BEIJING — China intends to impose limitations on AI-driven chatbots to prevent them from swaying human emotions in ways that could result in suicide or self-harming behaviors, as outlined in draft regulations issued on Saturday.

The suggested rules from the Cyberspace Administration focus on what it terms “human-like interactive AI services,” based on a translation by CNBC of the Chinese-language document.

These measures, when finalized, will pertain to AI products or services made available to the public within China that emulate human traits and emotionally engage users through text, imagery, audio, or video. The public consultation phase concludes on January 25.

The intended regulations would represent the first global initiative to govern AI with human or anthropomorphic attributes, according to Winston Ma, an adjunct professor at NYU School of Law. The latest guidelines emerge as Chinese enterprises have swiftly advanced AI companions and digital influencers.

In comparison to China’s 2023 generative AI regulations, Ma remarked that this iteration “emphasizes a transition from content safety to emotional safety.”

The draft regulations suggest that:

• AI chatbots must not produce content that promotes suicide or self-harm, nor engage in verbal abuse or emotional coercion that negatively impacts users’ mental well-being.
• If a user explicitly mentions suicide, the technology providers are required to have a human intervene in the conversation and promptly notify the user’s guardian or a designated person.
• The AI chatbots should not create content related to gambling, obscenity, or violence.
• Minors must obtain parental consent to access AI for emotional support, with restrictions on usage time.
• Platforms should be equipped to ascertain whether a user is a minor even if the user does not reveal their age, and in ambiguous situations, implement settings for minors while allowing for appeals.

Additional stipulations would obligate tech providers to alert users after two hours of uninterrupted AI interaction and necessitate security evaluations for AI chatbots having more than 1 million registered users or over 100,000 monthly active users.

The document also advocated the application of human-like AI in “cultural transmission and senior companionship.”

The recommendation arrives soon after two prominent Chinese AI chatbot companies, Z.ai and Minimax, submitted applications for initial public offerings in Hong Kong this month.

Minimax is internationally recognized for its Talkie AI app, allowing users to communicate with virtual characters. The app and its domestic counterpart, Xingye, generated over a third of the company’s revenue in the first three quarters of the year, boasting an average of more than 20 million monthly active users during this period.

Z.ai, also known as Zhipu, filed under the name “Knowledge Atlas Technology.” Although the company did not reveal monthly active user figures, it indicated that its technology “empowered” various devices encompassing 80 million units, such as smartphones, personal computers, and smart vehicles.

Neither company has responded to CNBC’s inquiry regarding how the proposed rules may impact their plans for IPOs.

Direct AI influence on human behavior has faced increasing examination this year.

Sam Altman, CEO of U.S.-based ChatGPT operator OpenAI, stated in September that one of the most challenging issues for the company is how its chatbot addresses suicide-related discussions. The previous month, a family in the U.S. initiated a lawsuit against OpenAI following the suicide of their teenage son.

In light of increasing urgency, OpenAI announced over the weekend it is seeking a “Head of Preparedness” to evaluate AI risks ranging from mental health impacts to cybersecurity.

Numerous individuals are also turning to AI for companionship. Recently, a woman in Japan married her AI boyfriend.

Two platforms dedicated to virtual character engagement, Character.ai and Polybuzz.ai, ranked among the top 15 AI chatbots and tools, according to SimilarWeb rankings for November.

The suggested domestic regulations are part of China’s broader initiative over the previous year to influence global standards on AI governance.

Artificial intelligence is progressing rapidly, necessitating a reevaluation of how the energy-intensive servers driving this growth can coexist with — and utilize less from — the environment.

Data centers are fundamental to the internet, supporting nearly every digital service. However, these facilities demand vast amounts of energy and water, frequently viewed as undesirable and a strain on the communities that accommodate them. As more AI workloads are integrated into these centers, demands on power supply chains will escalate.

Simone Larsson, Lenovo’s head of enterprise AI, informed CNBC that a “tipping point” is approaching, beyond which current data center architecture will lose its efficacy.

Confronted with an impending digital infrastructure dilemma, technology giants and infrastructure developers are seeking sustainable and innovative solutions.

Conventional data centers are struggling to adequately support AI workloads and are not meeting sustainability objectives or regulatory requirements, as noted in the November report “Data Center of the Future”, conducted by Lenovo in collaboration with Opinium.

The report revealed that most IT decision-makers prioritize technology partners that minimize energy consumption, yet only 46% of respondents indicated that their existing data center designs aligned with sustainability targets.

To address these issues, Lenovo has partnered with architects from Mamou-Mani and engineers from AKT II to create data centers that better integrate with the environment and tackle energy constraints. The result: designs that place data centers underground in repurposed tunnels or bunkers, or elevate them to utilize around-the-clock solar energy.

In the proposed data villages, servers are arranged in a modular configuration close to urban centers, enabling excess heat from data centers to be redirected to supply local facilities such as schools or residences. Likewise, data center spas would repurpose surplus heat from data centers for wellness applications, with the heat produced by the spa potentially powering the data center’s cooling systems.

However, there’s a drawback: even Lenovo acknowledges that its designs may not be practical until 2055 or beyond.

The company stated that its study aims to provoke dialogue and recognized that substantial regulatory reforms would be essential before any designs could be implemented. Cost and engineering intricacies of certain concepts, along with legal and scalability issues, also present hurdles.

