Recognizing AI's Impact on Data Projects

Recognizing AI's Impact on Data Projects — Interactive Notes
INTERACTIVE DATA NOTES

Recognizing AI's Impact on Data Projects

Explore how generative AI, deep learning, natural language processing, machine learning, and robotic process automation can affect the work of data professionals—and why validation, privacy, and responsible use remain essential.

Generative AIDeep LearningNLPMachine LearningRPA
GENERATE
UNDERSTAND
LEARN
VALIDATE
AUTOMATE
01

AI in Data Projects

AI tools can speed up activities such as generating suggestions, summaries, code, ideas, and analytical narratives. For data professionals, however, faster generation does not remove the responsibility to validate the underlying data, the process, and the AI-produced output.

Create

Generative AI can create text, images, audio, code, summaries, and ideas.

Analyze

AI-assisted tools can help explore large and complex datasets and support analytical work.

Validate

Outputs should be vetted for accuracy, bias, and suitability before they become part of a data project.

Protect

Data professionals should understand privacy and sharing settings before providing company or proprietary data to a tool.

Core responsibility

AI is a tool within the analytical process, not a substitute for professional validation, context, business knowledge, or organizational policy.

02

Generative AI, LLMs & Key Risks

Generative AI is a broad class of models that creates novel content. A large language model (LLM) is one type within that class focused on text and commonly built using transformer neural networks.

Prompt / Input
Generative AI
learned patterns
Text · Image · Audio · Code
Human Validation

Transformers

The transformer architecture enabled large language models to be developed at scale.

Diffusion Models

Generate data by learning how to turn random noise into realistic output, including generated imagery and video.

Bias

Biased training data or inappropriate data in a model can contribute to incorrect or unfair results.

Hallucinations

A model can produce nonsensical or factually inaccurate information while presenting it as though it were true.

Privacy & policy

Understand a tool's privacy, sharing, and training settings. Do not load sensitive company or proprietary data into a service unless its configuration and organizational policies permit that use.

03

Deep Learning & Neural Networks

Deep learning uses artificial neural networks to learn complex patterns from data. The arrangement and connections between layers influence the kinds of patterns a network can model.

INPUTHIDDEN LAYERSOUTPUT
Try It Yourself — Neural Network Playground

Use TensorFlow Playground to see how a neural network learns from data directly in your browser.

Open TensorFlow Playground ↗

Guided Exploration

1 · Choose the Data

Start with one of the simple classification datasets under Data. Notice the blue and orange data points. These represent the two classes the neural network is trying to separate.

2 · Examine the Features

Look at the Features column. Start with X₁ and X₂. These are the input values being passed into the neural network.

3 · Inspect the Hidden Layers

The boxes in the middle represent neurons in the hidden layers. Add or remove neurons and observe how changing the network architecture affects what the model can learn.

4 · Train the Network

Click the Play ▶ button at the top. Watch the Epoch counter increase as the network repeatedly learns from the training data.

5 · Watch the Connections

Observe the lines connecting the neurons. Their appearance represents the learned connection weights. The network changes these weights as it learns.

6 · Examine the Output

Look at the Output visualization. The background shows the regions the network predicts for each class. Compare these regions with the actual blue and orange data points.

Now Experiment

ChangeWhat to Observe
Add more neuronsDoes the network create a more complex decision boundary?
Add another hidden layerCompare what the deeper network learns with the simpler network.
Change the learning rateObserve how quickly or slowly training progresses.
Change the activation functionObserve how the shapes produced by individual neurons and the final output change.
Increase noiseSee how harder or less-clean data affects the network's ability to separate the classes.
Try a different datasetCompare how easily the same network architecture learns simple versus complex patterns.
Connect It to CompTIA Data+

Focus on the concept rather than the mathematics: inputs → hidden layers → learned weights → output. Training repeatedly adjusts the network's connections so that its predictions better match the data. A more complex problem may require a different network architecture, features, or training configuration.

