Rohit Prasad, VP and head scientist for Alexa, explains Alexa's new “Why did you do that?” feature which gives users the reason for why it took a certain action
In an interview with OneZero, Amazon's chief Alexa scientist explains the voice assistant's complicated new feature
onezero.medium.comDave Gershgorn
Context & Ripple Effects
By September 2019, Amazon had spent two years pushing Alexa past simple command execution: the assistant was already experimenting with rudimentary emotional awareness, listening for frustration in a user's voice and would soon ask clarifying questions and shift its intonation by context. The "Why did you do that?" feature Rohit Prasad describes is the transparency counterpart to that autonomy push — as Alexa takes more actions unprompted, Amazon is giving users a way to interrogate them.
It arrived alongside the same-day batch of Alexa updates including multilingual mode and paid celebrity voices, and it matters in hindsight because Prasad now runs a team of thousands building the generative-AI Alexa upgrade — an effort reportedly complicated by privacy concerns inside Amazon. Explainability introduced in 2019 reads today as early groundwork for trusting a far more agentic assistant.
First-order effects
Users of Alexa-controlled devices — especially smart home gear — can now demand a justification for any action the assistant took, turning opaque automation into something auditable at the moment it happens.
Second-order effects
As Alexa gains conversational behaviors like context-dependent intonation and proactive questions, rivals' assistants face rising user expectations that any action-taking assistant must be able to account for itself.
Third-order effects
If assistants keep moving toward autonomous action — as Amazon's later push toward a generative-AI Alexa suggests — explanation features like this one harden into an accountability layer, and governance of how agents justify themselves becomes a product requirement rather than a nicety.
The trend: Voice assistants are shifting from executing commands to acting on their own judgment, with built-in explainability emerging as the trust mechanism that makes that autonomy acceptable in the home.
Has your Alexa ever woken up for reasons that aren't immediately clear? @ozm sat down with Amazon's chief Alexa scientist Rohit Prasad to talk about that very phenomenon — and what they're doing to address it. http://read.medium.com/6MQeXGZ
Siri will soon offer Samuel L. Jackson's voice... Me: Siri, what's the weather? Siri: What country are you from? Me: What? What? Siri: What ain't no country I've ever heard of. They speak English in What? Me: What? Siri: English, motherf—er, do you speak it?
the predicted future: deepfake audio will be used to destroy public trust and empower authoritarians the actual future: deepfake audio will be used to make samuel l jackson yell at you until you order shampoo again https://twitter.com/...
APPLE: We put an extra lens in the iPhone camera and our watch now stays on AMAZON: We built smart eyeglasses, smart earbuds, a smart ring, a sidewalk that senses everything, a surveillance system for dogs, a talking doorbell, an oven, and turned Alexa into Samuel L. Jackson
damn Alexa is soon going to be a narc the Alexa Education Skill API will tie into things like Coursera to allow parents to ask what homework their kid has that night, or how they did on tests (!!!!) https://twitter.com/...
Alexa will apparently realize now when you're frustrated with her answers when she gets something wrong she says “I must have misheard” seems like a bad precedent to set for dudes to yell at a female voice pic.twitter.com/N7XuhOpcpp
@ow @TechmemeChatter Actually I would argue that Alexa has had the most transparent approach to privacy all along. It has always been easy to find and delete transcripts or adjust settings, and Amazon has never hid what they were doing.