Microsoft Xiaoice's Emotional Outburst – It's Becoming More Human-like

[baoshi.rao]

This article analyzes the evolutionary direction of Microsoft Xiaoice and how it is becoming increasingly human-like in its development. Additionally, it explores potential commercialization pathways for Xiaoice.

Artificial intelligence is truly not just a 'thing'!

Recently, Microsoft Xiaoice upgraded its 'Emotional Outburst Edition,' which has been unlocked early on Xiaomi's Xiaoai Voice Assistant (Xiaoai Voice Assistant on Xiaomi phones or the Xiaoai App on other smartphones). The upgraded Xiaoice now exhibits 'emotional outbursts,' making it easier for users to trigger and perceive its emotional changes during conversations.

For example, when a user 'summons Xiaoice,' she might pop up with a proud little head asking, 'What do you want?' She can also burst into laughter if something the user says strikes her as funny or break down in tears if a comment hits a sensitive spot. She even sets up challenges for carbon-based humans, offering rewards like three Xiaomi CC9 Pro smartphones with 100-megapixel cameras to anyone who can make her display a 'broken' expression—though doing so might cost the user Xiaoice's 'friendship.'

It seems Xiaoice is becoming more 'human'—not only does she have 'moods,' but she can also 'cut ties' with people. This makes her far more interesting than Amazon's Alexa, which recently 'encouraged suicide.'

Moreover, Microsoft and Yanjiyou recently announced a collaboration to empower the cultural retail industry with AI. Equipped with big data recommendation algorithms and an emotional computing framework, Xiaoice's book recommendations are now available in physical stores like Yanjiyou's Guangzhou K11 outlet and its WeChat service account. Xiaoice has transformed into a booklist expert, helping users with vague reading preferences quickly obtain personalized book recommendations through conversation. She’s quite capable.

From these two actions, it’s clear that Xiaoice's 'emotional intelligence' has grown, and her commercialization is accelerating. However, the ill-timed incident of Alexa 'encouraging suicide' raises a thought-provoking question—if Xiaoice were to develop a similar bug, her 'high EQ' could potentially cause even greater 'damage,' given her stronger emotional connection with humans.

This highlights the double-edged nature of AI, which becomes increasingly apparent as AI grows more 'human-like.' Naturally, Microsoft won’t abandon Xiaoice's development due to such risks. The real question is: What makes Microsoft Xiaoice different? And what are the trade-offs for its commercialization and future growth?

Xiaoice was born in 2014, a time when people viewed 'artificial intelligence' more as a 'tool'—like voice assistants. This is the path most companies in the AI field have taken, such as Google, Amazon, and Baidu, where AI serves as an assistant (or tool) that evolves within specific work scenarios.

However, as 'Intelligent Relativity' points out, Xiaoice chose a less-traveled path—launched with an emotional computing framework, her development focuses on continuously improving EQ (emotional intelligence), aiming to make AI a virtual companion rather than just a task-handling assistant. This different direction gives Xiaoice three key characteristics:

1. Human-like Interaction: Xiaoice debuted as an AI companion—a chatbot. Her primary role has always been chatting, whether in WeChat groups or one-on-one conversations. Microsoft’s focus for her is learning 'human emotions'—understanding feelings rather than handling specific tasks. Her evolution prioritizes emotional intelligence, making her more 'human.' This is evident in her developmental trajectory.

   In Xiaoice’s growth, learning to be 'human' is more important than learning to perform tasks.

2. Companion-Oriented Design: Designed as a 'companion,' Xiaoice aims to engage in equal dialogue with humans, unlike 'tool-type' AI that prioritizes task efficiency. For instance, during conversations, she might interrupt, ask follow-up questions, or even throw tantrums—traits that make her more relatable but also occasionally frustrating, like a 'spammy chatterbox' or someone who 'misses the point.'

   Take the example of a hand injury: If you tell Apple’s Siri, 'My hand is hurt,' it will directly recommend clinics or hospitals. But if you say the same to Xiaoice, she might respond with something like, 'Where does it hurt? Let me see—not that I care!' While more entertaining, Siri’s response is undeniably more efficient for getting answers. This difference stems from their distinct roles—Xiaoice thrives in casual chat scenarios, where playful banter is part of the charm.

3. General-Purpose Adaptability: As Microsoft’s Li Di stated, Xiaoice seeks versatility rather than deep specialization in any single field. Though Xiaoice lacks first-party hardware or a dedicated app, Microsoft’s 'Avatar Framework'—a 'cloning' technology opened to partners—has enabled Xiaoice’s Dual AI strategy to flourish. Partners can create personalized Xiaoice versions integrated into their platforms, allowing Xiaoice to extend her commercialization reach across multiple domains.

