AI in Industry 4.0 and 5.0: The Innovation Race Beyond Automation

Artificial intelligence (AI) is increasingly shaping how industries plan production, control machinery, manage energy, sort waste, treat wastewater and coordinate workers with intelligent systems. However, the next industrial transformation will not be judged only by faster output or lower costs. It will be judged by whether AI can make industrial systems more resilient, sustainable and genuinely human-centred, reveals a new study published in the journal Applied Sciences.

Titled Applications of Artificial Intelligence in Industry 4.0/5.0: Innovations, Challenges, and Future Directions, the editorial by Filipe Pereira, Carlos Felgueiras, Vítor Carvalho and Pedro M. B. Torres synthesises ten contributions covering predictive manufacturing, renewable-energy planning, reinforcement learning, computer vision, industrial wastewater treatment, robotic process automation and bias in AI-supported industrial research. Together, the papers show that AI is moving from a supplementary technology to a strategic layer of modern industry.

It also captures a deeper shift in industrial thinking. Industry 4.0 focused heavily on automation, connectivity and productivity. Industry 5.0 retains those capabilities but adds three more demanding goals: human-centricity, sustainability and resilience.

From the Automation Race to a New Human–Machine Bargain

Industry 4.0 transformed factories by connecting machines, sensors, cloud platforms and cyber-physical systems. Manufacturers gained real-time visibility into production, predictive maintenance reduced unplanned shutdowns, and data analytics improved quality control. Much of this transition was driven by a familiar logic: automate more, produce faster and reduce human error.

Industry 5.0 challenges the limits of that model. It treats workers not simply as costs to be removed from production, but as participants whose knowledge, judgment and adaptability remain essential. Collaborative robotics, intelligent decision support and human–machine interaction are therefore becoming as important as autonomous machinery.

Human-centric industrial policy must therefore go beyond reskilling programmes introduced after technology has already been deployed. Workers should be involved in system design, testing and oversight. Their practical knowledge can identify problems that historical datasets or laboratory simulations miss.

It also changes how industrial performance should be measured. Productivity and cost remain important, but they cannot be the only indicators. Industry 5.0 systems should also be assessed against worker safety, energy use, adaptability during disruption, transparency and the distribution of economic gains.

The editorial reflects this transition by presenting AI not merely as an automation engine, but as a foundation for collaboration between people and intelligent machines. The vision is ambitious, but it will require stronger evidence on how human-centred systems perform in real workplaces and whether they actually improve working conditions.

AI Is Moving from Prediction into the Machinery of Control

The ten contributions illustrate how rapidly AI is expanding from analysis into direct industrial decision-making.

One study uses real-time predictive models to support manufacturing digitalisation and proactive operational decisions. Another applies reinforcement learning to an automated irrigation system, allowing the controller to adapt to changing conditions rather than follow fixed rules. A separate contribution combines programmable logic control with Long Short-Term Memory neural networks to optimise industrial wastewater treatment.

Other applications extend AI into energy and environmental management. Neural networks are used to estimate wind-power density for national-scale energy planning, while computer vision supports automated construction-waste sorting. Machine-learning methods are also examined for laser-welding optimisation, and reinforcement-learning applications are mapped across process industries.

These applications matter because they demonstrate a shift from AI as a forecasting tool to AI as part of the operational control system. Once an algorithm influences water flows, energy planning, industrial treatment or machinery settings, its errors can carry financial, safety and environmental consequences.

This makes reliability and oversight central. A model that performs well under stable conditions may struggle when sensors fail, production inputs change or unexpected events disrupt the system. Industrial environments are rarely as clean or predictable as research settings. Dust, vibration, incomplete data, software incompatibilities and ageing equipment can all degrade performance.

The collection also shows how AI can support sustainability, but those benefits should not be assumed. More efficient wastewater treatment, renewable-energy planning and automated waste sorting clearly offer environmental potential. Yet AI infrastructure also consumes energy, depends on computing hardware and may encourage companies to pursue optimisation without addressing broader patterns of material use.

