modal AI’. Unlike models of today that work with only one type of data, say text or images, DALL-E can ‘read and see’. Check out the examples to see why this matters.
3. What are the four levels of human-AI collaboration?
4 minutes | GetSmarter | Thomas Malone
It's far more productive to think of AI-human collaboration as a continuum, rather than a competition. Malone, Professor at MIT lays out a useful framework to plan this collaboration. Today, we are in level-2, still trying to make AI a ‘useful assistant’.
III. From my Desk:
1. Mention among top data science newsletters on Substack - Yay!
A recent article by Analytics India Magazine listed a set of useful substack newsletters on data science. It was good to see “Our Data-driven future” featured in this list. Check out the full list for other interesting newsletters on ML research, business applications, and AI ethics.
2. Why many marketing leaders don’t use analytics for their decisions
The top reason for marketing leaders not using analytics - because the data findings conflicted with the decisions they ‘wanted’ to make. That’s a classic case of confirmation bias at play. This is very common in organizations. Check out this post and the comments for useful tips on how to tackle it.
-> View the Post on LinkedIn (5400 views)
Did you see the viral video of a lawyer who showed up as a cat in a Texas Court’s Zoom meeting?😆 (Thanks to Ginny Hamilton for sharing this clip!)
Thank you for subscribing and reading the newsletter. I appreciate your attention,
Ganes.
PS: Did someone forward this to you? You can subscribe here.
My Website | LinkedIn | Twitter | YouTube