On a rainy Tuesday in San Francisco, Apple executives took to the stage in a packed auditorium to unveil the fifth generation of the iPhone. The phone, which looked identical to the previous version, came with a new feature that people were quick to talk about: Siri, a virtual assistant.
Scott Forstall, then head of software at Apple, pressed a button on the iPhone to summon Siri and ask her questions. At his request, Siri checked the time in Paris (“8:16 p.m.,” Siri replied), defined the word “mitosis” (“cell division in which the nucleus divides into nuclei containing the same number of chromosomes” ) and compiled a list of 14 highly rated Greek restaurants, including five in Palo Alto, California.
“I’ve been working in AI for a long time and I’m still amazed,” Forstall said.
It was 12 years ago. Since then, people have been far from blown away by Siri and competing artificial intelligence (AI) assistants like Amazon’s Alexa and the Google Assistant. Technology has remained largely stagnant, and talking assistants have become the butt of jokes, notably in a skit by Saturday Night Live of 2018 featuring a smart speaker for the elderly.
Rapid improvement
The tech world is now getting excited about another type of virtual assistant: chatbots. These AI-enabled bots, such as ChatGPT and San Francisco-based OpenAI’s new ChatGPT Plus, can promptly improvise answers to questions typed into a dialog box. People have used ChatGPT to perform complex tasks like coding software, writing business proposals, and writing novels.
ChatGPT, which uses AI to guess the next word, is improving rapidly. A few months ago, he couldn’t write a proper haiku; today, he can do it brilliantly. On Tuesday, OpenAI unveiled its next-generation AI engine, GPT-4, which powers ChatGPT.
The chatbot craze exemplifies how Siri, Alexa, and other voice assistants – which once garnered similar enthusiasm – have lost their lead in the AI race.
Over the past decade, these products have run into obstacles. Siri faced technological limitations, including cumbersome code that took weeks to update with basic functionality, said John Burkey, a former Apple engineer who worked on the assistant. Amazon and Google miscalculated how voice assistants would be used, leading them to invest in technology areas that rarely pay off, according to former employees. When those experiments failed, enthusiasm for the technology waned within companies, they said.
Voice assistants are “stupid as a rock,” said Satya Nadella, Microsoft chief executive, in an interview this month with the FinancialTimes, claiming that recent AI would lead the way. Microsoft has worked closely with OpenAI, investing $13 billion in the startup and incorporating its technology into the Bing search engine and other products.
Complement or control?
Apple declined to comment on the Siri situation. Google said it’s committed to providing a great virtual assistant to help people on their phone and in their home or car; the company is also testing a chatbot called Bard. Amazon said it saw a 30% increase in customer engagement globally with Alexa over the past year and said it was optimistic about its mission to build world-class AI.
Assistants and chatbots are based on different forms of AI. Chatbots are powered by so-called large language models, which are systems trained to recognize and generate text from huge datasets mined from the web. They can then suggest words to complete a sentence.
In contrast, Siri, Alexa, and the Google Assistant are essentially what are known as command and control systems. They can understand a limited list of questions and queries such as “What’s the weather like in New York?” or “Turn on the bedroom light.” If a user asks the virtual assistant to do something that’s not in their code, the bot simply replies that they can’t help them.
Siri was also designed in a way that made adding new features cumbersome, said Burkey, who was given the task of improving the virtual assistant in 2014. Siri’s database contains a gigantic list of words, including names of musical artists and places such as restaurants, in nearly two dozen languages.
It made it “one big snowball,” he said. If someone wants to add a word to Siri’s database, “it ends up in a big pile,” he added.
So seemingly simple updates, like adding new phrases to the set, would require rebuilding the entire database, which could take up to six weeks, Burkey said. Adding more complex features, like new search tools, could take nearly a year. This means Siri doesn’t have the ability to become a creative assistant like ChatGPT.
low paying
Alexa and the Google Assistant rely on Siri-like technology, but companies have struggled to generate significant revenue from assistants, former Amazon and Google officials said. (By contrast, Apple has successfully used Siri to lure shoppers to its iPhones.)
When Amazon launched the Echo, an Alexa-powered smart speaker, in 2014, the company hoped the product would help it boost sales for its online store by allowing consumers to talk to Alexa to switch orders, said a former Amazon executive. But while people had fun playing with Alexa’s ability to answer weather questions and set alarms, few asked her to order items, he added.
Amazon may have overinvested in making new products like now-discontinued alarm clocks and microwaves that worked with Alexa and sold at or below cost, the agency said. former leader.
Nor has Amazon invested enough in creating an ecosystem that allows users to easily extend Alexa’s capabilities, much like what Apple did with its App Store, which helped boost interest in the iPhone, said John Burkey.
While Amazon offered a “skills” store that allowed Alexa to control third-party accessories such as light switches, it was difficult for users to find and set up skills for the speakers – unlike the frictionless experience. downloading mobile applications from application stores.
Amazon’s failures with Alexa may have led Google astray, said a former executive who worked on Google Assistant. Google engineers spent years experimenting with their assistant to mimic what Alexa could do, including designing voice-activated smart speakers and tablet screens to control home accessories like thermostats and light switches. . The company then integrated advertisements into these domestic products, which did not become a major source of revenue.
Convergence in sight
Most of the big tech companies are now scrambling to find answers in ChatGPT. Last month, Apple headquarters hosted its annual AI Summit, an internal event for employees to learn about its big language model and other AI tools, said two people who were informed about the program. Many engineers, including members of the Siri team, tested language-generating concepts every week, the people said.
On Tuesday, Google also said it would soon release generative AI tools to help businesses, governments and software developers build apps with built-in chatbots and incorporate the underlying technology into their systems.
In the future, chatbot and voice assistant technologies will converge, according to AI experts. This means that people will be able to control chatbots with speech, and those who use products from Apple, Amazon and Google will be able to ask virtual assistants to help them with their work, and not just with tasks such as than consulting the weather.
“These products have never worked in the past because we didn’t have human-level dialogue capabilities,” said Aravind Srinivas, founder of Perplexity, an AI startup that offers a powered search engine. by a chatbot. “Today it is. »
This article was originally published in the New York Times.
