SDS 489: Monetizing Machine Learning — with Vin Vashishta

Bias Challenges

Bias in data science models poses significant challenges, as these biases can remain undetected for years, impacting society negatively. Vin Vashishta emphasizes the importance of evidentiary support for models to uncover biases, especially when using massive datasets and opaque modeling techniques 1. He notes that while corporate investment in AI has accelerated research, it also leads to ethical dilemmas, particularly when models intended for social good inadvertently cause harm 2.

We often go in with best intentions on projects, especially social good projects, where we think we're going to improve... but will they end up doing more harm than good?

--- Vin Vashishta

Jon Krohn adds that static datasets can evolve over time, leading to models that may become biased in the future 1.

   

Societal Impact

AI applications hold the potential to benefit society, though most investments focus on financial returns. Jon Krohn highlights examples like machine learning for climate change and medical technologies, which demonstrate AI's societal benefits 3. Despite these positive applications, Vin Vashishta warns of ethical concerns, such as the use of personal data that could lead to privacy invasions 4.

They know about everything you've bought... and if all of a sudden you switch back to chocolate, yeah, maybe your healthcare insurance might be going up.

--- Vin Vashishta

The dilemma lies in balancing the use of data for innovation with the ethical implications for consumers 4.