Human Assistance: Another Possibility for AI Implementation

[baoshi.rao]

To achieve the goal of 'disability without barriers' for people with disabilities, human assistance is an overlooked vertical track, which also represents a feasible way for AI to be implemented.

(Note: This article is not a startup report but an analysis of specific scenario needs. To avoid misunderstandings, all related company, industry, and business names have been removed.)

Recently, Time magazine announced its list of the 100 Best Inventions of 2019.

The most eye-catching category was 'Accessibility,' or human assistance functions. This section featured inventions like AI-powered hearing aids that can speak, eye-tracking technology, smart canes, intelligent prosthetics, and audio-visual conversion tools.

While artificial intelligence remains a concept hyped by many listed companies, a buzzword in venture capital circles, a showcase for experts, and a source of anxiety in media narratives, some have already taken concrete action, targeting one of the most pressing vertical pain points.

On the other side of the ocean, in China, similar projects and stories are unfolding. Here are two examples.

Xiao Yu (pseudonym, same below) is a voice actor with visual impairment who also manages a WeChat public account. How does she do it?

Once upon a time, Xiao Yu was a bright and promising college student. She had played the piano for over a decade, was learning to drive, loved theater and acting, enjoyed painting in watercolors, and had a vivid imagination for the future.

But fate took a sudden turn. It began with bleeding in her pupils, followed by a minor surgery that failed and led to irreversible consequences. As a girl who had mastered Chopin and Mozart and was dexterous with her hands, she suddenly couldn’t see the piano keys anymore. She was only 19 that year.

She tore up all her unfinished watercolor paintings, kicked anything she could or couldn’t see to pieces, and was filled with hatred for the world.

Xiao Yu was unfortunate, but she was also stubborn. In her chaotic and blurry world, she maintained her hobby of collecting scripts and began voicing characters. But how could she read the densely packed small text?

Fortunately, screen-reading software could convert text to speech, and apps like iFlytek Input could accurately transcribe her voice into text, ensuring none of her creative ideas were forgotten.

Thus, after her vision descended into chaos, 'the mutual conversion of sound and text' became her key link to the world.

Her career as an anime voice actor quickly took off, and she started a public account focused on visual impairment. She remained the passionate and vibrant girl she always was—strong and resilient.

This is a heartwarming case.

In China, there are approximately 60 million people with physical, visual, or hearing impairments. A broader definition of the disabled population reaches 6% of the total population. AI technology is making rapid breakthroughs in several directions to address their core challenges:

1. How to help them see?

Through text recognition technology, unclear text can be converted to speech. Through image recognition, unclear people or objects can be described via audio.

2. How to help them write?

Through voice and semantic recognition, their speech can be converted into text.

3. How to make their daily lives easier?

There are many details here. For example, how do they handle image-based CAPTCHAs that require sliding puzzles? How do they navigate facial recognition systems that require blinking?

Of course, these technologies still face many challenges. Text recognition often requires visually impaired users to focus a camera—but how can they focus? Can irregular handwriting be accurately recognized? Can voice and semantic recognition be absolutely precise, and can it understand dialects?

In Guangdong, there is a communication club for people with disabilities, and one member became the 'best debater.' This young man, serving as the third debater, could engage in heated arguments with non-disabled opponents—despite being deaf and unable to hear any sound.

How can someone who can’t hear, including their own voice, debate with others?

To put it another way, in China, about 210 million people have hearing impairments, with 72 million suffering from moderate or worse hearing loss. Only 2.5% of them have access to effective hearing aids or cochlear implants.

Thankfully, with the rapid development of AI, many solutions have emerged. For example, specialized apps for the hearing impaired can:

• Convert spoken words into text in real time, allowing the 'best debater' from Guangzhou to engage in verbal exchanges. After long-term speech rehabilitation training, he can now 'clap back' directly, even if his speech isn’t perfectly natural yet. This is already a remarkable achievement.

Another challenge for this group is that deaf individuals often communicate via sign language, which is like a foreign language to most people. In the U.S., the ratio of sign language interpreters is 1:120, but in China, it’s far below 1:10,000.

How can this pain point be addressed?

Domestic companies have explored solutions like smart bracelets that track muscle signals from hand movements, translating gestures into text and animations.

However, two major challenges remain:

In Time’s list, we see examples like:

• The OrCam MyEye2, an assistive camera that clips onto glasses. Users can point at objects, and the device reads out text, recognizes faces, and identifies colors.
• The WeWALK smart cane, equipped with sensors, speakers, voice assistants, and navigation. It uses ultrasound to detect overhead obstacles and vibrates as a warning.
• Intelligent prosthetics are bringing brain-computer interfaces to life, allowing users to control them like natural limbs, complete with sensory feedback and autonomous intelligence.

Huang Zhisheng, founder of the Tree Hole Initiative, said, 'The economic benefits of AI advancements should be channeled to vulnerable groups, allowing them to fully benefit.'

People with disabilities are unfortunate—fate has unfairly deprived them of abilities others take for granted. But fortunately, in this era, the rapid progress of AI is breaking down barriers and bridging gaps, making the once-unimaginable possible.

Major companies in the AI field are taking action toward the goal of 'accessibility.'

• Tencent launched the 'Information Accessibility' initiative, enabling visually impaired users to add friends via voiceprints. Its image description feature helps them 'see' pictures, and its Yue Shi Sheng mini-program has been optimized for identity verification needs of those with language barriers.
• Alibaba released a disability support report, revealing that 170,000 disabled individuals run stores on Taobao and Tmall, generating nearly ¥30 billion in sales over three years. Alibaba DAMO Academy and Tsinghua University’s Human-Computer Interaction Lab developed a smartphone film for the blind.
• Since 2017, iFlytek has rolled out the 'Three Sounds of Hope'公益 plan, supporting disability-related projects with funding and technology, and openly sharing underlying tech for voice-text interaction projects. At its 1024 Developer Festival, related technologies and case studies moved many attendees.

Globally, smart speakers from Amazon, Google, and Apple have become a boon for people with disabilities, while advancements in autonomous driving are also aiding mobility.

'Disability without barriers' is becoming a reality, but progress remains slow. Human assistance is an overlooked vertical track, yet its products and business models are clear. Discussing the commercial aspects of this philanthropic field is not out of place.

From a market perspective, AI-based human assistance is the future of the rehabilitation industry. According to前瞻产业研究院, the rehabilitation market is projected to reach ¥103.3 billion by 2022.

From a demand and product perspective, human assistance solutions are maturing rapidly, targeting the daily communication needs and pain points of people with disabilities. These solutions are essentially smart upgrades to canes, hearing aids, and prosthetics, with performance improving alongside data accumulation and algorithm iterations.

From an industry perspective, 'disability without barriers' will integrate with more emerging technologies:

• IoT will bring more smart hardware and wearables.
• Robotics will introduce companionship and rehabilitation robots.
• VR will enable more situational simulation training.

Of course, AI-based human assistance also faces significant challenges: the algorithms and data requirements involved are quite high, making it difficult for ordinary startups to handle.

Perhaps a better model would be: led by the government with participation from large companies to complete the construction of some underlying infrastructure and databases, such as basic data for speech, images, and gestures, as well as algorithms for semantic analysis, and open-source them to the public, like the open platforms of iFlytek and Baidu; while more specialized startups can delve deeper to develop more targeted and cost-effective solutions.