The sustainability value of industrial AI will depend on what systems are designed to optimise. An algorithm programmed only to minimise costs may increase production intensity or shift environmental burdens elsewhere. Industry 5.0 therefore requires environmental and social criteria to be embedded directly into industrial decision models.

The Weakest Link Is Not the Algorithm

The editorial identifies scalability, explainability, interoperability and legacy-system integration as persistent obstacles. These challenges are closely connected and may prove more decisive than advances in model accuracy.

Many factories operate with machinery installed over several decades. Their equipment may use incompatible communication protocols, fragmented software and incomplete digital records. Introducing AI into such environments is not simply a matter of installing a new application. It requires linking sensors, production databases, control systems and human workflows without disrupting operations.

Interoperability is therefore a strategic industrial issue. If companies cannot exchange data across machines and platforms, AI systems remain confined to narrow functions. Heavy dependence on proprietary vendors may deepen the problem by locking firms into closed ecosystems and limiting their ability to adapt technologies locally.

Explainability is equally important. Industrial managers, engineers and workers need to understand why an AI system recommends changing a production schedule, rejecting a component or altering a control parameter. Black-box systems may perform well statistically but still be unsuitable for high-stakes industrial decisions where accountability is essential.

One contribution in the collection examines bias in AI-supported Industry 4.0 research and proposes mitigation strategies. Bias in industrial systems may be less visible than in consumer applications, but it can still have serious effects. Models trained on data from advanced factories may perform poorly in smaller plants. Systems built around one type of equipment or workforce may fail when transferred to another context.

The editorial itself has limitations. It is a synthesis of a Special Issue rather than a single empirical investigation. The ten papers use different methods, industries and performance measures, so their findings cannot be directly compared or combined. The editorial does not provide common indicators for cost, emissions, worker impact or long-term reliability.

It also offers limited evidence on what happens after pilot projects end. Many industrial AI systems demonstrate technical feasibility, but fewer studies track maintenance costs, organisational resistance, cybersecurity exposure or performance over several years. These unanswered questions matter because successful demonstrations do not automatically translate into scalable industrial transformation.

Industry 5.0 Could Narrow Development Gaps, or Deepen Them

For developing economies, the transition to Industry 5.0 presents both an opportunity and a risk. AI could help manufacturers improve quality, reduce resource waste and compete in higher-value production. Intelligent irrigation, energy forecasting, recycling and wastewater systems also have applications beyond factories, particularly in economies facing infrastructure constraints and climate pressures. These technologies connect with Sustainable Development Goals on industry and innovation, clean energy, water management, responsible consumption and climate action.

But the infrastructure required for industrial AI remains unevenly distributed. Reliable electricity, high-speed connectivity, cloud access, specialised skills and affordable financing are prerequisites that many firms still lack. Small and medium-sized manufacturers may struggle to digitise basic operations, let alone deploy advanced AI.

This creates the possibility of a new industrial divide. Large multinational companies may use AI to strengthen productivity and supply-chain control, while smaller firms become increasingly dependent on external platforms or are excluded from digitally integrated production networks.

Governments therefore need an industrial-AI strategy that extends beyond funding research laboratories. Short-term priorities include interoperable standards, cybersecurity support, testing facilities and workforce training. Public programmes should help smaller firms modernise legacy systems and evaluate technologies before committing substantial capital.

Development agencies and multilateral institutions can support shared infrastructure, open technical standards and regional innovation networks. They should also prioritise applications with broad public value, including clean energy, water management, circular production and climate-resilient manufacturing.

Businesses, meanwhile, should resist adopting AI simply to appear technologically advanced. The strongest projects will begin with a clearly defined operational problem, measurable outcomes and a plan for human oversight. Companies must also assess whether efficiency gains justify implementation costs and whether systems can be maintained without permanent dependence on outside vendors.