The case against conversational interfaces

Screens, Childhood, and “Computerized Humans”

• Several commenters reflect on how rapidly humans adapted to screens and remotes, arguing that people clearly like screen-based, finger-driven control and won’t abandon it for voice.
• Debate over early childhood exposure: Waldorf-style “no tech before ~6” is praised for supporting imagination and easier “switching off,” but others say early screen use (especially games) was life‑saving, opening access to ideas beyond hostile or anti-intellectual environments.
• Consensus that “screens aren’t the problem, bad content and absent parenting are”; quality and boundedness of games/TV (e.g., Ghibli vs. Cocomelon, Factorio vs. gacha games) matter more than the medium.

Voice / Conversational Interfaces: Where They Work

• Seen as a good secondary channel: setting timers, controlling lights, checking weather, simple reminders, querying smart speakers while hands/eyes are busy.
• Useful for rarely used or complex functions where users know what they want but not how to do it (BI queries, “book a flight around 7pm Friday within this budget,” ad‑hoc scripting, advanced app features).
• Works well as a proxy for a human assistant: small businesses or executives issuing high‑level intents instead of clicking through complex tools.

…and Where They Fail

• Repeated claims that speaking is slower, more tiring, and socially impractical; catastrophic in shared or noisy environments and for dense, precise tasks (coding, driving, gaming, flight search, form-filling).
• Serial, low-bandwidth, and memory-heavy: worse than visually scanning options, comparing many alternatives, or making fine-grained adjustments (car controls, shopping, flight selection).
• Real-world chat/“Copilot Studio” UIs often degrade UX: linear, confirmation-heavy flows that replace a simple form/datepicker with slow back-and-forth.

Augmentation, Not Replacement

• Strong alignment with the article’s core view: natural language should augment, not replace, GUIs and keyboard/mouse.
• Many want an OS-level agent that observes context and supports “telepathic” automation of routine tasks, but without constantly guessing and rearranging interfaces.
• Pushback against “tools trying to be smart”: predictive UIs, autocorrect, algorithmic feeds, and hidden options are seen as hostile to user agency; stable spatial interfaces and explicit commands are preferred.

LLMs, Ambiguity, and Hybrid Design

• Some praise LLMs as mediators that can turn vague human intents into exact commands and summarize complex outputs; others stress nondeterminism, hallucinations, and security risks.
• Natural language is framed by some as a poor “data-transfer mechanism” but by others as a powerful way to negotiate intent, much like working with senior engineers or travel agents.
• Broad agreement that the future is multimodal: keyboard shortcuts, pointing, and structured UIs for speed and precision, with conversational layers for discovery, delegation, and edge cases.