Artificial intelligence (AI) is a set of technologies that enable computers to perform tasks traditionally associated with human intelligence. Its practical value lies in specific jobs: finding patterns in data, supporting decisions, responding to questions, or helping automate a process. A useful AI system need not think like a person to do one of these jobs well. section 1.1 section 10.1
AI is built from human-designed algorithms that examine data for patterns and relationships. In supervised learning, an algorithm uses labeled examples: show it which fruits are apples and bananas, and it can learn to identify new ones. Unsupervised learning instead looks for patterns in unlabeled data. Neither approach makes data alone a substitute for common sense, context-based reasoning, or critical thinking. section 3.7 section 1.1 section 1.6 section 7.3
A useful task, not a general mind
Online fraud prevention is a well-defined application of AI. As purchases move online, retailers and payment providers face transactions where identity and card ownership can be difficult to verify. AI can help examine transaction data for useful signals. That is a narrower claim than saying a system understands every customer or can make every consequential decision without oversight. section 10.1
Match the promise of AI to the conditions in which it will operate:
- Look for a clear task and realistic goals. Access to large datasets and powerful computers does not make an intelligent system appear at the push of a button; projects need a plan. section 10.4 section 10.4.2
- Check the whole real-world process, not just one impressive component. A pizza venture used robots and AI to coordinate production and deliveries but ceased operations after encountering challenges. section 8.1
- Test for harm as well as performance. AI can make consequential errors, reflect bias, and be misused to produce deceptive synthetic media. section 1.6 section 10.2
The field has a longer history than today's generative tools suggest. Early researchers modeled logical operations, Alan Turing proposed a test based on whether an observer could distinguish machine answers from human ones, and the name “artificial intelligence” emerged at a 1956 Dartmouth workshop. More recently, tools such as ChatGPT have made AI accessible to many nontechnical users. Visibility, however, is not evidence that artificial general intelligence has arrived. section 1.1 section 1.6 section 1.0
For an AI project, decide what the system must accomplish, involve both technical and domain expertise, and test it against varied, unexpected, and erroneous data. In critical settings, quality assurance also needs to consider security and resilience. Treat AI as a tool that can augment human capabilities—not as an infallible replacement for judgment. section 10.4.5 section 11.10 section 10.4.2
