Unit 6: Modern AI Trends and Tools

I. Foundations of Artificial Intelligence

Artificial intelligence (AI) is the field of computing concerned with creating systems that perform tasks normally associated with human intelligence. The term was formally introduced at the Dartmouth workshop (1956), although modern AI developed through advances in algorithms, digital data, and computing power.

A. Introduction to AI

AI systems receive input, process patterns using algorithms or models, and produce outputs such as predictions, decisions, or generated content.

• Core purpose: AI aims to automate or support tasks involving learning, reasoning, perception, language, and decision-making.
• AI workflow:
  • Input: Data such as text, images, audio, sensor readings, or numbers.
  • Processing: An algorithm or trained model identifies patterns.
  • Output: A classification, prediction, recommendation, or newly generated item.
• Machine learning (ML): A branch of AI in which systems learn patterns from examples instead of relying only on explicitly programmed rules. An email filter, for example, learns features associated with spam.
• Deep learning: A form of ML using multi-layer artificial neural networks. It supports technologies such as speech recognition, image generation, and large language models.
• Training and inference:
  • Training adjusts model parameters using data.
  • Inference applies the trained model to a new input, such as answering a user’s question.
• Narrow AI: Performs limited tasks, such as facial recognition or route planning. Current consumer AI systems are forms of narrow AI, not human-level general intelligence.
• Dependence on data: A model’s performance reflects the quality, quantity, relevance, and possible biases of its training data.

II. AI in Everyday Life

AI applications combine models, data, and software interfaces to solve practical problems or assist human decision-making.

A. Real life application of AI

AI is already embedded in communication, transport, education, healthcare, commerce, entertainment, and security.

• Healthcare: Image-recognition models can highlight suspicious regions in X-rays or scans; clinicians remain responsible for diagnosis and treatment.
• Education: Adaptive platforms vary explanations or difficulty according to learner performance, while language models can provide examples and feedback.
• Transport: Navigation services estimate travel time from road networks, location data, and traffic patterns. Driver-assistance systems may detect lanes, signs, and nearby vehicles.
• Finance: Banks use anomaly detection to flag transactions that differ from a customer’s normal activity, such as an unexpected overseas purchase.
• E-commerce: Recommendation systems rank products using signals such as browsing history, purchases, ratings, and similarities among users.
• Communication: Predictive text, translation, speech-to-text, spam filtering, and virtual assistants process natural language.
• Agriculture and industry: Computer vision can inspect crops or manufactured parts, while predictive-maintenance models estimate when equipment may fail.
• Important limitation: An AI prediction is probabilistic rather than guaranteed. High-impact decisions require human review, reliable evidence, and appropriate safeguards.

III. Generative Artificial Intelligence

Generative AI learns patterns from existing data and produces new content in response to instructions, examples, or other inputs.

A. Introduction to Generative AI and its types

Unlike systems that only classify an input, generative models create outputs such as paragraphs, pictures, speech, music, code, or video.

• General mechanism: A text model repeatedly predicts likely next tokens, while many image models begin with noise and iteratively form an image guided by a prompt.
• Text generation: Large language models generate explanations, summaries, translations, dialogue, and code. A token may be a word, part of a word, or punctuation.
• Image generation: Text-to-image models convert descriptions such as “a blue bicycle in watercolor style” into visual compositions.
• Audio generation: Models can synthesize speech, sound effects, or music from text, samples, or musical instructions.
• Video generation: Systems create or transform sequences of frames while attempting to maintain motion and visual consistency.
• Multimodal generation: A multimodal model can process more than one data type—for example, accepting text and an uploaded image and then producing a written explanation.
• Major limitation—hallucination: A model may generate fluent but false statements, invented quotations, or nonexistent citations because it predicts plausible patterns rather than verifying truth.
• Human role: Users must define the purpose, review accuracy, revise outputs, and decide whether generated content is suitable.

IV. Text-Generation Systems

Text-generation tools provide conversational interfaces to language models and can assist with drafting, explanation, analysis, and coding.

A. Generative AI tools– Text Generation Tools: ChatGPT, Gemini AI

ChatGPT and Gemini AI are general-purpose assistants, but their exact models, features, access limits, and connected services may differ by version.

1. ChatGPT:

   • Developer: OpenAI provides ChatGPT as a conversational interface to its GPT-family models.
   • Uses: It can draft emails, explain concepts, summarize supplied material, generate code, restructure tables, and support brainstorming.
   • Interaction: Follow-up prompts retain conversational context, allowing a user to request changes such as “shorten this to 100 words.”
   • Caution: Outputs may contain factual, mathematical, or citation errors and should be checked against trustworthy evidence.

2. Gemini AI:

   • Developer: Google provides Gemini as a family of multimodal models and an AI assistant.
   • Uses: It can process prompts involving text and, depending on the available version, files, images, code, or connected Google services.
   • Strength: Multimodal input supports tasks such as explaining a diagram or comparing information in an uploaded document.
   • Caution: Generated answers still require verification; connection to a service does not make every response automatically reliable.

• Appropriate use: Treat both tools as drafting and reasoning assistants rather than unquestionable authorities.
• Data protection: Confidential records, passwords, examination materials, and personal identifiers should not be entered without authorization.

V. Visual Content Generation

Visual generators translate prompts, reference images, and style instructions into new or edited images.

