Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Amazon is about to launch its digital assistant Alexa as an artificial intelligence “agent” that can perform practical tasks, while the technology team is running to solve the problems that caused the system’s AI revolution.
The $2.4tn company in the last two years has been looking to revive Alexa, its communication system is installed in the devices of 500mn customers worldwide, so the “brain” of the software is installed with AI producing.
Rohit Prasad, who heads the artificial intelligence (AGI) group at Amazontold the Financial Times that the voice assistant still needs to overcome several technical hurdles before it goes live.
This includes solving the problem of “hallucinations” or imaginary responses, its response speed or “latency”, and reliability. “The feedback should be close to zero,” Prasad said. “It’s still an open problem in the industry, but we’re working hard on it.”
The vision of Amazon’s leaders is to transform Alexa, which is currently used for narrow tasks such as playing music and setting alarms, into an “agency” product that acts as a personal concierge. This can include anything from restaurant recommendations to adjusting the lighting in a room based on a person’s bedtime.
The renaissance of Alexa has been on the train since the launch of OpenAI’s ChatGPT, supported by Microsoft, in late 2022. Although Microsoft, Google, Meta and others have added artificial AI to their platforms computer and improve their software services, analysts have questioned whether Amazon can. solve its technical and organizational problems in time to compete with them.
According to several employees who have worked on Amazon’s voice assistant teams in recent years, its efforts have been challenging and follow years of AI research and development.
Several former employees said that the long wait was largely due to the unexpected challenges involved in adapting and integrating the simple, predefined processes Alexa is built on, with and powerful but unexpected forms of speech.
In response, Amazon said it was “working hard to make it even more helpful and powerful” for its voice assistant. It added that the technical implementation of this scale, being a live service and equipment used by customers around the world, was unprecedented, and not easy like covering LLM in Alexa service.
Prasad, Alexa’s former chief architect, said last month’s release of the company’s Amazon Nova models — led by his AGI team — was driven by the unique speed requirements. and high, cost and reliability, in order to help AI. Tools like Alexa “to reach the last mile, which is very difficult”.
To act as an agent, Alexa’s “brain” must call hundreds of third-party software and services, Prasad said.
“Sometimes we underestimate how many services are integrated with Alexa, and it’s a huge number. These devices receive billions of requests per week, so when you’re trying to do reliable actions quickly . . . you have to be able to do it in a cost-effective way,” he added.
The complexity comes from Alexa users who expect fast responses as well as very high levels of accuracy. Such characteristics are incompatible with the natural way to produce today’s AI, mathematical software that predicts words based on language and speech patterns.
Some former employees also point to the difficulties of maintaining the assistant’s original characteristics, including its stability and functionality, while infusing it with new features such as innovation and free chat.
Thanks to personalized, conversational LLMs, the company also plans to hire experts to create AI personalities, voice and diction to remain familiar to Alexa users, according to the person. one who knows this matter.
A former senior member of the Alexa team said that while the LLM was very sophisticated, it did come with risks, such as producing answers that were “completely made up at times”.
“At the rate Amazon operates, that could happen several times a day,” they said, damaging its brand and reputation.
In June, Mihail Eric, a former Alexa machine learning scientist and founding member of its “conversation group” said publicly that Amazon had “dropped the ball” by being “the undisputed market leader in AI communication” with Alexa.
Eric said that despite having strong technical talent and “huge” financial resources, the company was “plagued by technical and administrative problems”, suggesting that “data was not well defined” and “documents were missing or out of date”.
According to two early employees working on Alexa-related AI, the historical technology behind the voice assistant was inflexible and difficult to change quickly, weighed down by a code base that didn’t unorganized and disorganized and the engineering team is “too scattered”.
The original Alexa software, built on technology acquired from Britain’s Evi in 2012, was a question-answering machine that worked by searching through a defined universe of data to find the right answer. , such as the weather of the day or a particular situation. a song in your music library.
The new Alexa uses a group of different types of AI to recognize and interpret voice questions and generate answers, as well as identify violations, such as receiving inappropriate answers and views. Building the software to translate between legacy systems and new AI models has been a major hurdle in integrating Alexa-LLM.
The models include Amazon’s in-house software, including the latest Nova models, as well as Claude, an AI model from Anthropic startup, in which Amazon has invested. $8bn above during the last 18 months.
“(T)he most difficult thing about AI agents is to make sure they are safe, reliable and predictable,” Anthropic chief executive Dario Amodei told the FT last year.
Agent-like AI software needs to get to the point “where . . . people can trust the system”, he added. “Once we reach that point, we will release these systems.”
One current employee said more steps are still needed, such as covering child safety filters and testing custom integrations with Alexa such as smart lights and doorbells.
“Reliability is the issue – making it work close to 100 percent of the time,” the employee added. “That’s why you see us . . . or Apple or Google is sending slowly and slowly.
Many third parties developing “skills” or features for Alexa say they are unsure when the new AI tool will be released and how they can make it perform new tasks.
“We are waiting for details and understanding,” said Thomas Lindgren, co-founder of Swedish content developer Wanderword. “When we started working with them, they were more open . . . then over time, they changed.”
One partner said that after the first period of “pressure” that was placed on Amazon developers to start preparing for the next generation of Alexa, things were quiet.
An enduring challenge for Amazon’s Alexa team — which was hit by major layoffs in 2023 — is monetization. Finding a way to make the assistants “cost-effective enough to run at scale” will be a big task, said Jared Roesch, co-founder of AI product group OctoAI.
Options being discussed include creating a new Alexa subscription service, or cutting back on sales of goods and services, a former Alexa employee said.
Prasad said Amazon’s goal was to create different types of AI that could serve as “building blocks” for different applications beyond Alexa.
“What we’re always focused on is consumers and AI in action, we don’t do science for science’s sake,” Prasad said. “We are doing this . . . providing customer value and impact, which in this era of AI innovation is more important than ever because customers want to see a return on investment. ”
