Why Siri is not as smart as Alexa, Google Assistant

At any time, most of us are within earshot of a virtual assistant. They are in our pockets, our homes and our cars.

Whether you use Apple's Siri to remind you of an appointment, asking Alexa of Amazon to play a song for you or consulting the Google Assistant to get a local weather report, let them know. interaction with these non-human assistants has become normal.

Siri arrived on the iPhone in 2011, but the underlying technologies are actually older than you think.

The first machine capable of synthesizing speech was created by the Bell Labs 80 years ago, in 1939.

In 1952, Bell Labs invented a machine capable of understanding spoken numbers from 0 to 9. Two years later, in 1954, an IBM machine, in collaboration with linguists from Georgetown, was able to translate 60 sentences into Russian in English. .

In 1962, IBM created the Shoebox, which could include 16 spoken words. In 1976, Carnegie Mellon increased that number to over a thousand. And in the mid-1980s, machines could include tens of thousands of spoken words.

Since then, scientists have begun to combine these processes with artificial intelligence, a field that has existed since the 1950s.

As a result, we now have tools such as Alexa, Siri, Google Assistant and Microsoft Cortana, able to understand us when we speak.

Artificial intelligence is one of the main reasons why computer scientists have been able to simplify the use of wizards, but there could be a difference between what you might think when you hear it. AI and what it really means.

"There are two types of AI," says Joyce Chai, a professor of computer science and engineering at Michigan State University.

"The powerful artificial intelligence mainly concerns developing systems capable of reasoning or thinking or acting like a human. The other type is weak AI, which is more focused on specific tasks, and this also includes virtual assistants, very far from a strong AI. "

Traditionally, in order to make a decision about something, a computer needs a set of rules predefined by a human. By relying on machine learning, which is a type of artificial intelligence, computers are able to deduce rules themselves after having traveled huge amounts of data.

In this case, they can learn to understand the language by looking at how people speak and interact. This requires a lot of data.

Natalie Schluter, Associate Professor of Computer Science at the University of Copenhagen, explains.

"The main challenge for these companies is to get enough data in forms that are diverse enough to actually do something for many people." It could be very interesting in a lab to create a product that can understand and understand you. But of course, there are different dialects, different accents, different tonalities of voice. "

And it's not just the amount of data, the nice data also counts. If your training data only comes from San Francisco Whites, you will end up with an AI that can understand a very small group of people.

"They have smart people working at Apple and very smart people working at Amazon, "says Schluter. But at some point, we need to make sure that these people are involved in the data and make sure they are exposed to the right amount of data. data from a diverse number of people ".

So, why does not Siri always understand what you are looking for?

In part, it concerns things that have nothing to do with science and everything with the reality of the operation of different companies.

"One of Siri's challenges is the negative image they created because of the early and underperforming results," says Keyvan Mohajer, co-founder and CEO of SoundHound, a company offering a competing virtual assistant Siri, as well as that music recognition technologies and voice tools that can be used by other companies.

"The other challenge they have, is that they have not really grown the knowledge base as quickly as expected." Amazon has gone from a few skills to hundreds , thousands, tens of thousands. "Apple has not actually created an ecosystem of developers."

Siri has also fallen behind because of Apple's strict privacy standards. While many virtual assistants collect as much data as possible to train their artificial intelligence, Apple has emphasized the importance of minimizing and anonymizing this type of data collection. Although it is suggested that this results in a less useful assistant, Apple strongly disagrees.

"We reject the excuse that making the most of technology means giving up your right to privacy," Apple CEO Tim Cook said in a speech delivered at the conference. Duke University in 2018.

Beyond that, Apple is a notoriously secretive company.

"What are people working on, what do they think are really important issues at Apple? We have no idea of ​​that, "said Schluter.

"Usually at Amazon, at Google, at other companies, Microsoft, we researchers, we all work in the same field and we attend the same conferences, we publish, we collaborate together, Apple is a completely closed book."

But it seems that Apple has begun to take these things more seriously. Last year, he hired John Giannandrea, a renowned computer scientist, out of Google, to take the position of Senior Vice President, Machine Learning and AI Strategy. And earlier this year, Ian Goodfellow, one of Google's leading researchers on AI, was recruited as director of machine learning.

A study by Loup Ventures at the end of 2018 showed that Siri was not in the lead yet, but was gaining on his competitors.

In addition, this week, at the Apple Worldwide Developer Conference, he announced updates to Siri Shortcuts, which allows developers a better integration of Siri, as well as a synthetic engine update. Siri, which now uses a fully software-generated voice.

But there are still things that Apple could do if it wants to make Siri more impressive.

"The first version of Siri did 12 things," says Mojaher, "but to be really useful, you have to do everything.It's about covering up and adding more content and having an architecture that allows you to add content and increase understanding faster than linearly. "

"I think one of the most promising things that Apple can do is to create a very successful developer community around Siri, I do not think anyone has achieved that in the field of AI voice." . "