Information Systems Midterm

What are the 3 ways IT creates value?

Automates processes to offload low value and repetitive work

democratize data to get information to the people deciding

reduce user friction to make the thing easier to use

What is a problem?

gap between what is and what should be

What is a computer?

history: mechanical turk in 1770 was a chess playing atomaton with a person inside

was also a job title

it became one machine that could be told to do anything

What are the 4 things a computer can do?

take input, store data, process data and have instructions run agianst it, and output(make data available)

Computers vs Humans

Humans think in stories and context, intuitions and judgement, and leaps and shortcuts while computers need steps and sequences, rules and logic, and explicit instructions

What is decomposition

breaking a whole into the parts you can name

4 steps to closing the gap

decompose to break the problem down, find the pattern, abstract by dropping details that don’t change the answer, and design the algorithm so someone or something can run them

What are feedback loops

a continuous cycle where the output of a system circles back to become an input, shaping future behavior

What is example of reinforcing

when one car hits the pothole, then the other cars would realize and know

Example of balancing

the cobra effect

Computational thinking and when you use it

excels when you need to build or automate something

• designing a process, writing requirements, creating a workflow

What is systems thinking and when is it useful?

excels when you need to diagnose or predict something

• understanding why a process fails

• anticipating unintended consequences

What is a system?

set of connected parts that continually produce a pattern

Example of universities as a system

• Things: students, professors, researchers, administrators

• Connections: lectures, research, publications, projects

• System: knowledge flowing between these groups

Examples of looking at both inflows and outflows

Don't just focus on what enters a system. Think about what happens to it afterward.

Examples:

• Company → Don't just hire people; retain and develop them.

• Learning → Don't just consume information; apply it.

• Networking → Don't just meet people; nurture relationships.

• Money → Don't just accumulate it; think about what you will do with it.

What is a reinforcing loop

A reinforcing loop amplifies change.

Example:

Marketing spending ↑ → Sales ↑ → Revenue ↑ → Marketing spending ↑

More of one variable creates more of the next, which eventually feeds back and creates even more growth.

Think:

Reinforcing = growth or decline accelerates

What is a balancing loop

A balancing loop counteracts change and pushes the system toward stability.

Example:

Backlog ↑ → Hiring ↑ → Tickets cleared faster → Backlog ↓ → Hiring pauses

The system responds to the original change and pushes the variable back toward its desired level.

Think:

Balancing = stabilizing / goal-seeking

What is delayed feedback

when the consequence of an action occurs much later

ex: smoking

What are the 4 types of systems?

System	Cause & Effect	Best Approach
Clear	Obvious	Follow process/checklist
Complicated	Exists but is difficult to discover	Analyze + use experts
Complex	Emerges and is understood in hindsight system constantly changing	Experiment + adapt
Chaotic	Cause/effect relationship is broken information is incomplete	Act + stabilize first

Clear ex: following a recipe

Complicated ex: financial modeling or medical diagnosis

complex ex: integrating 2 company cultures, implementing AI

chaotic ex: natural disaster

What is the DART framework?

• used to determine what kind of system

• D- Deconstruct

• A- Analyze

• R- Recognize

• T- Task

What are 4 ways information moves?

Live - happens now, both sides see it (ex: gate scan)

Batch- happens on a schedule, usually overnight(ex: your miles on account)

Manual - a person carries it(ex: the gate agent phoning the ramp to ask if bag is there)

None

Can a connection itself be a system?

Yes, the layer that sits in between 2 things and moves data between them can act as a system and it can break

What are the parts that help a computer function?

CPU, GPU, RAM, SSD

What is a CPU?

a processor that does one thing at a time extremely fast

What is a GPU

the other processor that does the same thing to enormous amounts of data at once

What is RAM

Memory- holds what it’s working on right now. Fast, expensive and forgets when the power goes

What is SSD

storage- keeps it when the power goes. slower, chepaer, remembers

What do the parts look like

Where does all the memory actually go?

“AI data center construction and hyperscaler demand are absorbing a disproportionate share of global memory capacity.” A data center is a building full of the parts on the lastslide. Somebody bought them. Somebody powersthem. Somebody ownsthe building. Nothing about “the cloud” is soft.

What is the cloud?

