Data Science Podcasts

The Turing Podcast, from The Alan Turing Institute, makes the clearest case for data science as something done to the world rather than to a spreadsheet. In Navigating Change: AI in Arctic Sea Ice Forecasting, the hosts explain that ice centres still employ human analysts. Those analysts turn satellite data and model runs into vetted ice charts, the maps people on the ground actually rely on. The episode's claim for IceNet is measured. AI models speed that work up and have been shown to predict better than physics-based models over different time scales, but for the present human expertise remains "still vital." The conversation with Dr. Ellie Bowler of the British Antarctic Survey then widens to indigenous communities whose hunting has followed caribou and whales.

Training_Data was the podcast of IQT CosmiQ Works, a geospatial analytics lab, and it is best heard as a finished archive. In #28 Introducing the SpaceNet 7 Challenge: Multi-Temporal Urban Development, the team admits that assigning unique IDs to 10 million building footprints was "rather daunting." They then explain why tracking an individual building matters more than spotting changed pixels: in theory buildings "don't really move around," and in practice it is trickier. The vocabulary is dense, so newcomers will need patience. #29 Celebrating CosmiQ Works is billed as the final episode, recorded as In-Q-Tel was shutting the lab and stepping back from SpaceNet. Near the end someone invokes a rule: "It's not over until it's over and it's never really over."

Training_Data discussed building datasets. Chai Time Data Science, Sanyam Bhutani's interview show, is about winning with them. Top Kagglers Panel on Best Practises for Training Models is a working session. One panelist argues, against intuition, that a higher learning rate usually means less overfitting. Another, Gabor, defends blending an "army of weaker models" for competitions while conceding it "would be insane to put into production." That candour about the gap between a leaderboard and a live system is the show at its best. The panel speaks fluent Kaggle, so it suits people who already train models.

Plenty of data science happens in meetings, and three shows sit there. Danielle LeBlanc's #DataFemme talks shop with people who make and sell data products. In The People Dimension: Why Data Trust Starts With Us, her guest is a VP at Atacama who grew up in product at IBM. She lists her company's generative features, and says the people aspect is usually the hardest part of a data quality programme because it means new roles and new expectations. Later she tells an anecdote about a garbage truck called a convertible, as a reminder to use such tools with critical thinking. The mood is warm. The guests in recent episodes come from vendors and consultancies, so listen for where enthusiasm shades into pitch.

Great Data Minds publishes recorded webinars. Innovating Under the Government Umbrella with Adam Carpenter earns its time with one story. In a state averaging fewer than six people per square mile, a file for a child in danger once had to be dug out of a district cabinet and shipped. Carpenter describes digitising nine million records for what he thinks was a total project cost of 208,000 and saving 400,000 a year in labour. His procurement rule is blunt: "don't experiment with my tax money please." He also carves out an exception for programmes whose stated purpose is research. On Elephant in the Room: Ethics, Governance + Security with Generative AI, one panelist frames AI's new risk as scale: a minor decision repeated "20 million times by lunch."

go Beyond the Data, hosted by Dave Mathias, is quieter, and its newest episode dates from April 2021. In Ep 43 - Michel Guillet - Leveraging product and data-driven thinking to enhance sales, Guillet weighs the ideal chart against the familiar one a salesperson will actually use, and the host worries that the move toward AI may be happening too quickly. The Enterprise Technology page goes further into this territory.

Openness at Oxford dates from 2015, and some of its references show it, but its argument about verification has not aged. In Simon Benjamin on Open Science, the quantum physicist explains that he hopes his group has released the code behind all its papers from the last year under permissive licences, because finding a bug early beats building five years of science on it: "I actually want other people checking my stuff." The short documentary Open Data: The Golden Age of Discovery makes the same case with maps. It describes readers who emailed "this can't be right" within days, prompting checks and corrections where needed.

Gen.T: A Spotlight For Bright Young People is a general careers show, but Gen.T: Data Science and Urban Development With Leonardo Nicoletti #21 belongs here. Nicoletti describes scoring Vancouver blocks for impermeability to see which areas were vulnerable to flooding. He also describes a tool that finds the best places to plant trees, built for local planners.

Adversarial Learning, with Joel Grus and Andrew, is the counterweight to all that purpose. Adversarial Distancing, Episode 4: Joel's Fizz Buzz Book explains the beginner coding test and recalls Grus's humorous TensorFlow solution. Grus also describes replacing a chapter on heaps after early readers found it confusing. Then the talk drifts to homeschooling and children's play, and whether that drift appeals depends on you. Adversarial Distancing, Episode 3: Two Conversations with Jowanza runs 92 minutes. The guest speaks plainly about racism, including new colleagues assuming he worked in marketing rather than writing code. The Tech Culture page is the natural next stop.