No one would dare deny Amazon’s pioneering status in many things, but the truth is that Alexa and its family of devices have long lacked the momentum of its early days. Home automation, music playback, voice queries… a whole catalog of services backed by the famous virtual assistant now look outdated and awaiting renewal. All the more so since the advent of generative AI makes it look old-fashioned. This is the key to the innovations presented in September with the undisguised intention of preserving its instrumental role in a secondary line of business, hitherto subordinated to the company’s core business, e-commerce.
The idea, Amazon acknowledges in its advertising, is to customize the user experience so that Alexa and company continue to facilitate the transactions that are the reason for Amazon’s existence: shopping on the one hand, entertainment on the other. Activating them with a simple voice command was amazing in 2014, but nine years later it is a triviality that many users have relegated to the back room. There was, therefore, an urgency to catch up with the new chatbots, which not only process simple queries but also offer a simulated conversation.
Several circumstances came together. Amazon’s new CEO, Andy Jassy, is on a mission to straighten out loss-making businesses, even if the amount may seem venial for a company worth $1.3 trillion on the stock market. The pretext that those losses are generously offset by the profits of other businesses that Alexa supports is useless, if it ever was. One proof that the losses are no longer tolerable is that the hardware division has been punished with disproportionate share of the 27,000 layoffs reported so far this year. Not that founder Jeff Bezos was magnanimous, but Andy Jassy is a grateful employee.
It coincides, not coincidentally, with a new professional destination for David Limp, who has been the head of the devices branch and who is leaving to head Blue Origin, Bezos’ space whim. A surprise that forks: to take Limp’s place, Amazon has snagged a big fish, Panos Panay, executive VP and chief product officer at Microsoft. It should be recalled that under Panay’s command the Surface line of laptops and tablets was developed and that he personally oversaw the launch of Windows 11.
It’s too early to tell how Panay will fit in at Amazon, after 19 years in senior roles at Microsoft, another corporate culture. The timing seems ideal for someone who is said to have been unhappy with the cutbacks imposed on him in experimental device development, a reversal dictated by the market situation.
On the other hand, Amazon was hungry for new ideas, which were in short supply in the latest crop of products unveiled by Limp last month. Neither the new Echo Show 8 nor the second generation of smart
glasses bring a revolution, even if the addition of generative AI is presented as such. Nor do the Marvel and Disney character-inspired children`s speakers have any great attributes. The security cameras and the Fire TV remote are better, because at least they simulate more “natural” interactions as they better understand the language of the users.
Alexa, undoubtedly the centerpiece of this ecosystem, can be asked for ideas to organize an evening, be told to write a poem or update the shopping list, or search for that action movie whose title you can’t remember. These are advances, but they are made with wisdom: what Amazon wanted with them is not a disruption (sic) but for its assistant to do its main tasks better: control home automation and digital entertainment, communicate information and news in real time. The impression remains that other added functions are not more than a mere occurrence, such as the attempt to humanize Alexa with a more expressive voice and provide it with opinions and preferences, without going over the top. Making it identify with the user’s favorite team may be a good trick to support Amazon’s well-known intentions in the broadcasting of sporting events.
Of course, incorporating generative AI into a chatbot comes with difficulties, in the current state of technology. To written queries, its competitors ChatGPT and Bing Chat return responses with some latency, a nuisance also caused by Alexa, however much Amazon claims its assistant is five to ten times faster than rivals. Unlike ChatGPT, it would have a better understanding of the real world – said Limp – because while the knowledge base d OpenAI stops in early 2022, Alexa’s LLM offers “real-time” information and is “more conversational” than previous versions.
Rohit Prasad, Amazon’s chief AI scientist, explained to the audience that the superiority of Alexa’s LLM is a consequence of being built with four components: the model itself, real-world devices and services, personal context, and responsive AI.
There is no reason to doubt it, but, in interactions with journalists during the presentation, Alexa gave several wrong answers, which can either be taken with patience or else pay attention to David Limp’s intriguing explanation. He said that with large language models (LLMs) there is a risk of hallucinations. This is a euphemism coined in haste to explain the failures of AI, but it ends up hiding the real problem: the possibility of an assistant becoming a vehicle for chain misinformation.
In home devices, mishaps can occur if Alexa misinterprets a request and, for example, opens the garage door in the middle of the night. In anticipation of failures that it does not yet control, Amazon has chosen to limit the type of devices enabled to receive responses from the assistant, among which it excludes any security mechanism (such as supposedly smart locks), which it promises to integrate later in its offer. Nothing has been said, however, about the potential dangers of hacking.
The privacy issue is the most problematic. Generative AI has a need for a great deal of data to refine its merits and adjust to user needs. Amazon boasts that it does not sell the information it collects from its customers to third parties, but it is very useful for making decisions about its business and for ‘sharing’ trends (anonymously) with its marketplace subscribers.
The truth is that the capacity of these devices to capture and retain information is overwhelming. There is another delicate factor: the anticipation that there will be a separate Alexa for children has given pedagogues the creeps: is it ethically acceptable to put this technology within the reach of the little ones?
These are questions yet to be elucidated and will give much to talk about. As will the economic rationale for the assistant. When a user frequently uses generative AI, training it has a low but material cost per model inference in the cloud. In what will be his last presentation before stepping down, Limp made it clear that the current version of Alexa will be free, but in the future there will be another one or two with additional capabilities only available to subscribers. Analysts have caught the message on the fly: Amazon is in a hurry to find its own monetization model.
This is a crucial point. Because it’s not just Amazon that’s having a hard time. Google Assistant has also been unsuccessful as a source of revenue and so far has not bothered it. Amazon’s speakers and other devices have been sold at cost price in the hope of making a profit from purchases or subscriptions to digital content. Limp’s warning would indicate a change in strategy that his successor will have to manage.
It should be noted that the Echo is the most popular ‘smart’ speaker worldwide. In the United States alone, 71 million people used them on average monthly in 2022, according to Insider Intelligence, while it is estimated that Amazon’s hardware division would have ‘burned` about $ 5 billion. The new approach seeks to reverse this trend.