The Intersection of Science and Technology in Society

For this week's Tech Policy Press podcast, I spoke to Dr. Alondra Nelson about AI, democracy, and the future. Some takeaways in brief: • The traditional framing of technology as a separate domain of policy is insufficient. Tech is a horizontal force that intersects with all societal challenges, from climate change to healthcare to democracy itself. • The increased use of AI in federal agencies risks creating a “black box” government—one where accountability and transparency are eroded, public trust diminishes, and citizens have even less oversight of the systems shaping their lives. • The move toward AI-powered government services—like replacing human workers in the Social Security Administration—raises urgent questions about efficiency versus human oversight. The “worst-case scenario” could leave vulnerable people unable to access essential services during crises. • The current AI narrative suggests the future preferred by some in Silicon Valley is unavoidable, but Nelson argues that society still has agency. The public must challenge corporate-driven visions of AI and assert democratic control over how these technologies are integrated into life.

🎯 Bias in Machine Learning: From Physics to Humans I recently had an engaging discussion with some visitors about bias in machine learning—a topic that sits at the intersection of technology, science, and society. One of the persistent challenges is that the word bias is used in very different ways: from statistical bias in modeling to bias in social and ethical contexts. These meanings are often conflated, even though they apply to entirely different domains. From a purely scientific standpoint, physical laws are inherently unbiased. Gravity doesn’t care who you are. When machine learning is used to model physical phenomena, any bias is a technical one - related to simplifications, approximations, or model underfitting. These are well-understood challenges, and we have established tools for dealing with them. However, once we move beyond the realm of physics and mathematics into the human domain - language, social systems, or policy - bias becomes both more complex and more consequential. It can creep in through unbalanced datasets, opaque selection criteria, or even implicit assumptions in model goals. At this point, technical strategies alone are not enough. We need to ask deeper questions about how the data was collected, why we’re modeling certain outcomes, and who benefits from the predictions. From data perspective, causal inference and root cause analysis are solution. These tools allow us to disentangle correlation from causation and help identify where biases originate in the data or model structure. When properly applied, they offer a principled way to correct for biases. Yet, there is a deeper layer of complexity that even the best technical tools cannot resolve: the design of reward functions and policy approximations. Every machine learning system is, explicitly or implicitly, optimizing for something. Whether it’s predictive accuracy, profit, or engagement, the model’s behavior is shaped by the reward signal it’s trained on. The problem is that rewards are often defined without sufficient reflection on their long-term implications or alignment with human values. This is where principles such as ethics, honor, fairness, and responsibility come in. In a way, these principles act as policies optimized under uncertainty for long-term societal reward functions. They help guide behavior when the exact outcome is unknowable, and the stakes are high. Embedding these principles into our design choices, whether in algorithmic objectives or governance structures, is essential to prevent short-term metrics from dominating long-term well-being. Ultimately, machine learning is just a tool. It reflects the values, assumptions, and incentives we encode into it. If we want ML to support a fair and just future, we must be willing to take responsibility not just for how the models work, but for why we build them in the first place. #MachineLearning #Bias #EthicsInAI #CausalInference #AIAlignment #ResponsibleAI #HumanInTheLoop #MLStrategy

We’re living through a massive paradigm-shift that’s not getting enough attention. Today’s negotiations—on AI chip access, rare-earth control, and multibillion-dollar trade frameworks—show how deeply technology now drives geopolitical and economic decisions. In my lifetime - from the mid-80s through the 90s, 2000s and early 2010s, technology, finance and government policy were distinct. Each field performed their functions almost separately. Today’s environment has entirely flipped the convention of the last several decades on its head: Technology is shaping policy with today’s innovations, like generative AI, autonomy, reusable rockets, quantum computing, GLP-1s, bitcoin, new energy solutions and social media. These technologies as well as many others - increasingly in the physical world in addition to the digital one - have profound influence over prices, individual well-being, information dissemination, power, safety, and personal finance. Arguably, the most significant levers in current geopolitical negotiations are those around advanced chips access, data center placement and energy access, and defense technology. Media (& social media), traditional finance (and now cryptocurrency), labor (and now AI labor and autonomy), are all evolving rapidly as a function of these blurred lines. Policymakers are closer than ever to technology and business - in fact, they are now technologists, VCs, and business people. Are you optimistic or pessimistic about this shift? No matter which side you fall on, you can’t ignore it. https://lnkd.in/gRvj7zuB