AI Technology Landscape: From the Turing Test to “AI Plus” — Evolution and Practice

1. Disciplinary Position and Technical Boundaries

Artificial intelligence (AI) is a discipline that studies and develops the theories, methods, and technologies used to simulate, extend, and augment human intelligence. It is not the product of any single technology, but the result of deep cross-disciplinary convergence among computer science, mathematics, neuroscience, cognitive science, cybernetics, and related fields. Within the broader field of intelligent science and technology, its core question has never changed: how can machines exhibit behavior that humans would recognize as intelligent?

Understanding this question requires distinguishing two levels. The first is the capability level: which tasks that traditionally require human intelligence can machines perform, such as language understanding, image recognition, and decision-making and planning. The second is the mechanism level: through what computational processes machines achieve these capabilities, which involves algorithm structures, representation learning, search, and optimization. In engineering practice the two advance together — capability requirements drive mechanism innovation, and mechanism breakthroughs in turn expand the boundary of achievable capabilities.

In terms of technical architecture, AI covers machine learning, natural language processing, computer vision, speech recognition, knowledge representation and reasoning, and planning and decision-making. Deep learning is currently the dominant technical route, built on a supporting system in which intelligent chips, algorithm models, and large-scale data advance together. That system can be abstracted into three layers:

• Compute layer: centered on parallel computing chips and accelerator cards, providing the baseline throughput for large-scale matrix operations and gradient computation;
• Framework layer: provides general-purpose capabilities such as tensor computation, automatic differentiation, and distributed training, serving as the infrastructure for algorithm development;
• Data layer: covers data collection, cleaning, annotation, version management, and quality assessment, and directly determines a model’s performance ceiling.

The three layers are not simply stacked on top of one another; they constrain each other. Compute scale shapes model capacity choices, data quality limits generalization performance, and framework efficiency determines iteration speed. Current technical evolution is extending toward frontier directions such as embodied intelligence and brain-inspired computing: the former emphasizes real-time interaction between agents and the physical environment, while the latter draws on sparse coding and spiking mechanisms in nervous systems to explore computing paradigms with better energy efficiency.

2. Origins and Historical Evolution: From the “Imitation Game” to the Dartmouth Consensus

The intellectual roots of AI go back to earlier foundational work in computation theory, but the discipline’s recognized starting points follow two clear threads.

The first thread is the proposal of the Turing test. In 1950, Alan Turing published “Computing Machinery and Intelligence” in the journal Mind, replacing the philosophical question “can machines think?” with an operational behavioral criterion — the “imitation game,” later widely known as the Turing test. Turing’s insight was that rather than debating the definition of “thinking,” it is more productive to examine whether a machine can exhibit behavior indistinguishable from a human’s. He further demonstrated the universality of digital computers: given enough time and storage, a digital computer can simulate the behavior of any other digital machine. This argument provided the theoretical basis for the idea that intelligence can run on a general-purpose computing device.

The second thread is the naming and chartering of the field. In August 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester, and Claude Shannon jointly submitted a landmark proposal for a two-month study of artificial intelligence to be held at Dartmouth College in the summer of 1956. The proposal not only gave the discipline its name, but also put forward a far-reaching core hypothesis: every aspect of human learning or any feature of intelligence can in principle be precisely described and therefore simulated by a machine. This hypothesis established the field’s basic conviction for decades — intelligence is formalizable, computable, and reproducible.

Starting from these two threads, AI’s evolution can be divided into several broad stages: early work centered on symbolic reasoning and search algorithms, emphasizing logical representation and problem solving; expert systems then rose to prominence, encoding domain knowledge into rule bases and delivering practical value in specific vertical scenarios; in the data-driven era, statistical learning became mainstream, and deep learning, with its representation-learning capability, greatly reduced the reliance on manual feature engineering. Behind every paradigm shift lies a different answer to the same question — how does knowledge enter the system: through hand-coded rules, or through representations learned automatically from data?