It is someone else’s computer — the parts we just listed, thatsomebody else bought. Those computers live in the buildings

A virtual environment and delivery model that uses software and virtualization to slice up, manage, and share the hardware inside those data centers. It allows you to access compute power, databases, and storage over the internet on-demand

Where did internet come from?

was a government project. DARPA/ defense research money built the first 4 modes in 1969. Then the NSF money ran the backbone every univeristy rode. Then congress passed a 1991 Act that paid for the next wave if capacity. Then, provate money bought the stuff

What exactly is the internet?

set of agreements on how to address something and how to routie it

• you message is broken into pieces and the pieces take different roads and are reassembled at the dar end

• this is called the cloud which is someone’s hardware located in a building

What is TCP/IP

transmission control protocol / internet protocol — how to cut a message into pieces, addressthem,send them by any road, and put them back together in order at the far end.

What is SMTP?

simple mail transfer protocol — how to hand a message to a mailserver,say who it isfor, and have it accepted or refused.

What is protocol?

agreement everyone follows

What is an API?

same idea od protocol narrowed to one company- it’s an agreement they publish about how to ask them

What does API Stand for?

Application (software you want something from), Programming(asking for it in code), Interface(published place where the asking happens)

Why is API a promise?

it’s published, kept, and what’s behind it can change completely

Example of how information systems talk

Logging into Canvas:

• You → Canvas here

• 02 Canvas → Duo

• 03 Duo → your phone a push, and a wait

• 04 Duo → Canvas

Canvas only knows yes or no which is the design

The 2 systems connecting doesn’t mean they’re merging. The only thing that corsses is your name and then a yes or no from Duo

Example of interface

A waiter is the interface between customers and the kitchen

• menu is the “promise”

What is analytics?

Using data to answer a question about a business

What are the 4 kinds of questions/analytics?

descriptive, diagnostic, predictive, prescriptive

What is descriptive analytics?

What happened? Counting what is already in the record

What is Diagnostic?

Why did it happen? Explaining a difference the count exposed

What is predictive?

What will happen? Estimating a number nobody has recorded yet

What is prescriptive

What should we do? Choosing an action, and defending the choice

What is Data

the numbers

raw record

What is information

the numbers with date and sales

data in context

What is knowledge

the shape you can act on by looking at graphs and conducting an analysis

what you do differently because of it

What is a database

structure for storing data so it can be found again

What is a count

a definitional claim, not a measurement

What is data quality?

fitness for intended use - so the same record can be perfectly good and completely wrong, depending on who is reaching for it

Who does the cost land on

Not the system. It lands on the pwrson who needed the number and couldn’t get it and on the company which makes decisions on numbers it cannot defend

What are the seven questions to look at when testing a data advantage?

1) Value Add

2) Marginal Value

3) Obsolescence

4) Proprietary

5) Imitability

6) Self vs Others

7) Speed

What is a pivot table?

summarizes a table you cannot read by eye

• alternative for pivot tables is writing a formula for every question

What is a pivot table made of?

Rows, columns, values, filters

• one row per distinct value of the field you put here

• one column per distinct value of the field

• what gets computed for the records that land in each cell

• which records are allowed into the grid at all

What is the difference between counting rows and counting things?

a row is a record and a thing is a person, account, or id.

What is the difference between a spreadsheet and a database?

you type things into a spreadsheet. It has one table with formulas in cells and stored as a file

• database is fed automatically and many tables are linked. there are queries across all of it and is written in SQL and holds far more data than a spreadsheet

What is a database made of

collection of files organized as tables

• tables are made of records - rows

• rows are made of fields - columns

What is the difference between a flat file and relational?

flat file is a single table and relational is 2 or more tables with relationships betweent them

What is metadata

data about the data: what a column means, where it came from, who owns it, when it last changed

• shared key is a promise that the same name means the same thing. Metadata is where that promise is written down

What is the cost of bad data?

costs money and is a tax —> tax because usually taxes are paid by every piece of work downstream of it and a model trained on bad data ays them tax on eery answer it gives

What are the 6 dimensions of data quality?

completeness, validity, consistency, uniqueness, timeliness, accuracy

What are the 3 questions that decide whether cleaning can happen?

1) Will the organization accept a standard?

2) can the source system be fixed, or only the copy?

3) When 2 systems disagree, who decides which one wins?

What are the 3 ways to find a defect class?

1) Profile the column

2) Compare against a rule

3) Compare against another system

Then standardize it or flag and leave it

What does PowerQuery keep that editing cells by hand does not?

Every step in order. Applied steps is a procedure you can rerun on multiple (or updated files)

What are the 2 kinds of compute?

Deterministic and probabilistic

What is deterministic compute?

a person wrote the rule. the output is exact: same input, same output, every time

ex: a formula, a SQL Query, power query steps

What is probabilistic compute?

a model learned the pattern from past data. the output is an estimate and can be wrong

ex: a fraud score or a chatbot’s answer

How do we run analytics across millions of records?

machine learning and AI run all 4 types across millions of records

What does the AI landscape look like?

