How to Navigate AI’s Growing Social Footprint

We already live in a world shaped by powerful algorithmic systems, and our ability to navigate them effectively is, at best, shaky—very often through no fault of our own.

We may want to think, like Spider-Man’s Uncle Ben, that with great power comes great responsibility; in the real, non-comic-book world, the two don’t always arrive simultaneously. The companies driving most AI innovation often rush to release products even when the latter have the potential to disrupt lives, careers, and economies and to perpetuate harmful stereotypes; responsible deployment isn’t always their creators’ top priority.

To help us survey the current state of affairs—risks, limitations, and potential future directions—we’ve put together a strong lineup of recent articles that tackle the topic of AI’s social footprint. From medical use cases to built-in biases, these posts are great conversation-starters, and might be especially helpful for practitioners who have only recently started to consider these questions.

• Gender Bias in AI (International Women’s Day Edition)
  In a well-timed post, published on International Women’s Day last week, Yennie Jun offers a panoramic snapshot of the current state of research into gender bias in large language models, and how this issue relates to other problems and potential blind spots lurking under LLMs’ shiny veneer.
• Is AI Fair in Love (and War)?
  Focusing on a different vector of bias—race and ethnicity—Jeremy Neiman shares findings from his recent experiments with GPT-3.5 and GPT-4, tasking the models with generating dating profiles and playing matchmaker, and revealing varying degrees of racial bias along the way.
• Seeing Our Reflection in LLMs
  To what extent should LLMs reflect reality as it currently is, warts and all? Should it embellish history and current social structures to minimize bias in its representations? Stephanie Kirmer invites us to reflect on these difficult questions in the wake of Google’s multimodal model Gemini generating questionable outputs, like racially diverse Nazi soldiers.

• Emotions-in-the-loop
  Invoking a near future where the line between sci-fi and reality is blurrier than ever, Tea Mustać wonders what life would look like for a “scanned” person, and what legal and ethical frameworks we need to put in place: “when it comes to drawing lines and deciding what can or cannot and what should or should not be tolerated, the clock for making these decisions is slowly but steadily ticking.”
• ChatGPT Is Not a Doctor
  After years of having to contend with patients who’d consulted Dr. Google, medical workers now need to deal with the unreliable advice dispensed by ChatGPT and similar tools. Rachel Draelos, MD, PhD’s deep dive unpacks the obvious—and less obvious—risks of outsourcing diagnoses and treatment strategies to general-purpose chatbots.

Looking to branch out into some other topics this week? We hope so—the following are all top-notch reads:

• For a comprehensive, one-stop introduction to prompt engineering, don’t miss Anand Subramanian’s debut TDS contribution.
• Discover the power of hierarchical navigable small worlds (HNSW) and how you can use them in nearest-neighbor search by following along Ryan McDermott’s explainer.
• Ready to take a step back and expand your knowledge of the history and future of AI? Thu Vu presents seven book recommendations to add to your reading list.
• In an accessible tutorial, P.G. Baumstarck invites us to take a detailed look under the hood of Torch’s stochastic gradient descent (SGD) optimizer.
• LLMs have become very popular programming assistants, but how sound is the code they produce, really? Andrea Valenzuela outlines some of the major shortcomings that users need to keep in mind.

Thank you for supporting the work of our authors! If you’re feeling inspired to join their ranks, why not write your first post? We’d love to read it.

Until the next Variable,

TDS Team

How to Navigate AI’s Growing Social Footprint was originally published in Towards Data Science on Medium, where people are continuing the conversation by highlighting and responding to this story.