3. Talent Pipeline: A Complete Path from Undergraduate to Doctoral Studies

Technological progress depends on a steady supply of talent. In China, top-level policy design and academic program development have advanced in coordination. Under the Ministry of Education’s Action Plan for AI Innovation in Institutions of Higher Education, artificial intelligence was formally established as an undergraduate major in the 2018 round of university program registrations and approvals, with results announced in 2019: 35 universities were approved in the first batch, under the major code 080717T. This institutional step extended AI talent development from the graduate level down to the undergraduate level, forming a more complete talent pipeline.

A complete training system covering both foundational theory and applied technology is now in place, with distinct objectives at each level:

• Undergraduate level: core courses build the foundation across intelligent perception, machine learning, knowledge representation, pattern recognition, and optimization methods, with dual emphasis on mathematical foundations and programming skills, so that students can understand and implement classic algorithms;
• Master’s level: strengthened cross-domain engineering practice, focusing on applying AI methods to concrete industry scenarios and building comprehensive skills in data governance, model training, and system integration, with an emphasis on deployability;
• Doctoral level: focused on original system construction and theoretical breakthroughs; researchers are expected to independently pose new problems, design new methods, build new systems, and produce original research results.

Admission to the major is highly competitive — in 2025, for example, minimum admission scores at some top universities were among the highest — reflecting strong demand for AI talent. A maturing training system and high-quality student intake together form the intellectual foundation for technological innovation.

4. Applications and Impact: From Technical Capability to Industry Empowerment

AI’s value ultimately shows in how it transforms production processes and research paradigms. China’s State Council has issued action opinions on the “AI Plus” initiative, mapping out the path for deep integration of AI with key sectors. Its core logic is to move AI from a “single-point tool” to a “general-purpose enabling capability” that is deeply combined with each industry’s business processes, data assets, and domain knowledge.

From an engineering perspective, AI adoption typically follows a reusable path:

# Illustrative workflow: typical stages of bringing AI capability to production
# 1. Scenario definition: clarify task boundaries and evaluation metrics
task = {"input": "business data", "output": "decision/prediction/generation results", "metric": "accuracy/recall/latency"}
# 2. Data preparation: collection, cleaning, annotation, splitting
dataset = prepare_data(raw_source, schema=task["input"])
# 3. Baseline modeling: start with a simple, interpretable baseline
baseline = train_simple_model(dataset)
# 4. Iterative optimization: adjust features, models, and training strategies based on error analysis
model = iterate(baseline, dataset, budget=...)
# 5. Deployment and monitoring: watch for drift, latency, and resource overhead
deploy(model, monitor=["data_drift", "latency", "resource_usage"])

The key to this path: scenario definition comes before model selection, data quality before algorithmic complexity, and engineering constraints before metric chasing. Many failed deployments are not caused by insufficiently advanced models, but by ambiguous task boundaries, training data distributions that do not match production, or inference costs beyond what the business can bear.

As a driving force of the new round of technological revolution, AI is empowering industries at an unprecedented pace. In manufacturing, visual inspection and predictive maintenance improve product quality and equipment utilization; in scientific research, AI-assisted literature analysis, molecule screening, and experiment design are reshaping research workflows; in services, intelligent interaction powered by natural language processing significantly lowers the barrier to accessing information. At the same time, governance issues such as model explainability, data privacy, and safety alignment have become prominent and must be addressed in step with large-scale deployment.

5. Conclusion

The history of artificial intelligence is, at its core, a history of exploring how to formalize and engineer intelligence. From Turing’s behavioral criterion to the Dartmouth charter, from symbolic reasoning to deep learning, and onward to embodied intelligence and brain-inspired computing, every leap has deepened our understanding of “intelligence” and upgraded the engineering means to realize it. For developers, understanding the layered structure of this technology stack, its evolutionary logic, and its deployment constraints delivers more long-term value than chasing any single hot topic. Against the backdrop of “AI Plus,” practitioners who can effectively connect algorithmic capability with scenario requirements and engineering constraints will be the key force in turning technology into real productivity.