For new entrants to the profession, data science is a glamorous area promising interesting problems, experiments and advanced machine learning models. However, this video interview tears down the glamour and discusses how data scientists utilise their work hours in practice.
 

For years, public concern about technological risk has focused on the misuse of personal data. But as firms embed more and more artificial intelligence in products and processes, attention is shifting to the potential for bad or biased decisions by algorithms. Inevitably, many governments will feel regulation is essential to protect consumers from that risk.

Apologies for the link error last time - again, here is ‘Where Do We Meet’ - a simple interactive tool built in RShiny showcasing how ‘sf’ for geometric operations and Leaflet for maps can find available outdoor facilities within e.g. 5km or 10km of two addresses.

 

AI has delivered spectacular results in specific applications. Unfortunately, this has led to the ‘AI’ label becoming a marketing tool, and firms making unrealistic promises of model performance that at best, disappoint, and at worst, create biased and discriminatory outcomes.  

In the same vein - for small to medium data, machine learning can often have limited advantages and a simpler model may be not only just as predictive, but more transparent as well. A warning while reading the article; ethics-conscious actuaries should resist the temptation to sell snake oil instead.

Read it here.

Previously the newsletter mentioned GitHub Copilot, which uses AI to automate programming and generate code. This video shows a demo for data science coding by OpenAI of Codex, the model that powers the tool. Based on ‘Plot the results. Label both axes (y axis is max temperature), rotate the x ticks, and add a title’, the model is able to generate working code. Human oversight is still needed, but it appears AI has the potential to improve productivity for data scientists too.

The privacy act is changing – how will this affect your machine learning models and what can you do? This article looks at changes to the Australian privacy act currently under consideration and how to address the potential impacts for industry and consumers.

The continuation of the previous article, exploring the potential to apply data privacy approaches used in traditional settings to machine learning models. It describes the challenge with data privacy, and how “differential privacy” and emerging “unlearning” techniques can help.

Read more.

Looking for bite-sized introductions to AI and data ethics topics? This collection of 11 videos covers topics such as bias, over-reliance on KPIs, and diversity - in clips of less than 15 minutes long.

Watch them here.

For notebook aficionados, Jupyterlab App is now available from the Jupyter organisation - a standalone, self contained app that includes Python, popular libraries and Jupyterlab. For those using Jupyter Notebooks, this may be a significant upgrade. Download links to the installers can be found under ‘Download' here.

Investigating a new dataset that’s completely foreign to you and don’t know where to start? Or wanting to play around with SQL without installing anything?

Try sqliteviz - a simple and intuitive web app that allows you to visualise your data with a wide range of interactable charts within seconds. The data can be read directly from an SQLite database or imported from CSV.