The award-winning sci-fi movie Her features an artificial intelligence (AI) system that serves as a romantic companion for its human protagonist. Far from being a fancy of imagination, the AI system Xiaoice, which was first developed in 2014 by Microsoft’s Asia-Pacific R&D Group, has made waves by turning this uncanny concept into reality with its emotional computing framework and suite of products.
The bot was modeled after the personality of a teenage girl and became quickly popular in China, with many users referring to it as their virtual girlfriend, according to TechCrunch.
On July 13, Microsoft (NASDAQ: MSFT) announced that Xiaoice would be spun off as an independent entity, creating an independent future for the AI platform and the company behind it. Chinese tech media outlet Deep Echo scored an interview with Xiaoice’s CEO, Li Di, for a peek into the future.
Localization in Asia first
About Xiaoice’s split from Microsoft, Li said ties with the company remain strong. Harry Shum, a renowned scientist and Microsoft’s former executive vice president of artificial intelligence and research, will be the chairman of Xiaoice under the invitation of Microsoft, Li said.
Li explained that the reason behind the split was a desire to localize more effectively. “We have always been envious of domestic companies and their ability to use innovative business and operations models [to obtain customers]. Xiaoice obtained all its current customers ‘organically,’ but other domestic businesses used customer-focused advertisements or different promotional methods to first attract customers before thinking about retention. We have not used these tactics before, but could try.”
Still, localized sales and marketing is only one part of the puzzle. More importantly, localization engenders speed. “For example, five or six months ago, our commercial revenue had already exceeded RMB 100 million (USD 14.6 million)—without us having our own sales team. [Being part of a] large company gave us access to supporting facilities, but this is not comparable to the speed of an independent team.”
China is only a starting point, but it is not where Xiaoice’s vision ends, Li affirmed. “At inception, we wanted to be a ‘China First’ product, because, at that time, China’s innovation trends had taken off. But now, we want to start from China but later advance globally.”
Li wants to start by localizing the platform in Asian markets such as China, Japan, and Indonesia, which are “emerging markets with strong growth and large populations,” said Li. “Our starting point is the population size because our AI products are very correlated to this metric.”
A precise roadmap
Xiaoice originated as a small China-based research project in Microsoft in 2014. The chatbot uses neural networks trained on huge datasets with a context-aware strategy. Microsoft chose to develop emotional intelligence (EQ) capabilities for Xiaoice’s empathic computing framework, going against the industry tide of focusing on cognitive intelligence (IQ) applications like Amazon’s Alexa, Apple’s Siri, and Microsoft’s other intelligent voice assistant, Cortana.
Since its inception, Xiaoice has gone through several iterations of product development. Its eponymous bot was able to chat with users, draw, write poetry, and even sing. Xiaoice was also able to process multiple reports from financial markets and provide financial information to its users.
According to Li, financial products are just part of a wider framework that comprises business-to-consumer (B2C) chatbot functionalities. “We have been engaged in the financial industry for many years but kept quiet about it. We actually believe that many business-to-business (B2B) industries are ultimately business-to-consumer in the final analysis,” said Li.
In the B2B play, Xiaoice first ventured into the financial information sector and ostensibly established market-leading share. Microsoft claimed in 2018 that Xiaoice was the world’s largest provider of AI financial information, with 90% of traders in Chinese financial institutions now using Xiaoice to generate summaries of market reports.
Xiaoice identified synergies and developed financial products before the spin-off, Li explained. “Documents are passed from department to department, and through robotics process automation, we position ourselves in this delivery process to speed up transfers and interaction.”
Despite Xiaoice’s expansive product ambitions, Li remains vexed about the careless use of the term “artificial intelligence” in technology circles.
“In the beginning, people in our industry all focused on making one thing—smart speakers. But when you go to smart speaker conferences, the same three selling points are touted: industrial design, built-in content, and price. Yet, these have nothing to do with AI. When an ‘AI assistant’ can rely on voice commands to switch on lights, this is not an AI product, but a speaker product with AI technology,” explained Li.
“The reason why Xiaoice builds content generation [in its products] has nothing to do with interaction. When a user tells a smart speaker to play a song, the command registers for two seconds, with the song playing for 20 minutes. Such interaction has no higher existence and only consists of voice commands.”
Xiaoice tries to go above and beyond such simple functionality. Upon launch, the team consciously discarded data based on user requests for facts or commands to do simple tasks, focusing instead on data that would help its bot build up a truly engaging personality.
Cutting across platforms with strong partnerships
Xiaoice wants its interactive products to be accessible across platforms.
The firm has made important advances. For instance, its bot is present across more than 20 Chinese platforms, including Xiaomi, Huawei, Vivo, and Oppo mobile phone products. Users can also access their personal Xiaoice-customized AI boyfriends, girlfriends, and companions across diverse platforms, including WeChat and Weibo.
“We aim to have our Xiaoice personalities accessible on any platform that is utilized by our users,” said Li. “Originally, we were slightly concerned [that brands like Xiaomi would not allow Xiaoice’s technology to operate in their products]. But eventually, they all let us in, including Tencent’s QQ and its other QQ products.”
“We formed a set of reciprocal norms with partners. Platforms are reassured that we will not deprive them of their independence, and in return, they share their user base with us. More importantly, the value of our products outweighs their concerns,” Li said.
Li credits his successful AI business to focusing on continual iteration and a steady accumulation of systemic capabilities, as opposed to AI companies that change their strategic direction frequently. Xiaoice claims to distinguish itself from competitors by its high degree of systemization.
“There are two aspects to competing in B2B—vertical territory and third-party AI use cases. Our core competency when it comes to depth is our high degree of integration and systemization. Our competitors in the industry often have specific advantages in certain fields or single-point technologies, but they have not formed systemic, large-scale competencies as we have,” suggested Li.
“One simple example is how our peers have computerized voice technologies but lack natural language processing (NLP) capabilities as strong as ours. And if they do have NLP capabilities, they are not as competent at product development as us. . . When we announced that our bots could write poetry, we simultaneously developed their ability to process financial information as well, but kept quiet about that,” explained Li.
An industry leader offers advice
Li noted that exercising restraint in seeking commercialization and marketing opportunities is vital when playing a long game. “There is too much to improve. . . Xiaoice exercises greater restraint mainly because it has greater flexibility. Therefore, we do not pay special attention to short-term key performance indicators, such as monthly and daily active users. Ultimately, traffic will come to us.”
Since Xiaoice’s inception, the AI industry has also opened up to robots with EQ. “Initially, the industry believed that developing EQ capabilities had an impact on the efficiency of a robot’s IQ. But Xiaoice uses a completely different way of object-oriented thinking rather than process-oriented thinking,” said Li. “EQ does not mean emotional—in fact, being very emotional is indicative of insufficient EQ. EQ means that I can control the interaction between participants, including controlling my interactions and the direction of the dialogue.”
Over the past few years, other companies such as Google, Amazon, and Facebook have been developing their own assistants that are like Xiaoice. But Li plays down any notion of competition. “We are actually not that worried—we have a six-year head start. Rather, we are afraid that a lack of reference points due to our unique position in the industry could be to our detriment,” said Li.
This article was translated and adapted from an interview carried out by Hong Jian of Deep Echo.