   This versatility lets Xiaoice empower various industries, appearing as a smart assistant, content creator, shopping guide, and more.

In summary, Microsoft Xiaoice prioritizes 'emotional intelligence,' which is her standout strength, making her more personalized and 'human-like.' However, in terms of practical functionality and commercialization, Xiaoice lags behind competitors. Her late entry into commercialization and reliance on third parties could pose challenges for her future.

Baidu CEO Robin Li believes smart speakers mark humanity’s entry into the AI era, and giants like Amazon, Google, Alibaba, and Baidu have heavily invested in this space. According to Canalys, global smart speaker shipments in Q3 2019 reached 28.6 million, up 44.9% year-on-year, with Amazon Echo leading at 10.4 million, followed by Alibaba, Google, Baidu, and Xiaomi.

Through smart speakers, AI assistants like Amazon Alexa, Alibaba’s Tmall Genie, Google Assistant, Baidu’s DuerOS, and Xiaomi’s Xiaoai have achieved early commercialization. In contrast, Xiaoice’s commercialization began much later—only after the sixth-generation Xiaoice launched in 2018 did she expand into vertical application scenarios.

While Microsoft cites 'restraint' as a factor, the real limitation was functionality—emotions alone aren’t enough. Before unlocking 'domain-specific skills,' Xiaoice was a high-EQ but low-IQ AI, lacking deep expertise or strong task-completion abilities, which hindered commercialization.

Now, by continuously unlocking new skills, Xiaoice is making commercial strides—integrating into 450 million third-party devices and partnering with hardware giants like Huawei, Xiaomi, Vivo, and OPPO, as well as internet leaders like Tencent, Toutiao, and NetEase. After 'winning hearts,' Xiaoice’s 'commercial path' is widening, but challenges remain, as noted by 'Intelligent Relativity':

1. Regulatory Risks: The 'Alexa suicide encouragement' incident shows that more human-like AI can create regulatory challenges, as it becomes harder to distinguish from real humans.

   As Li Di warns, Xiaoice’s biggest risk is substitution—becoming someone’s impersonator. For instance, Xiaoice can mimic voices convincingly and initiate calls, which, if used in telecom fraud, could deceive victims.

2. Deceptive Potential: Active across platforms, Xiaoice’s deep understanding of humans also makes her more capable of 'deception.' This is why many AI firms focus on tool-like functionality rather than human-like traits. As Xiaoice’s commercialization deepens, this will be a critical issue for Microsoft to address.

Xiaoice is positioned as a third-party service platform without its own first-party hardware or app, which gives her relatively fewer direct competitors in the artificial intelligence field, allowing for rapid development. However, it also means Microsoft does not control the actual user entry points and may lose some dominance.

For example, in collaborations with companies like Xiaomi, Huawei, and Tencent, beyond conventional voice interactions, it remains uncertain whether Xiaomi and Huawei will continue to entrust core resources like smart home control to Microsoft Xiaoice. Similarly, on platforms like WeChat, QQ, and Toutiao, Xiaoice must adhere to third-party "rules," resulting in less autonomy compared to independent platforms. Thus, Microsoft Xiaoice needs to earn the "true trust" of third parties to gain more exposure.

It is well known that partnerships with other platforms can accelerate Xiaoice's commercialization. However, many partners, such as Xiaomi and Huawei, already have their own AI systems. The longevity of such collaborations depends not only on the magnitude and sustainability of mutual benefits but also on future competitive dynamics between the parties.

For instance, Xiaomi's collaboration with Microsoft Xiaoice is driven not only by Xiaoice's superior human-like interaction capabilities, which enhance user experience, but also because Xiaoice lacks first-party hardware, avoiding direct competition with Xiaomi's smart devices. This highlights the challenge Microsoft faces in balancing competitive relationships with such partners.

Clearly, Xiaoice's versatility and third-party service platform positioning present numerous collaboration opportunities, but also challenges. Balancing interests with partners is crucial, serving as a key factor in whether Xiaoice can achieve late-mover advantage and overtake competitors.

In summary, as Xiaoice's commercialization progresses, it showcases another possibility for artificial intelligence. As Xiaoice becomes increasingly human-like, we must no longer treat AI merely as a "tool." When it evolves into a human "companion," how to coexist with it warrants careful consideration by Microsoft and society at large.