1. Understanding AI Tools and Their Limits

Artificial intelligence is a broad field involving systems that perform tasks commonly associated with human cognitive activity, including recognising patterns, processing language, making predictions, generating media, and supporting decisions. Generative AI produces new text, images, audio, video, code, or other content in response to instructions. Large language models generate language by modelling patterns in data; they can produce useful explanations and drafts, but fluent language should not be confused with verified knowledge (Bender et al., 2021).

1.1 Major categories

Conversational assistants answer questions, brainstorm, explain concepts, and help develop drafts. Research assistants help search for literature, summarise papers, compare sources, and organise evidence. Writing tools revise grammar, style, clarity, and structure. Data tools help inspect datasets, create formulas, write code, and generate visualisations. Creative tools generate or edit images, slides, audio, and video. Coding assistants explain code, suggest implementations, find likely errors, and support software development. Workflow agents can use connected tools to perform multi-step tasks, but they require clear permissions and human oversight.

1.2 What AI does well

AI can reduce the friction of starting a task, provide alternative explanations, transform content into different formats, help organise complex information, identify possible patterns, generate code drafts, and support iterative problem-solving. It is particularly useful when a human can define the task, supply context, assess the output, and correct errors.

1.3 What AI does poorly or inconsistently

AI may invent facts, cite nonexistent publications, misunderstand ambiguous instructions, overlook context, reproduce bias, make arithmetic or logical errors, and state uncertain claims confidently. It may fail to distinguish a plausible answer from a demonstrated answer. Search-connected systems can still misread sources or cite pages that do not support a claim. Research has documented fabricated and inaccurate references in AI-generated academic writing, making source verification essential (Fabrication and errors in the bibliographic citations generated by ChatGPT, 2023; Picazo-Sanchez & Ortiz-Martin, 2026).

1.4 Human–AI collaboration

A sound model is to treat AI as an assistant, not an authority. The human defines the problem, determines what evidence is acceptable, checks the response, makes decisions, and accepts accountability. UNESCO (2023) recommends a human-centred approach to generative AI in education and research, including attention to privacy, inclusion, age-appropriate use, and pedagogical purpose.

1.5 The verification habit

For consequential claims, verify names, dates, quotations, statistics, laws, scientific findings, references, formulas, and recommendations against primary or authoritative sources. Ask the tool to distinguish facts from assumptions, identify uncertainty, and list evidence needed to verify its answer. Then independently check that evidence. Never treat a generated citation as real until the publication and its bibliographic details have been confirmed.