Feedforward Networks

A foundational neural-network architecture.

CNNs

Convolutional neural networks are strongly associated with image recognition.

RNNs

Recurrent neural networks are suited to sequential data such as text and time-series information.

Hardware Accelerators

GPUs and TPUs can accelerate machine-learning workloads that would be slow using only a CPU.

04

Natural Language Processing

Natural language processing (NLP) is a branch of AI that analyzes, understands, and generates natural human language. Applications include sentiment analysis, chatbots, web search, natural language generation, and named entity recognition.

Interactive Example — How NLP Interprets Text

Choose a sample sentence, then move through the NLP steps. This simplified demonstration shows how raw human language can be transformed into useful analytical information.

RAW TEXT
What to notice

A computer does not begin with human understanding. NLP techniques first represent language in a form that software can process, then use patterns and context to perform tasks such as sentiment analysis or named entity recognition.

Sentiment Analysis

Evaluates the emotional tone of text, such as positive, negative, or neutral customer feedback.

Named Entity Recognition

Finds and labels entities such as people, places, departments, job titles, or other defined concepts.

Chatbots

NLP enables conversational interaction so users can request information using natural language.

Context Matters

Words must be interpreted in context. “Not bad at all!” and “This was just bad.” contain the same word but express different sentiment.

05

Preparing Text for NLP

Text is data. Before it is modeled, raw language can be cleaned and transformed into a more useful representation. The appropriate preprocessing depends on the problem being solved.

Example Text Pipeline

Raw text:
"Customers are REALLY enjoying the new reports!"

Standardize:
"customers are really enjoying the new reports"

Tokenize:
["customers","are","really","enjoying","the","new","reports"]

Possible stop-word removal:
["customers","really","enjoying","new","reports"]

Possible stemming / lemmatization:
["customer","really","enjoy","new","report"]

→ structured features suitable for later analysis or modeling
Do not clean blindly

Preprocessing choices depend on the business problem. For example, punctuation or currency symbols may contain useful meaning, and stop-word removal can be harmful for contextual models.

Interactive Dataset Lab — Prepare Text for NLP

Use the same five customer-feedback records to see how each preparation technique changes text. Click a technique or use Next Step to move through the pipeline.

IDRaw customer feedback
End-to-end idea

Raw text → Standardize → Tokenize → Remove selected stop words → Stem/Lemmatize → Extract patterns with RegEx. In a real NLP project, you choose only the preparation steps that support the analytical goal.

06

Machine Learning Approaches

Machine learning applies AI to data so a computer can learn patterns through training rather than relying only on explicit programming. Different learning approaches use different kinds of feedback.

ApproachTraining InformationMain Idea
SupervisedLabeled data; correct outcomes are knownLearn to predict an outcome, such as fraud/not fraud
UnsupervisedUnlabeled dataDiscover hidden patterns or groupings
Semi-supervisedSome labeled + more unlabeled dataLearn with relatively few labeled examples
ReinforcementFeedback from environmentOptimize decisions through rewards and penalties

Interactive Dataset Explorer — Four Machine Learning Approaches

Select an approach to inspect the type of data it uses and interact with a simplified learning example.

Compare the learning signal

Supervised: known labels · Unsupervised: no labels · Semi-supervised: a few labels plus many unlabeled examples · Reinforcement: rewards and penalties from interaction with an environment.

07

Machine Learning Techniques & Foundational Models

Within the learning approaches are techniques used for different analytical problems. Foundational models are large neural networks trained on broad data and can be adapted to specialized applications.

Clustering

An unsupervised process that organizes similar objects into groups without a training set. K-means is one example.

Regression

A supervised statistical method used to understand relationships between variables and support prediction.

Ensemble Learning

Methods such as bagging and stacking combine multiple models to improve performance.

PCA

Principal component analysis can support dimensionality reduction, feature extraction, and visualization.

