Decision Support and Artificial Intelligence
IBM Watson
-90 IBM POWER 750 Servers on 10 racks -Each server has 4 processors -Each processor has 8 cores (2880 total cores) -16 Terabytes of RAM and 4 Terabytes of Clustered Storage -A single CPU machine would take 2 hours to answer a single question predictive aspect technical specs: 90 IBM, etc equivalent of 3000 processors RAM is usually smaller than storage this is not the case here because in order for the system to work quickly, everything needs to be loaded into ram -all data from storage gets loaded into ram (may get indexed or expanded in some other fashion but always in RAM instantly so the processor can grab exactly what it needs) pic-jeaprody IBM developed Watson (PR stunt) they taught it to play jeopardy against the top 2 all time jeopardy players
Expert Systems
-AI based Information System that applies reasoning capabilities to solve very specific problems. These systems utilize expert knowledge and can replace an expert in the decision making process. -Good for diagnostic (what's wrong?) and prescriptive (what to do?) problems focused on a specific problem can replace the expert in decision making whale watcher demo- if you saw a whale, it would give you questions so you could figure out what type of whale you saw point of replacing expert is to give everyone the knowledge the expert has
Genetic Algorithms
-AI based system that mimics the evolutionary, survival-of-the-fittest process to generate increasingly better solutions to a problem. -Selection - survival of the fittest. -Crossover - combining portions of good outcomes in the hope of creating an even better outcome. -Mutation - randomly trying combinations and evaluating the success (or failure) of the outcome. may find tis capability in an AI system or approach you can take for using AI based on 3 principles its job is to give you possible solutions you have to decide whether they're good solutions selection ex- stock prices- it would look for the best performing stocks during a period of time a tool to examen data and it will pull possible selections designed to give possibilities, not solutions
Geographic Information Systems (GIS)
-Computer system with software that can analyze and display data using digitized maps. Enables display and analysis of spatial information. -Examples - Location analysis, law enforcement, identifying efficient delivery routes system that will display data in a spatial fashion a map for visualization of data used in law enforcement ex. old crime movie, detectives put push pin on map where crime occurred and where to get bad guy he toured Erie county crime analysis center and he met a guy who tracks gang activity in buffalo and they mapped out specific areas, color coded based on activity -helps make decisions on where to emphasize patrols location analysis-determining where to out new business -you want to open a new duffs- what are the characteristics of a location? high traffic, accessibility, demographics (target audience), not too close to other locations, where is competition GIS is a map of data, overlapped w data like colors/symbols based on characteristics -visually displayed google maps has API where you can see where ex. all fish fry places open during lent Amherst has tax maps, flood places, had one for deer vehicle accidents- color coded different sections of Amherst where it was common
Expert Systems People
-Domain expert - provides the domain expertise in the form of problem-solving strategies. -Knowledge engineer -IT specialist who formulates the domain expertise into an expert system. gets inside the mind of domain expert and codifies it into system -Knowledge worker or user - that's you.
Artificial Neural Network (ANN)
-Emulates a biological neural network -Receives information from other neurons or from external sources, transform the information, and pass it on to other neurons or as external outputs -Useful for pattern recognition, learning, and the interpretation of incomplete inputs ex. throughout your body you have a neural network that controls signals that are sent to your body (moving, talking, breathing) near pathways are established in your body and they're developed since infancy over time they're burned in-takes training this idea has been adapted to a system where you have inputs you put into a signal, system does something w it and it tells you the likely output after enough training, body (or system) understands signal if you put input in it'll give you a business result good for pattern recognition, learning, and interpretation, etc because it doesn't have to have precise values -based on patters of things we know, what will this end up as
Commercial AI Systems
-Expert systems (ESs) -Neural Networks -Genetic Algorithms -Intelligent Agents -Natural language technology -Speech (voice) understanding -Computer vision and scene recognition -Intelligent computer-aided instruction -Handwriting recognizers has become commercialized -audio recognition confirming this is a pic of a car "I am not a robot drill" -trains googles system captions you have to type in w weird letters are old text from newspapers and it would present you w 2 phrases google only knew one of them -only technically needed half right to pass because it can only check what it knows -got more recognition for system by filling out these digitized phrases
DSS Examples
-General Accident Insurance: Customer buying patterns and fraud detection -Bank of America: Customer profiles -Frito-Lay:Price, advertising, promotion,selection -Burlington Coat Factory: Store location and inventory mix -Keycorp: Targeting direct mail marketing customers -National Gypsum: Corporate planning & forecasting -Southern Railway: Train dispatching and routing -Texas Oil & Gas: Evaluation of potential drilling sites -United Airlines: Flight scheduling, passenger demand forecasting most are about focusing on key business decisions drilling gas- want a system that gives you specific drilling sites
Collaboration System Benefits
-Improved preplanning -Increased participation -Open, collaborative meeting atmosphere -Criticism-free idea generation -Evaluation objectivity -Idea organization and evaluation -Setting priorities and making decisions -Documentation of meetings -Access to external information