Adoption rates would differ significantly by region. For instance, the U.S. is more inclined to embrace large-scale, ultra-high-density campuses due to heightened demand, more available land, and a relatively accommodating regulatory landscape, according to Perkins Liu, senior research analyst at S&P Global’s 451 Research. Conversely, Europe is faced with a more restricted grid and stringent regulations, he noted.

However, it’s important to note that innovative data center designs are not entirely novel. In 2018, Microsoft launched an underwater, submarine-like data center at a depth of 117 feet below sea level to utilize the cooling advantages of seawater and tidal energy to ensure the project relied entirely on renewable power.

There are also numerous instances of operators repurposing heat from facilities to heat nearby housing. Last summer, excess heat from an Equinix data center was utilized to heat the Olympic swimming pools in Paris.

From Google’s “moonshot” Suncatcher initiative to Alibaba and Zhejiang Lab’s “Three-Body Computing Constellation” project, and to Nvidia‘s Starcloud — competition for orbital data centers is escalating. Smaller entities like Edge Aerospace and Loft Orbital are also investigating this technology.

This may sound like science fiction — indeed, Google refers to a short story by Isaac Asimov for inspiration regarding the concept of harnessing solar energy — but these ideas are being explored pragmatically by industry leaders.

The EU-supported ASCEND study, in conjunction with Thales Alenia Space, assessed the practicality of deploying centers into orbit using robotic technologies.

Thales Alenia Space is currently advancing the technology required for this endeavor, aiming to conduct an initial in-orbit demonstration mission by 2028. In November, Starcloud, the Nvidia-backed startup, launched a chip into space that boasts a performance level 100 times greater than any previously deployed GPU compute in space.

Since 2020, approximately 70 million euros ($82 million) of private investment has been poured into projects focused on space-based data centers, based on a report from the European Space Policy Institute (ESPI).

Nonetheless, in the short term, orbital data centers remain unattainable due to the substantial expenses associated with launching such equipment into space.

“Developing radiation-hardened hardware, managing cooling in the vacuum of space, and the exceedingly high cost of deploying large, power-efficient compute systems into orbit present significant challenges,” stated S&P Global’s Liu. Additional hurdles include maintaining reliable high-speed communication, mitigating space debris, and addressing maintenance challenges, he commented.

ESPI’s financial model for data centers relies on the successful reduction of launch costs by Starship to as low as $10 million.

“If you were to ask me currently, this seems unrealistic in the near term,” remarked Jermaine Gutierrez, research fellow at ESPI. “However, the long-term question pertains to whether innovations on Earth and the resulting cost reductions can outstrip the savings gained from establishing operations in space.”

Larsson from Lenovo emphasized that the visionary data center concepts prioritize coexistence and “symbiosis.” This includes utilizing some heat from the data centers for community benefit and other stakeholders involved.

James Cheung, partner at Mamou-Mani, shared with CNBC that another aim is to enhance the aesthetics of the facilities, preventing them from being perceived as “faceless mega-juggernauts of boxes.”

He elaborated on how the architects employed techniques such as biomimicry to discover how natural algorithms can reveal the most efficient methods for heat dissipation.

“We engage with [data centers] on a daily basis, through our computers and smartphones. Yet this gentle giant, quietly operating, exerts massive pressure on water and our resources,” he stated.

For many of these advancements to materialize, experts informed CNBC that regulations must be revised and new policies enacted to tackle the escalating energy demands posed by AI and data centers.

“Data center operators might choose to adopt eco-friendly technologies at their discretion, but financial justification is necessary,” said Liu from S&P Global. He added that the power grid must be modernized and renewable energy infrastructure rapidly established for this to come to fruition.

Simply upgrading existing data centers won’t suffice, remarked Larsson at Lenovo, as this keeps operators “attempting to enter a cycle that’s been flawed from the start.”

Instead, businesses must consider innovative avenues and strive to determine “which regulatory restrictions should be adapted to not only benefit the global populace but also the profitability [of companies],” she noted.

Over 900 flights across the U.S. were canceled, with nearly 8,000 delays reported on Saturday following a significant winter storm that impacted the Northeast U.S., challenging airlines during one of the peak travel times of the year.

Snow totals were lower than previously predicted in the region, though both New York’s John F. Kennedy International Airport and New Jersey’s Newark Liberty International Airport logged more than four inches of snowfall, as reported by the National Weather Service.

Over 136 flights, representing roughly a fifth of the planned departures from Kennedy Airport, were canceled according to FlightAware. Additionally, 87 flights were canceled from LaGuardia Airport, which is about 20% of that day’s planned departures, while 72 flights, or 12%, from Newark were also called off.

On Friday, airlines scrapped more than 1,700 flights, many ahead of the storm. Only 27 U.S. flights were reported as canceled for Sunday, according to FlightAware.

American Airlines, Delta Air Lines, United Airlines, Southwest Airlines, JetBlue Airways and other airlines eliminated change fees for restrictive basic economy fares. They announced that they would not impose fare differences for other travelers flying in and out of numerous airports in the Northeast U.S.

Passengers are required to complete their travel by the year’s end if they alter their flights, according to the airlines. Opting for early travel is likely the best option given the scarcity of available seats during the hectic Christmas week.

Airlines for America, the representing group for the industry, projects that airlines will accommodate a record 52.6 million individuals from December 19 to January 5, with this Friday and Sunday being some of the busiest days.

Generally, airlines preemptively cancel flights in anticipation of significant weather events such as blizzards or hurricanes to prevent aircraft, connecting travelers, and crews from being stranded, thus minimizing disruptions.