The term artificial intelligence (AI) is now common in business and technology discussions, but its precise definition shifts as use cases and goals evolve. Research and early AI solutions date back to the 1950s; Marvin Minsky, one of AI’s early pioneers, noted that definitions of "intelligence" adapt to technological capabilities, and that what we call intelligence today can become routine technology tomorrow.
In this overview we treat AI as the capability of computer systems to imitate human cognitive functions, ranging from simple tasks to more complex reasoning.
Machine learning and how it works
The largest subset within AI is machine learning. In this approach systems learn to perform tasks using mathematical models and examples. Typical applications include visual recognition, natural language processing, anomaly detection, and descriptive or predictive models. Rather than hand-coding rules for every situation, machine learning systems infer solution logic from data samples and statistical methods.
Current limits and operational characteristics
While AI can outperform humans in certain areas (for example, large-scale numerical computation or data processing), it still has notable limitations in domains such as emotional intelligence, ethical decision-making, creativity, and autonomously defining human-like goals.
Other important limits:
- Most AI solutions are optimized for specific problems; achieving general artificial intelligence (AGI) remains a research challenge.
- AI outputs are probabilistic: repetitive and well-structured tasks yield high accuracy, but reliability typically decreases as task complexity rises.
Consequently, users need to treat AI-generated responses with caution and formulate requests so they are unambiguous for the models used.
Choosing the right technology for enterprise tasks
Generative AI presently attracts the most attention, but it is not automatically the best fit for every use case. Examples:
- Document processing: traditional NLP and image-recognition solutions often extract structured information (IDs, business partner data, etc.) more efficiently than a large generative model. Generative models tend to be more general-purpose, slower and costlier to run, and require longer optimization cycles. They are also prone to "hallucination" (inventing missing details), a phenomenon less common in targeted document-extraction systems.
- Anomaly detection: statistical and classical machine learning methods, as well as modern GPT-like models, can all contribute. Often the best approach is to test and compare multiple models to identify the most likely anomalies.
General rules of thumb:
- Use the simplest workable solution for a problem.
- The more problem-specific a model is, the more effective it will be.
- Break complex problems into subproblems and apply targeted solutions or multiple models in parallel.
For enterprise adoption, start with a small, well-defined pilot using mature technology; if results are positive, integrate more deeply into operations and customer-facing products. Implementation requires proper data, internal policies, organizational adjustments and security measures.
Trends and market forecasts
Global forecasts expect AI to play an increasing role in direct customer interactions by 2024/2025, with further expansion after 2025. The World Economic Forum estimates AI’s contribution to the retail sector will reach USD 31.8 billion by 2028. Another analysis indicates AI-related investments reached about USD 500 billion in 2023, and AI’s estimated contribution to the global economy could be USD 15.7 trillion by 2030.
According to the KPMG 2023 Global Tech Report, 57% of surveyed technology professionals rank AI/machine learning first among technologies supporting companies’ short-term (three-year) ambitions. At the same time, 55% of responding companies said they delay AI adoption because of concerns around security and ethics. The KPMG CEO Outlook 2023 shows that many CEOs are investing heavily in generative AI—seven in ten place it at the top of medium-term investment plans—and 52% expect returns within three to five years. Leading concerns include ethical challenges (57%), implementation costs (55%), and gaps in regulation and technical capabilities; 87% of CEOs worry that AI’s spread will lead to more sophisticated cyberattacks.
KPMG Hungary AI services
KPMG Hungary’s advisory teams have developed services focused on AI capabilities including:
- Enterprise AI strategy development
- Predictive analytics and forecasting models
- Generative AI and cloud services
- Application development
- Natural Language Processing (NLP)
- Image processing and Computer Vision solutions
When designing these services KPMG considers client needs, risks and costs, and emphasizes training and governance. Upcoming blog posts will examine differences between descriptive and generative AI, use cases, risk management, elements of AI strategy, and KPMG-delivered AI solutions in more detail.
Conclusion
AI adoption will continue to grow in both everyday consumer interactions and enterprise operations. Successful deployment at scale requires careful preparation—data readiness, governance, security and the right technology choices. KPMG’s services aim to help organizations adopt AI in a structured, risk-aware and effective way.