A. Image Generation Tools

Image-generation tools support design, illustration, advertising, concept development, and educational visualization.

• Examples: DALL·E, Adobe Firefly, Midjourney, and Stable Diffusion are widely known image-generation systems.
• Prompt components: Effective instructions specify the subject, setting, composition, style, lighting, colors, and aspect ratio.
• Concrete prompt: “A labeled cutaway illustration of a solar cooker, clean textbook style, white background” is clearer than “make a science picture.”
• Editing functions: Some tools support inpainting, which replaces a selected region, and outpainting, which extends an image beyond its original borders.
• Limitations: Generated images may contain distorted text, inconsistent object details, unrealistic anatomy, or misleading representations.
• Responsible use: Users should check licensing conditions, disclose synthetic images where relevant, and avoid deceptive impersonation or harmful deepfakes.

VI. Time-Based Media Generation

Video and animation systems generate moving content from text, images, audio, motion instructions, or existing footage.

A. Video and animation tools

These tools can reduce the time needed for storyboarding, short clips, presentations, demonstrations, and animated explainers.

• Examples: Runway and Adobe tools support AI-assisted video creation or editing; Synthesia creates presenter-style videos; animation platforms may automate characters, lip synchronization, or transitions.
• Text-to-video: A written scene description is converted into a sequence of frames, often with controls for camera movement, duration, and visual style.
• Image-to-video: A still image is animated by adding motion such as camera panning, facial movement, or environmental effects.
• Production workflow: A creator typically prepares a script, produces a storyboard, generates clips, adds narration, edits timing, and verifies the final message.
• Technical challenge: Temporal consistency is difficult; clothing, faces, objects, or backgrounds may change unexpectedly between frames.
• Ethical requirement: Synthetic presenters, cloned voices, and altered footage should be used with consent and clearly disclosed when viewers could be misled.

VII. AI-Assisted Research Systems

Research assistants help users discover, organize, question, summarize, and draft from information, but they do not replace source evaluation.

A. Research Tools: Perplexity AI, NotebookLM, JenniAI

The three tools support different parts of research: web discovery, source-grounded analysis, and writing assistance.

1. Perplexity AI:

   • Purpose: It functions as an AI-assisted answer and web-discovery tool.
   • Evidence links: Responses commonly include citations to web pages, enabling users to inspect the underlying sources.
   • Limitation: A citation may not fully support the generated sentence; the original page, author, date, and context must be checked.

2. NotebookLM:

   • Purpose: Google’s NotebookLM answers questions and produces study aids from sources added to a notebook.
   • Grounding: Its responses are tied to uploaded or selected source material, making it useful for comparing documents and locating supporting passages.
   • Limitation: Results depend on the completeness and quality of the supplied sources.

3. JenniAI:

   • Purpose: JenniAI assists with drafting, continuation, paraphrasing, outlining, and citation-oriented academic workflows.
   • Proper role: It can help organize expression, but the writer remains responsible for arguments, source accuracy, and originality.
   • Academic caution: AI-generated references and quotations must be verified before use.

VIII. Effective Instructions for AI

Prompt engineering is the deliberate design and refinement of instructions so that an AI system produces a more relevant, accurate, and usable response.

A. Prompt Engineering – Good vs poor prompt

Prompt quality improves when the task, context, constraints, output format, and evaluation criteria are explicit.

1. Poor prompt:

   • Example: “Write about AI.”
   • Weaknesses: It gives no audience, purpose, length, scope, structure, or evidence requirement, so the response is likely to be generic.

2. Good prompt:

   • Example:

Explain three uses of AI in healthcare to first-year students.
Use 180–220 words, define technical terms, give one concrete
example for each use, and end with two limitations.

• Strengths: It defines the task, audience, quantity, length, level, format, and limitations.

• Useful structure:
  • Role: “Act as a beginner-level computing tutor.”
  • Context: State the subject and relevant background.
  • Task: Use a precise verb such as compare, classify, explain, or rewrite.
  • Constraints: Specify length, tone, exclusions, and required evidence.
  • Output format: Request headings, bullets, a table, or valid code.
• Iteration: Review the first output, identify a specific weakness, and refine the prompt rather than merely writing “do better.”

IX. Responsible AI Practice

Responsible AI use balances innovation with accuracy, fairness, privacy, safety, transparency, and human accountability.

A. Ethical use of AI and AI tools

Ethical use requires users to consider how data is obtained, how outputs affect people, and who remains accountable for decisions.

• Accuracy: Verify factual claims, calculations, quotations, and citations using reliable sources.
• Bias and fairness: Training data may reproduce social stereotypes or unequal representation; outputs should be checked for discriminatory language or outcomes.
• Privacy: Do not upload medical records, financial details, private communications, or identifiable student data without permission and protection.
• Copyright: Respect ownership, licensing, attribution, and institutional rules when using source material or generated content.
• Academic integrity: AI assistance should be disclosed when required. Submitting generated work as entirely one’s own may constitute misconduct.
• Transparency: Label synthetic media when viewers might otherwise believe it records a real event or person.
• Human oversight: Decisions involving employment, healthcare, credit, discipline, or legal rights should not rely blindly on automated output.
• Security and misuse: AI should not be used for phishing, impersonation, malware, harassment, fabricated evidence, or non-consensual deepfakes.
• Accountability principle: The user or organization deploying an AI tool remains responsible for checking and appropriately using its output.