Amazon Alexa to Get ChatGPT-Like Capabilities

Internet

Big tech is going big on generative AI. Almost every tech product is set to add generative AI features in some form or the other. Amazon Alexa, the popular virtual assistant, is also getting its dose of conversational AI tools that can help accelerate the company’s vision to create a great AI-powered personal assistant. “It’s now a question of just how do we implement it,” said Dave Limp, senior VP of devices and services, in a recent media briefing at the company’s headquarters in Seattle, US.

The goal is to enable Alexa to have more natural conversations with users. Generative AI can help achieve that goal even faster. During the briefing, Limp also explained why Amazon cannot simply take ChatGPT and throw it into Alexa. He explained that such large language models tend to “hallucinate” from time to time. For a virtual assistant like Alexa, that could spell trouble. Customers rely on Alexa’s capabilities to offer accurate information — be it prices, weather, or simply turning on your smart home equipment. Limp also confirmed that a lot of generative AI models in the background are already in production.

Limp revealed how consumers are now using Alexa more than ever. He said that year on year, Alexa engagement is up over 30 percent. “It has become their new normal, and it has become a new habit,” he said. Consumers are listening to music and podcasts, shopping, building lists, setting up alarms, etc. Amazon is also working on a bunch of new Alexa-powered products that are “basically in the oven and baking, and ready to come put.”

We spoke to Michele Butti, Head of Alexa International, to understand how Alexa is doing in India, when Alexa will get ChatGPT-like features, and what the company is planning for its Indian audience in the near future.

Note: Some responses have been condensed and slightly edited for clarity.

Gadgets 360: Can you talk about Alexa’s engagement in countries like India?

Michele Butti: Definitely, I’m thinking about where to start, there’s so much to say about India. I’ll probably start from Echo devices, because it’s a bit more similar to what we have learned everywhere else. We started in a similar way where we built a natural authentic experience that we built in English. That was a bit limited, and then we added Hindi, and then we made the multilingual mode available. But even that doesn’t fully reflect the complexity of how customers interact with Alexa in general.

In India, the interaction is in multiple languages, so much deeper and complex that the switch (language) happens in the same phrase, or sentence. We need to go beyond the idea of speaking once in Hindi and once in English, we also need to recognise that many customers are also interacting in a lot more languages. Today, Alexa recognises basic commands in multiple languages, in part because it learned from customers.

Cricket and Bollywood content continue to drive engagement for Alexa in India
Photo Credit: Amazon

Since the beginning, we tried to create content that was engaging, such as deep knowledge of cricket and Bollywood. There are slightly unique expectations in behaviour in countries like India. We made services like music available in India right out of the box, unlike other countries where you have to have a different service provider. A big amount of music consumption is the newly released titles, driven primarily by Bollywood music. This involves updating the catalogue more frequently to be ready to play something that has just been released.

We also experimented with adding Alexa to the Amazon app, adding the convenience of voice interaction and Alexa intelligence to the Amazon app. Customers were able to ask for recommendations, add products to the cart, and other features. We are encouraged by the fact that some of the fastest-growing interactions involve shopping, even though people continue to use other use cases such as creating lists, setting alarms, etc.

One of the unique features we developed for India was Pay with Alexa. The payment industry in India is way more dynamic and exciting in India than anywhere else. Amazon Pay is a very interesting initiative, and adding the convenience of voice makes it easier than the touch experience.

Alexa drives all the voice-operated features on Amazon devices including Fire TV Cube
Photo Credit: Amazon

Gadgets 360: What are some of the biggest challenges you face when developing for markets like India?

Michele Butti: India is a bit unique; people speak a number of different languages. The first barrier was indeed understanding and that’s why we have made changes to our model to understand some of the commands in multiple languages. You can ask to play music in around 10-12 languages in India. The other challenge is the availability of broadband Internet, this is still not unique to India though. It will require broader industry development.

Gadgets 360: Do you plan to expand support for more regional languages in India?

Michele Butti: Probably the first way to answer your question is that we always listen to customers, you know when they’re passionate about something. We have clearly learned that even customers who are perfectly fluent in English, might have more engagement when English isn’t their preferred language. To add support to a language, it doesn’t just involve plain translation. It is the culture, the appreciation of preferences, history, and what is appropriate and what is not, that’s something that can change with region and language. This is where LLMs (large language models) can be an accelerator to help reduce the effort.

Gadgets 360: That means Alexa is getting generative AI support?

Michele Butti: Yes, we are already working on it. It is already training more languages than what we have made available so far.

Gadgets 360: At what point can Alexa become so smart that it can literally train itself?

Michele Butti: I think there are probably multiple components to this question. It starts from data labeling and transcription; we have already been able to dramatically reduce the amount of transcription needed to train the models. There is already supervised learning applied. LLM applications will likely help to further speed things up. AI will allow Alexa to react to more complex questions that were previously trained. There are use cases where LLMs (large language models) can make things easier, such as asking Alexa to record a basketball game (it will involve multiple steps and decision-making).

Gadgets 360: What’s the next big thing for Alexa?

Michele Butti: I don’t think there’s a single silver bullet. I’m personally very excited about making Alexa available in multiple smart properties. We mentioned hospitality, but it could be healthcare facilities or senior citizen facilities. We have launched some of those, that’s one other area where we’re making Alexa available. We have also launched Alexa in real estate, and smart properties. That way real estate developers can integrate Alexa in smart homes right from the beginning. Alexa is a big simplifier for companies that manage such large properties.

Disclosure: Amazon sponsored the correspondent’s flights and accommodation for the event in Seattle, US.