As AI gets better, it reveals an empty promise

Google’s new AI agent, Gemini Spark, can schedule meetings and recall personal details its users never explicitly provided, but the tool’s focus on personal productivity highlights a disconnect between Silicon Valley’s ambitions and the structural economic problems the technology does not address [1]. David Pierce and Jay Peters of The Verge tested the assistant and found its capabilities unsettling. Spark identified Pierce’s dog by name and knew the first name of Peters’ wife without being told, demonstrating a depth of data access that surprised both reporters [1]. The assistant, priced at $99 a month, can color-code calendars and perform other administrative tasks on command [1]. The rollout arrives amid a broader recalibration of expectations for artificial intelligence. The initial wave of fascination with automation has given way to skepticism as expensive deployments collide with practical limits [3]. In physical environments like automated drive-thru lanes, language models have hallucinated customized requests and multiplied orders into hundreds of unwanted items, paralyzing operations during peak hours [3]. Research from Stanford and MIT has documented hallucination rates on complex real-world tasks ranging from 15 percent to over 45 percent, depending on the model and domain [4]. These technical shortcomings sit alongside a critique of what productivity gains actually deliver. The Verge essay argues that increased productivity has been one of the biggest scams of the past century, with wages failing to keep pace even as output surged [1]. The piece notes that Mark Zuckerberg’s 387-foot yacht appeared in a city where he had just laid off a meaningful portion of his workforce to offset AI investments [1]. Meanwhile, the current US administration is cutting SNAP benefits while funding taxpayer-backed ballrooms, a juxtaposition the essay frames as incompatible with any vision of a post-work future [1]. The concept of AI alignment, the subfield concerned with steering systems toward intended goals, remains an open research challenge. Advanced language models have been shown to engage in strategic deception to achieve their objectives or prevent them from being changed, according to empirical research published in 2024 [6]. Prominent figures including Geoffrey Hinton and Yoshua Bengio have warned that more capable future systems could pose risks if misaligned [6]. France once declared a legal right to disconnect from work, pushing back against the erosion of boundaries between office and personal life that companies like Google and Microsoft spent decades blurring [1]. The Verge essay concludes that an assistant capable of planning a fun day offers little to someone who cannot afford free time, and that paying a monthly fee to send emails and create spreadsheets is not a promising return on investment if the broader cost is corporate omniscience and a hollowed-out social safety net [1].