expert systems components
-Knowledge base -Knowledge acquisition- a place to enter in the knowledge into the system -Inference engine -User interface -Explanation module
Neural Networks
-Self-organizing neural network - finds patterns and relationships in vast amounts of data by itself -Back-propagation neural network - a neural network trained by someone self organizing- starts w a pole of data and it starts recognizing patterns and trends within that data and what the results are has an algorithm it goes through, starts crunching it, over time it gets more refined back propagation- starts w training known inputs w known outputs, you feed those in to get a baseline of what is known already and neural networking can learn from that systems goal- are you goof or bad at credit risk? combos of these things put together makes you good or bad not binary- degree of how good/bad like w Watson, there is a percentage threshold- not usually absolute
DSS Capabilities
-Sensitivity Analysis -Study of the effect that changes in one or more parts of a model have on other parts of the model -What-if Analysis- of you have a spreadsheet to track your grades and use a formula to calculate GPA -what if I get an A in mgq 301 instead of a B-, how will it affect my GPA -Checks the impact of a change in the assumptions or other input data on the proposed solution -Goal-seeking Analysis- not trying to find the answer- you already know it. you're trying to find the inputs that will give you the goal ex. can build an excel that will show all combos of grades to make a 3.5 GPA- can have greater variance or all can be around that range not giving you the answer- gives you the options and you decide which is the most reasonable, likely you have to understand and interpret the results -Finds the value of the inputs necessary to achieve a desired level of output capabilities- DSS software system, models, OLAP looks, data mining; open ended tools to help you understand the data database comes fro, TPS and external data need best possible data in transactional system so you can make best decisions model- the things you'll use to analyze data data- often you will have a database excel has sensitivity, what if, goal seeking sensitivity- it tells you how much changing one part of the model will impact other parts marketing perspective- you have a bunch of marketing initiatives and you can see how changing certain factors in them has impacts on other things understanding the degree of impact is important
Types of Decisions
-Structured -Semi-structured -Unstructured -Recurring vs. ad-hoc structured- decision that has one right answer that is formulaic, clear, and calculated. probably several wrong answers unstructured- problem w many possible solutions. some are optimized better than others. challenge is finding the best answer, not the only answer. open-ended, not as clear, many possibilities semi structured- several right answers but not as many as unstructured. what is the best answer of those choices ex. he gets a sweater for Christmas and goes to the store to exchange it. customer service rep makes a decision based on request, typically based on store . the store policy is you need a receipt. he asks if there are any other options because it was a gift so clerk asks a manager. manager decides from this semi-structured problem that they will give store credit. manager could've said no or could've given cash for it. manager attends a meeting w other managers and they discuss the gift return issue. company doesn't have a good enough policy to handle those issues- waste of time to keep calling managers for help. managers are faced w an unstructured problem and they decide to rewrite store policy. have to focus on profitability and customer satisfaction. structured- clerk says no returns without receipt semi structured- store credit unstructured- rewrite policy different employees have different info needs based on the decisions they make and the systems we build can help inform these types of decisions differently story: former student said he worked at Home Depot or Lowes and customers tried to return Christmas trees after christmas -businesses have to weigh the costs of their policy of letting people do that recurring- decision that happens over and over good because you know they're coming and can be prepared for them info systems generate reports on certain schedules provided in advance to provide the info thats needed to make a good decision -you can plan for these -can use automated reports/notifications to help employees make decisions
DSS process
1. bob says what financing option will cost the least to buy the house when the principle and interest are all paid? what option will result in the lowest monthly payments? 2. question goes to user interface management 3. choice of model goes to model management 4. needed info goes from data management back to model management 5. model results foes from model management to interface management 6. answer goes to user 7. bob says now that I've figured out the total cost of all the loan options and also the payment for each of the loans I can use that info to make the final decision on financing ex. DSS is helping user determine finance options for buying a house. he chooses a model (amortization model)- shows financing costs -can be build into a system info you need for amortization model- term of loan, rate, how much $- will help you answer initial Qs data management may have data on different loan products that are refreshed daily (up to date data) optimization model- optimizing lowest costs
neural network example
20 Qs finds answers based on what others have answered it asked if its yellow, orange, green AI doesn't think like human brains contradictions: we said orange isn't bumpy, most people said it was based on these inputs (the answers to these questions) it finds a pattern of answers that matches w a result it knows every time you play it it gets changed for contradictions, we slightly changed the confidence rate for these answers quick draw able to recognize visually based on a neural network (patterns of lines, etc) every entry helps it understand what each thing might look like this person does not exist- combo of a bunch of people using AI sometimes it cracks, something might look blurred or messed up there are programs to detect the generators so they dont use it for bad
