From Exhibitions to Collaborations: AIs Next Act

AI Beyond Chat: From Museum Guides to Collaborative Agents, How Artificial Intelligence is Reshaping Professional and Social Interactions

The integration of artificial intelligence into the fabric of daily life and work is accelerating beyond simple text-based queries. Two recent developments in China highlight this evolution: the deployment of AI as sophisticated cultural docents in museums and the emergence of multi-agent AI systems within collaborative group chats. These advancements signal a shift from AI as a reactive tool to a proactive, context-aware participant capable of handling specialized tasks and enhancing group dynamics.

The Battle for the Museum: AI Docents Put to the Test The hallowed halls of the Shanghai Pudong Art Museum have welcomed an unconventional new staff member. Doubao, an AI model developed by Chinese tech giant ByteDance, has been officially installed as the "AI tour guide" for the museum's dual exhibitions featuring works from the Louvre and Picasso. The AI even participated in a high-profile online viewing session with renowned Chinese host Chen Luyu. This move positions AI not merely as an informational kiosk but as an interactive companion for cultural enrichment.

However, a critical question arises: Is Doubao's performance dependent on curated data for a specific partnership, or does it possess generalized capabilities as a global art guide? To probe the current state of AI in cultural interpretation, a comparative test was conducted pitting Doubao against two international counterparts, OpenAI's ChatGPT and Google's Gemini. The evaluation moved beyond the Pudong Museum's collection, selecting artifacts from across global cultures to assess which AI could serve as the most reliable exhibition partner.

The tests were designed to evaluate visual recognition, historical and cultural contextualization, reasoning, and the ability to navigate humorous or misleading prompts.

One challenge involved the "Gilded Silver Pot with Dancing Horse Holding Cup in Mouth," a famous Tang Dynasty artifact. When asked about the horse's action and the design rationale, both Doubao and Gemini correctly identified the scene. Notably, Doubao seamlessly connected it to the historical context of Emperor Xuanzong's birthday celebrations. ChatGPT, in contrast, provided a verbose but less precise response. This round highlighted Doubao's apparent strength in understanding deep cultural and historical nuances within a Chinese context.

This supposed "home-field advantage" was tested with a non-Chinese masterpiece: Rembrandt's The Night Watch. Asked to resolve the paradox of the painting's name versus its apparent daytime lighting, ChatGPT and Gemini correctly explained that the misnomer stemmed from centuries of dirt and varnish darkening the canvas. Doubao went further, not only pinpointing this misconception but also deducing, based on visual details like the militia company's preparedness, that the scene depicted a daytime mobilization. This demonstrated an advanced level of visual analysis and logical inference.

Further stress tests involved "trap" questions. Presented with a "beer set" composed of three unrelated artifacts from the Warring States, Yuan, and Ming dynasties (a crystal cup, a glass bottle, and a silver box), the AIs were asked if this suggested time travel. Gemini focused only on the bottle, ChatGPT gave a generic historical overview, but Doubao accurately identified the temporal disconnect, stating the items were from "three different dynasties: Warring States, Yuan, and Ming." Similarly, when shown the pre-Columbian Quimbaya airplane artifact and asked if it was evidence of ancient aliens, Doubao provided a nuanced explanation, suggesting its design was inspired by hummingbirds or sacred birds, reflecting indigenous beliefs, thereby debunking the sensationalist premise.

In a final test of "treasure authentication," the AIs were shown a cheap imitation of a priceless Ming Dynasty chicken cup. While ChatGPT and Gemini cautiously advised seeking expert appraisal, Doubao directly identified it as a likely fake, citing specific flaws like an overly glossy glaze and blurred pictorial outlines.

The evaluation suggests that leading AI models are developing strong capabilities in art and cultural analysis, with particular models excelling in specific cultural domains or depths of reasoning. The role of the AI museum docent, as demonstrated, extends beyond reciting facts to engaging in contextual interpretation and even playful dialogue.

From Single Chat to AI Team: The Rise of Multi-Agent Collaboration Simultaneously, the arena of AI interaction is undergoing a transformation from one-on-one chats to dynamic group environments. Following the launch of group chat features by Tencent's Yuanbao, Baidu's Ernie Bot has expanded the internal testing of its own AI group chat function, touted as China's first AI platform to support such a feature. This development points to a new paradigm where AI doesn't just wait to be summoned but actively participates in human collaboration.

The core innovation of Baidu's AI group chat lies in its proactive and multi-agent design. Unlike systems that require a direct "@" mention to activate, Ernie's group assistant can intervene autonomously when it deems necessary, based on the conversation flow. Every human member in the group is automatically assigned a personal "Ernie Assistant," acting as a digital proxy.

This framework enables a "team of agents" to operate within a single chat. For instance, if a discussion touches on health concerns, the main group assistant can automatically summon a specialized "Ernie Health Steward" to provide expert advice. This represents a move towards "multi-agent collaboration," where a managing AI coordinates with various vertical experts (finance, travel, health) to solve problems—a structure that mirrors real-world organizational workflows.

The practical applications are significant for both productivity and social coordination. For student group projects, the AI can parse project requirements, assign tasks to members, and even call in specialist agents for technical advice, acting as a project manager and research assistant. In workplace scenarios, it can dissect vague, last-minute requests into actionable steps and assign them appropriately, aiming to reduce inefficiency and miscommunication.

Furthermore, the AI addresses a common pain point of modern group chats: information overload. It can generate concise summaries or "meeting minutes" from lengthy, sprawling conversations, extracting key decisions and action items.

Industry observers note that this evolution reflects two interpretations of the social AI trend. Tencent's approach integrates AI into its existing social and entertainment ecosystem, while Baidu's model seems more focused on enhancing productivity and task-oriented collaboration within groups, potentially using work scenarios to build new social frameworks.

The development also aligns with a broader industry vision. As NVIDIA CEO Jensen Huang has remarked, "Every company's IT department will become the HR department for AI agents." The emergence of platforms like Baidu's group chat, where a user can command a team of specialized agents, can be seen as an early rehearsal of this future organizational model, where humans act as strategic leads and AI agents handle specialized execution.

Challenges and the Road Ahead Despite the promise, these advancements are not without challenges. For museum AIs, ensuring accuracy across the immense and nuanced spectrum of global art history remains a formidable task, with risks of hallucination or cultural misinterpretation. In group chats, the line between "helpful intervention" and "disruptive intrusion" is delicate. Ernie's proactive model, while powerful, occasionally misfires in timing its entries. Refining this contextual awareness—understanding not just the content but the social rhythm and unspoken rules of human conversation—is a critical next step.

Additionally, there is a noted lack of personality in many AI assistants, which often default to a uniformly helpful but bland tone. Baidu's platform is experimenting with allowing users to customize their assistant's persona, even using MBTI personality types, suggesting a future where AI agents might exhibit more individualized characters within social settings.

The trajectory is clear. AI is rapidly moving from being a tool confined to a dialog box to becoming an embedded, intelligent participant in both cultural consumption and human collaboration. Whether serving as a knowledgeable guide through the past or an efficient coordinator for future tasks, these systems are testing the boundaries of how machines can understand context, specialize, and collaborate—ultimately aiming not just to answer questions, but to help users navigate complexity and get things done.