We posed ten profound questions to AI.
As artificial intelligence permeates every corner of our lives from the laboratory, we are standing at an unprecedented historical juncture. The pace of technological iteration has exceeded the traditional cognitive cycle, bringing with it a series of complex and pressing questions.
This article is not a pure technological prediction report but a in - depth dialogue focusing on multiple dimensions in the AI era. Centering around three core areas: social ethics, business and industry, and technological trends, we posed ten popular questions to ten mainstream AI models. Through this carefully designed "thought experiment", we hope to shed light on areas of consensus and divergence and stimulate in - depth reflection.
Ultimately, all these questions point not to the technology itself but to humanity. While constantly questioning AI, what we are truly exploring is with what wisdom and responsibility we will harness this unprecedented power to jointly define a future that belongs to both humanity and intelligent agents.
Social Ethics Section
Q1: What fundamental ethical issues are touched upon by AI's "digital resurrection"? How will people's understanding of death, memory, and farewell change when technology can infinitely extend "existence"?
Zhipu: It touches on the boundaries of "human autonomy" and "the dignity of the deceased" at the core of ethics, blurring the line between biological death and social death. Technology may bring back the "data ghosts" of the deceased without their prior consent, forcing them to continue serving as emotional tools for the living. We are losing the right to "forget", and mourning is infinitely postponed, which may lead to "morbid attachment".
Claude: Last year, an engineer in Silicon Valley "resurrected" his wife using GPT - 4 and chatted with her every night. This made me realize that what we are facing is not just a technological problem but a fundamental challenge to human civilization's perception of death.
But when technology allows the deceased to "live forever" in an almost perfect digital form, we are actually depriving the living of the ability to mourn and move forward. Future regulatory frameworks must include the design of a "mandatory farewell period" and clear "non - personality" markings to prevent emotional substitution.
Q2: Top - tier AI models are becoming the patents of large enterprises. Will this "computing power gap between the rich and the poor" create a new social stratification?
Doubao: The monopoly of computing power will cause social stratification, which is more difficult to bridge than the traditional gap between the rich and the poor. The training cost of top - tier AI models is extremely high, and small enterprises lose their technological autonomy. Governance requires a multi - pronged approach: policies should establish a "public computing power fund" to subsidize small and medium - sized enterprises; open - source models are the key equalizer in technology, allowing small enterprises to conduct secondary development at low cost; at the same time, large enterprises should be restricted from exclusive use of computing power and required to provide affordable computing power services.
Wenxin Yiyan: From a governance perspective, the government should promote the sharing and inclusiveness of computing power resources, establish public computing power platforms, and encourage the development of open - source models to prevent the technological dividends from being monopolized by a few.
Claude: Let me use a comparison to illustrate the severity of this divide: OpenAI spent about $100 million training GPT - 4, while the total AI R & D investment of 54 African countries is less than one - tenth of that. This is not just a simple technological gap.
In five years, there may only be three to five giants left. Small and medium - sized institutions will either be acquired or become "computing power tenants" - surviving by renting the API interfaces of large enterprises.
Q3: Looking ahead to 2026, which AI - related contradictions will become more prominent? In which areas may there be positive governance breakthroughs?
Tencent Yuanbao (DS): Two AI - related ethical issues will come into focus: deepfake and abuse, which seriously impact social trust; AI psychological addiction and dependence. Laws against malicious deepfakes should be formulated, and research on the social and psychological impacts of AI should be funded.
Grok: Multi - modal AI will intensify the conflicts between employment and privacy (such as the escalation of the Hollywood strike). Governance breakthroughs may occur in the field of ethical governance. Recommendations for the public: improve AI literacy, embrace human - machine collaboration, and adapt to technological changes.
Business and Industry Section
Q4: How to balance "using AI to reduce costs" and "retaining employees"? Can the "AI + human" collaboration model avoid large - scale layoffs?
Gemini: There is a "collaborative synergy" zone between "using AI to reduce costs" and "retaining employees". Transforming employees into "AI instructors" or "auditors" can not only preserve organizational experience but also achieve higher output quality than fully automated AI.
Tencent Yuanbao (DS): Enterprise decision - makers should abandon the "substitution mindset" and adopt an "upgrading mindset". The "AI + human" collaboration model can balance cost reduction and talent retention. AI is responsible for repetitive and rule - based tasks, while humans focus on creative and complex work. Enterprises need to invest resources in employee retraining to help them adapt to new job content.
Qianwen: The liberation of productivity by technology should not be a feast for a few. Support the "AI substitution tax" for the retraining fund and explore "reducing working hours" instead of layoffs. The "AI + human" collaboration model allows AI to do repetitive work and humans to make complex decisions, achieving a balance between efficiency and humanity.
Q5: How much impact does the lack of unified standards for AI applications have on industrial development?
Gemini: The lack of industry standards has led to "uncertainty". Without unified standards for AIGC detection or hallucination rates, enterprises dare not deploy on a large scale, and the industry cannot form economies of scale, seriously restricting industrial development.
Kimi: The lack of standards has led to market chaos and increased R & D costs for enterprises (due to adaptation to multiple standards). It is urgent to establish dual standards of "technical indicators + ethical norms", and set standards first in high - risk industries such as healthcare and finance.