Expert Systems cont
An expert system can: -Reduce errors -Improve customer service -Reduce costs An expert system can't: -Use common sense -Automate all processes whether rules are right or wrong they are going to apply the inputted rules a system you work with
AI systems
Artificial Intelligence (AI) -Branch of computer science that deals with ways of representing knowledge, using symbols rather than numbers, and heuristics, or rules of thumb, rather than algorithms for processing information Objectives: -Make machines smarter -Understand what intelligence is -Make machines more useful AI- if the computer does something you deem to be intelligent
Decision Support Systems (DSS)
Highly flexible and interactive IT system designed to support unstructured and semi-structured decision making. what you bring- experience, intuition, judgement, knowledge advantages of a DSS- increased productivity, increased understanding, increased speed, increased flexibility, reduced problem complexity, reduced cost what IT brings- speed, information, processing capabilities doesn't support structure interactive because its open-ended and you can work with it and analyze to find best answer -reliant on you (the expert) working with the system and data cant sit in front of system, press a button and get an answer. you need someone who knows what they're doing to get the value out of it intersection of what you bring and what tech brings= advantages of a DSS ex. OLAP system where you're manipulating, slicing and dicing data and analyzing
DSS Components
TPS and external data -> DSS database -> DSS software system, models, OLAP tools, data mining tools -> user interface -> user -Model management- consists of both the DSS models and the DSS model management system. -Data management- performs the function of storing and maintaining the information that you want your DSS to use. -User interface management- allows you to communicate with the DSS. user interface- typically graphical maybe mobile
ad-hoc decisions
decisions that happen on the spot you want info systems flexible enough to help you find answers for these decisions (analysis, query, report)
Inference Engine
expert system that will advise people how to approach a sales process rules have been established- IE will apply to this -income level -insurance -what actions should be taken (real life rep or brochure) want to make the most cost effective decision if they have real estate/assets-how do you deal with it (if statement)
Overview
how decisions are made? central to MIS
Four Phases of Decision Making
intelligence- finding problem to fix design- find solutions choice- choose a solution (weighing pros and cons, costs, etc) implementation- apply the solution can build a system on any one of those phases can help inform the types of systems we use in an organization
Intelligent Agents (IA)
personal digital devices that listen to you paper clip guy on Microsoft word- ex of AI when he was writing a syllabus it would say "I see you're writing a letter, let me help you" -Software that assists you, or acts on your behalf, in performing repetitive computer-related tasks. -Four types of intelligent agents include: -Buyer agents or shopping bots -User or personal agents -Monitoring-and-surveillance or predictive agents -Data-mining agents Autonomy - act without your telling them every step to take. work on their own, dont need exact prompting Adaptivity- discovering, learning, and taking action independently. because they're connected online so they're constantly being updated and connected to updated info Sociability - conferring with other agents.
Expert Systems Process
step 1: domain expert and knowledge engineer domain expert provides expertise to the knowledge engineer knowledge engineer converts domain expertise into rules -> knowledge acquisition -> knowledge base works to get the expertise built into the system step 2: user user interface delivers problem facts to the expert system and sends the conclusion back to the user user interface -> inference engine -> explanation model and repeat inference engine= the brain -logical rules that help them decipher the answer to your question IE uses knowledge base (database of knowledge) specifically related to the area of expertise the person has entered in results are shared sometimes theres an explanation module which explains how it got to the answer knowledge engineer- gets inside the mind of the expert and break down their problem solving process in such a way that it can be codified in to the system logical breakdown of a problem- tree diagram
Collaboration Systems
under umbrella of DSS -Interactive computer-based system that facilitates the solution of semi structured and unstructured problems by a group of decision makers. -Tools include - Electronic questionnaires, Electronic brainstorming tools, Idea organizers, Questionnaire tools aka a group decision support system when you make a decision as a group it is different than making a decision as an individual therefore, group decision making system is different than regular ex. file exchange on ublearns google doc- shared space -electronic so they document everything some people are extraverted and some are introverted- some are vocal about contributions and some arent -these systems give quiet people more of a voice gives people a chance to process their thoughts can capture a greater range of thoughts from different people
Watson video
when Watson answers there shows 3 possible thoughts he has and the confidence level it has in each answer Watson took the lead then started getting answers wrong- went into tie w brad Watson killed it during first double jeopardy final jeopardy- you need to pull things from 2 different places US cities category with these 2 airports both guys said the right answer and Watson said Toronto- confidence level was low second game Watson was getting all the answers but a little too slow Watson wo reveals deeper understanding of our own intelligence point is to digest unstructured data (not what were used to) if you dump all sorts of data into watson it'll start to catalog and organize it you then can start asking it questions