Amazon launched its voice-first digital assistant Alexa to the world in November 2014. The know-how’s identify is alleged to be impressed by Star Trek’s pc system onboard the Starship Empire and underscored CEO Jeff Bezos’ ambition of making a conversational and clever assistant. However, a report claims that regardless of a tech demo final 12 months exhibiting a contextually conscious Alexa, it’s nowhere near being built-in with synthetic intelligence (AI) to be smarter. A former Amazon worker who was engaged on Alexa AI has additionally highlighted data silos and fragmented organisation buildings being detrimental to Alexa’s development.
Former Amazon worker highlights points with enhancing Alexa
In a protracted put up on X (previously referred to as Twitter), Mihail Eric, who labored as Amazon’s Senior Machine Learning Scientist at Alexa AI between 2019 and 2021, shared his expertise of working within the firm and the challenges he confronted. He additionally defined why Alexa was a undertaking doomed for failure.
Highlighting the “bad technical process” within the firm, Eric mentioned the corporate had a really fragmented organisational construction, which meant getting knowledge to conduct coaching for big language fashions (LLM). “It would take weeks to get access to any internal data for analysis or experiments. Data was poorly annotated. Documentation was either nonexistent or stale,” he added.
He additionally mentioned that completely different groups have been engaged on an identical points, which created an environment of inside competitors which was not productive. Further, he discovered managers weren’t fascinated with collaborating on tasks that didn’t reward them.
In the put up, Eric shared a number of cases the place the organisation construction and insurance policies got here in the best way of growing “an Amazon ChatGPT (well before ChatGPT was released).”
Amazon workers reportedly spotlight Alexa’s struggles
Fortune revealed a protracted report the place it cited greater than a dozen unnamed Amazon workers to spotlight the problems the corporate is dealing with in integrating AI capabilities into the digital assistant. One explicit difficulty that surfaced was that Alexa’s present capabilities make it more durable to combine a contemporary tech stack.
Reportedly, Alexa is educated to reply in “utterances”, which basically implies that it was created to answer a consumer command and announce that it was working the requested command (or that it couldn’t perceive the consumer). As a end result, Alexa was not programmed for back-and-forth dialog.
The publication cited a former Amazon machine studying scientist, who defined that the mannequin additionally resulted in Amazon clients studying a extra environment friendly manner of speaking with the digital assistant, which was to present a brief immediate for the motion. This created one other downside. Despite having tons of of thousands and thousands of customers who actively converse with Alexa each day, the information is suited to utterance coaching and never conversations. This has reportedly created an enormous knowledge hole within the organisation.
Further, the report claims that Alexa is a value centre for Amazon, and the corporate loses billions every year because the know-how can’t be monetised but. Meanwhile, Amazon Web Services (AWS) has an AI assistant dubbed Amazon Q which is obtainable to particular enterprises as an add-on and generates cash. Over the years, the Amazon Q division has seen extra investments and even integration with Anthropic’s Claude AI mannequin. However, Alexa’s AI workforce was not given entry to Claude because of knowledge privateness considerations.
When Fortune reached out to Amazon, a spokesperson reportedly denied the claims and mentioned these particulars offered by workers have been dated and didn’t replicate the division’s present state of LLM improvement. While that could be true, the extra conversational Alexa seen on the tech demo final 12 months is but to be launched to the general public by the tech big.