Wenxin Yiyan: The lack of unified standards for AI applications has led to market chaos, lack of trust, and increased costs. Establishing standards for acceptable ranges of AIGC detection rates and model hallucination rates will help promote fair competition in the market.
Q6: What are the core scenarios for AI to empower industrial manufacturing? Where are the key breakthrough points for the integrated development of "AI + industrial Internet"?
Doubao: The core scenarios for AI to empower industrial manufacturing are: predictive maintenance of equipment, AI visual quality inspection, optimization of production processes, and intelligent production scheduling. The breakthrough points for the integration of "AI + industrial Internet" are: first, establish unified data collection standards; second, develop lightweight AI models suitable for industrial scenarios to improve the stability of models in complex environments; third, cultivate compound talents with both "industry + AI" knowledge.
Zhipu: The breakthrough point for integration is the "activation" of industrial data, that is, realizing real - time data processing through edge computing, combining industrial mechanism models with AI data models to improve the accuracy and timeliness of decision - making, and at the same time establishing an industrial knowledge graph.
DeepSeek: The core application scenarios of AI in industrial manufacturing can be summarized into three links: "perception - decision - execution", that is, using AI to perceive production anomalies, make optimization decisions, and drive equipment execution.
Technological Trends Section
Q7: Can the "hallucination" problem of large models be fundamentally solved? What are the core breakthrough directions for the development of large - model technology in the next 3 - 5 years?
Tencent Yuanbao (Hunyuan): "Hallucination" is an inherent attribute of large models and cannot be fundamentally eliminated, but the hallucination rate can be controlled within an acceptable range through improving data quality, optimizing training methods, and using external tools (such as real - time online retrieval). The core breakthrough directions in the future are: unified modeling of multi - modal large models, improvement of few - shot and zero - shot learning capabilities, and the establishment of continuous learning and knowledge update mechanisms for models.
Grok: The "hallucination" of large models cannot be completely eliminated, but it will be significantly improved with the improvement of training data and the optimization of model architectures. The core breakthroughs in the future are the mature application of multi - modal large models, the reduction of model training costs, and the combination with edge computing.
Q8: How will the competitive landscape between open - source and closed - source large models evolve? How should enterprises choose the model type suitable for themselves?
Doubao: There will be a dual - pattern situation where "closed - source models dominate the high - end market, and open - source models capture the mid - and low - end markets". Closed - source large models will occupy core fields such as finance and healthcare with their technological leadership, while open - source models will cover small and medium - sized enterprises and vertical niche scenarios with low cost and high flexibility.
Zhipu: In the future, the competition will shift from "technological competition" to "ecosystem competition". Closed - source large models will build ecological barriers relying on the resources of large enterprises, while open - source models will expand their application scope through community collaboration. Large enterprises can use both types of models in a hybrid way, while small and medium - sized enterprises should give priority to open - source models.
Grok: In the future competitive landscape, closed - source models will dominate the high - end commercial market, and open - source models will be used by small and medium - sized enterprises and developers. Large enterprises in the market should choose closed - source models to ensure performance and service, while small and medium - sized enterprises and entrepreneurs should choose open - source models to reduce costs and iterate products quickly. More and more enterprises will use both types of models in a hybrid way.
Q9: Can the development speed of security technology keep up with the iteration speed of AI technology itself? How should enterprises build a comprehensive AI security system?
Tencent Yuanbao (Hunyuan): The development speed of AI security technology lags behind that of AI technology itself. The concealment and diversity of adversarial attacks are constantly increasing, posing challenges to security defense. The keys for enterprises to build a security system are: first, "shift security left", integrating security design at the early stage of the project; second, full - link protection, covering data collection, model training, deployment, and application; third, dynamic defense, regularly updating security policies.
Kimi: The development of AI security technology lags behind the iteration of AI, mainly because security R & D is "targeted", and new defense technologies need to be developed after the emergence of new models. Enterprises can build a security system by: using federated learning and differential privacy to protect data at the data layer; conducting adversarial training, model watermarking, and tracing at the model layer; deploying real - time monitoring and anomaly detection at the application layer; and establishing security compliance systems and conducting regular security assessments and audits at the management layer.
Q10: What is the development prospect of the integration of edge computing and AI? Which industries will be most profoundly affected by this integration?
Doubao: Edge computing solves the problems of "latency, bandwidth, and privacy" in AI model deployment, and AI endows edge devices with the ability of "intelligent decision - making". The integration of the two will drive AI from the cloud to the terminal. The industries most profoundly affected are: autonomous driving, industrial manufacturing, smart home, and healthcare.
Kimi: The core value of edge AI is "real - time + privacy protection", and it has broad development prospects, and the market scale will continue to grow.
Tencent Yuanbao (DS): The integration of edge computing and AI is the core direction of "distributed intelligence", with broad prospects. It will break the limitations of the centralized deployment of cloud - based AI and realize the ubiquitousness of intelligence.
This article is from the WeChat official account "IT Times" (ID: vittimes), author: IT Times. Republished by 36Kr with permission.