Transfer Learning

Supports adaptation to a domain or task, including situations with relatively few labeled data.

Foundational Models

Large neural networks trained on broad information can be adapted to many more specialized tasks.

08

Robotic Process Automation

Robotic process automation (RPA) uses software bots to perform routine tasks according to rules and information defined by humans. Unlike machine learning, the RPA layer follows predefined, deterministic workflow logic.

TRIGGERNew customerRULEIF conditionACTIONMove / update dataACTIONNotify / refresh report

Attended Bot

Includes a human in the process.

Unattended Bot

Can run from a trigger without requiring human interaction.

Integration

RPA can use webhooks and API calls to move information between systems and trigger actions.

Reporting Automation

Workflows can pull data, refresh reports, send notifications, distribute reports, and repeat on a schedule.

09

AI + RPA: Intelligent Automation

RPA and AI can be combined while retaining different roles. The RPA layer orchestrates deterministic workflow steps, while AI services can perform tasks such as OCR, document classification, or generative-AI processing within those steps.

PDF uploaded
RPA
detect trigger
AI service
OCR / classify
RPA
route data
Database / Report / Email
Key contrast

RPA executes predefined steps. Machine learning identifies patterns and can adapt based on data. Combining them allows deterministic workflows to call AI capabilities at the appropriate point.

10

Summary Concept Map

The central theme is not simply that AI can automate work. Data professionals need to understand what each technology does, prepare appropriate data, choose suitable methods, and verify the resulting output.

AI + DATA PROJECTSGENERATIVE AIcreate + assistNLPlanguage + preprocessingMACHINE LEARNINGlearn patternsRPArule-based automationBias · HallucinationClean · TokenizePredict · ClusterTrigger · Rule · ActionVALIDATE · PROTECT DATA · APPLY CONTEXT
11

Knowledge Check — 20 Questions

Test the major concepts: generative AI, deep learning, NLP preprocessing, machine learning approaches, and RPA.

1. What is generative AI designed to do?

Answer: B. Create novel content such as text, images, audio, or code

2. A large language model (LLM) is described as:

Answer: A. A type of generative AI focused on text

3. Why should data professionals validate generative AI output?

Answer: B. Outputs can contain inaccuracies or bias

4. What is an AI hallucination?

Answer: B. Factually inaccurate or nonsensical information presented as true

5. Which architecture enabled the development of large language models at scale?

Answer: B. Transformer

6. Deep learning primarily uses:

Answer: A. Neural networks

7. Which neural-network type is especially associated with image recognition?

Answer: A. CNN

8. What does NLP enable AI systems to work with?

Answer: B. Natural human language

9. Which NLP task determines whether text expresses positive, negative, or neutral emotion?

Answer: B. Sentiment analysis

10. What is standardization in NLP preprocessing?

Answer: B. Cleaning text by removing or normalizing unwanted elements

11. What is tokenization?

Answer: A. Breaking text into discrete chunks called tokens

12. Why might stop-word removal be used?

Answer: B. To remove common words that add little information

13. How does lemmatization differ from basic stemming in the notes?

Answer: A. It checks that the resulting root is a valid word

14. Supervised learning trains on:

Answer: B. Labeled data with known outcomes

15. Unsupervised learning is used to:

Answer: A. Discover patterns or groupings in unlabeled data

16. What does reinforcement learning use to guide behavior?

Answer: A. Rewards and penalties

17. Which technique groups similar data points without a training set?

Answer: A. Clustering

18. What distinguishes RPA from machine learning in the source?

Answer: A. RPA follows predefined rule-based workflows

19. What is an unattended RPA bot?

Answer: A. A bot that can run from a trigger without human interaction

20. How can AI and RPA work together?

Answer: A. AI services can perform intelligent tasks while RPA orchestrates when workflow steps occur
RECOGNIZING AI'S IMPACT ON DATA PROJECTS · INTERACTIVE NOTES · MARBLE LIGHT BLUE EDITION