What are the 3 ways a machine learns?

Supervised, Unsupervised, and reinforcement

What is supervised learning?

learns from labeled examples

What is unsupervised learning?

finds structure with no labels

What is reinforcement?

learns from reward

Example of supervised learning?

Churn- learns from labeled examples

• Which customers are about to leave

• method: classification with a decision tree

• analytics: predictive

Check Fraud

• which deposited checks are forged?

• method: classification with deep learning

• anlaytics: predictive

Example of unsupervised learning

Recommendations- finds structure with no labels

• question: what will a shopper buy next?

• method: association mining

• Analytics: prescriptive

Example of reinforcement

Super Mario Bros- learns form reward

• how does the program learn to clear a level?

• method: trial and error, scored by reward

• Analytics: prescriptive

Why should we do deep learning and not a decision tree for finding which deposited checks are forged?

A decision tree asks about columns someone has already defined, like months as a customer or support calls.

A forged check gives itself away in the pixels: a digit written over, a signature slightly off, ink that does not match.

Nobody can list all of those as columns, so a deep-learning model has to find its own.

It is still supervised learning: every past check it learns from is labeled forged or genuine.

How does a neural network learn?

by guessing, checking the label and adjusting

• ex: looks at a past check and guesses. if the check’s labels says forged and the guess was wrong, it can strengthen or weaken the connections that led to the wrong guess

• 
  

Reinforcement learning learns from _______ not ______

reward; label

What is a large language model

an advanced type of artificial intelligence trained on massive amounts of text to understand, summarize, translate, and generate human-like language

• the model predicts the next work: learned from huge amoint of texts and given what came before, it predicts what word is likely to come next then the next

What is hallucination?

perceiving something that is not there

What is confabulation?

filling a gap with something plausible, without knowing what it is made of. this is what language models do

• this is part of how these models work

What are the 3 ways to cut confabulation?

1) give it the source and tell it to answer only from that. this is reterival augmented generation(RAG)

2) check any specific it could not have known: a name, a date, a number

3) ask it to say when it does not know

What is anthropomorphization?

talking about a model as if it were a person

• we say it “thinks” or “knows” but that makes it sound like it has intentions and beliefs

• what it does is predicts text

What is the difference between a chatbot and an agent?

A chatbot answers you while an agent plans steps, uses tools, and carries out a text

• agents reach those tools thorugh connectios and the common standard is the model context protocol(MCP)

What is explainability

whether a person can say why a model produced a particular output

Difference between IT and IS

• Information technology: hardware, software, and networks used to process information.

• Information system: people, processes, data, and technology working together to accomplish a purpose.

Why does the number of connections grow faster than the number of systems?

• With \(n\) systems, the maximum number of direct pairwise connections is \(n(n-1)/2\).

• Adding systems can dramatically increase integration complexity.

What is the difference between reconciling and cleaning data?

• Cleaning: Correcting or standardizing data defects.

• Reconciling: Explaining why different systems or reports produce different results.

• Two systems may disagree because they use different definitions, even when neither contains incorrect data.

How does information travel across the internet?

• Information is broken into smaller packets.

• Packets travel across networks, potentially taking different routes.

• They are reassembled at their destination.

What is the difference between internet and world wide web?

• Internet: Global network infrastructure that connects devices.

• World Wide Web: A service that uses the internet to access websites and linked resources.

Why do relational databases use multiple connected tables instead of one giant table?

• Reduces unnecessary duplication.

• Improves consistency.

• Makes information easier to update and maintain.

• Connects related records through keys.

What is database normalization, and why does it matter?

• Organizing data into related tables to reduce redundancy and avoid inconsistencies.

• Prevents problems when inserting, updating, or deleting records.

How do primary and foreign keys connect tables?

• Primary key: Uniquely identifies a record.

• Foreign key: References a key in another table to establish a relationship.

What are the three ways companies can create business value through APIs?

• Unbundling: Break business capabilities into reusable services.

• Outside-in: Design APIs around what their users need.

• Ecosystem: Connect businesses and partners to exchange value.

What is the "rule of 10" for bad data?

• As a rule of thumb, work involving flawed data costs 10 times as much as work involving good data.

• Bad data creates rework, delays, and lost productivity.

How does a decision tree classify data?

• Uses a series of if-then rules to divide data based on features.

• Each split attempts to create more homogeneous groups.

• The final leaf nodes determine classifications.

What is the bias-variance tradeoff?

• High bias: Model is too simple and makes systematic errors.

• High variance: Model is too sensitive to training data and overfits.

• Goal: Balance complexity to minimize errors on new data.