Which of the following best represents a properly defined and scoped key performance indicator (KPI)?
Reduce customer complaints.
Increase customer satisfaction for the month of May.
Increase phone sales by 15% this year.
A properly defined key performance indicator must be specific, measurable, and bound by time, with a quantified target that allows progress to be tracked objectively. "Increase phone sales by 15% this year" satisfies each element. It identifies the subject (phone sales), supplies a numeric goal (15 percent), and specifies the period (this year). Anyone reviewing results at year end can state unambiguously whether the objective was met, and managers can monitor interim progress against it.
"Reduce customer complaints" names a direction but provides neither a magnitude nor a deadline. Without a target value, a reduction of one complaint would technically satisfy it, and without a time frame no point exists at which success or failure can be judged.
"Increase customer satisfaction for the month of May" supplies a time frame but lacks a quantified target and a defined measure of satisfaction. It does not state by how much satisfaction should rise or by which instrument, such as a survey score, it will be measured. The window is also narrow, which gives limited room to evidence a trend.
Only the phone sales statement combines metric, target, and period. This reflects the SMART criteria widely used when scoping indicators for AI initiatives, and the same discipline allows the value of an AI project to be demonstrated afterward.
Among the non-probability sampling methods, which one is the MOST susceptible to bias?
Judgment
Convenience
Quota
Convenience sampling is the non-probability method most vulnerable to bias. It selects whichever individuals are easiest to reach, such as passersby, volunteers, or readily available records, with no structure intended to make the sample reflect the wider population. Those who are accessible usually differ systematically from those who are not, so findings may be skewed and cannot be generalized with confidence. Because the process offers no safeguard against self-selection or availability effects, it carries the highest risk of an unrepresentative result.
Judgment sampling, also called purposive sampling, relies on a researcher's expertise to choose elements considered typical or informative. It is subjective and can introduce the researcher's own bias, but the deliberate selection criteria give it some degree of control over relevance, so it tends to be less exposed than a purely opportunistic approach.
Quota sampling sets targets for subgroups in proportion to known population characteristics, such as age and gender, and the sample is filled until each quota is met. Although the individuals within a quota are still chosen non-randomly, the structure compels representation of key segments and reduces the imbalance that unstructured selection creates.
Only the convenience approach has no built-in mechanism for balance, which makes it the most susceptible to bias. It is nonetheless popular for exploratory work because of its speed and low cost.
Which category of data is most typically augmented through methods such as pitch shifting and time stretching?
Text
Audio
Images
Audio is the data category most typically augmented through pitch shifting and time stretching. Both are signal-processing operations. Pitch shifting raises or lowers the perceived frequency of a sound without changing its duration, while time stretching speeds up or slows down a recording without altering its pitch. Applied to speech or environmental sound datasets, they create realistic variations that resemble different speakers, vocal ranges, and speaking rates. Training on these variants helps speech recognition and sound classification models cope with the diversity they will meet in deployment, and it increases the effective size of the training set without recording new material.
Text augmentation uses different methods because language is discrete and symbolic. Typical techniques include synonym replacement, random insertion or deletion of words, sentence reordering, and back-translation through an intermediate language. Frequency and tempo are meaningless concepts for written characters.
Image augmentation relies on geometric and photometric transformations such as rotation, flipping, cropping, scaling, and changes in brightness or contrast. These operate on pixel arrays and have no relationship to sound waves.
Since pitch and time are properties of acoustic signals, audio is the data type to which these two techniques apply. Matching an augmentation method to the nature of the data is necessary to ensure that transformed samples remain valid and labels remain accurate.
Given that the ethical framework surrounding AI is still in its early stages of development, which of the following represents the greatest concern?
Financial responsibility
Accountability
Intellectual property
Accountability is the foremost concern while the ethical framework for AI remains immature. Established norms, statutes, and professional standards have not yet settled who answers when an AI system harms someone, reaches a discriminatory decision, or simply errs. The candidates include the developer who built the model, the organization that deployed it, the vendor that supplied training data, and the operator who relied on its output. The problem is aggravated by autonomy and by the opacity of many models: if nobody can fully explain why a system produced a result, assigning responsibility and ordering remedies become far more difficult.
Absent clear accountability, affected individuals lack recourse, regulators lack enforcement levers, and organizations lack incentives to invest in safe design. Trust in AI deployments erodes accordingly. Resolving the gap requires governance structures, documented decision rights, audit trails, human oversight, and, where appropriate, legal liability rules.
Financial responsibility is a narrower matter, since the allocation of costs and damages follows once accountability has been determined. Intellectual property is a real and active area of debate, involving ownership of AI-generated works and the use of copyrighted material in training. It concerns rights over assets and does not address responsibility for harm.
Because both alternatives flow from the broader question of who is answerable, accountability constitutes the greatest concern.
Which of these approaches builds a predictive model by training it on input data?
Expert system
Artificial intelligence
Machine learning
Machine learning is the approach in which algorithms are fitted to input data so that they can generalize and generate predictions on new observations. The training process extracts statistical relationships from historical examples, and the resulting artifact, the model, is what performs forecasting or classification once deployed. Credit scoring, demand forecasting, and fraud detection are common illustrations.
An expert system follows a different design. Its behavior comes from rules and facts that human specialists articulate in advance and that engineers encode into a knowledge base. The system applies those rules through an inference engine, but it does not improve by observing data, so it does not build a predictive model through training.
Artificial intelligence is the broadest term in the list. It covers any technique that lets a machine imitate aspects of human intelligence, including rule-based systems, search, planning, robotics, and machine learning itself. Because the term spans so many methods, it does not name the specific activity of training on input data.
When a question asks which approach builds a model from data, the discriminating phrase is "training on input data." That phrase identifies machine learning as a subset of AI, distinguished from hand-authored logic by its data-driven origin of knowledge.
Which of the following serves as a reporting tool that enables the appropriate allocation of organizational needs?
Governance
Balanced scorecard
Business intelligence
The balanced scorecard is a strategic performance management and reporting framework. It converts an organization's mission and strategy into measurable objectives arranged across four perspectives: financial, customer, internal business processes, and learning and growth. Reporting results against these perspectives gives leadership a consolidated view of where the organization is succeeding and where it falls short, and that visibility supports deliberate allocation of budget, staff, and attention to the areas of greatest strategic need. This is the sense of a reporting tool that enables the appropriate allocation of organizational needs.
Governance refers to the structures, policies, roles, and accountability mechanisms through which an organization is directed and controlled. It establishes decision rights and oversight, but it is a system of authority rather than a reporting instrument. A scorecard may feed governance, though the two are not the same thing.
Business intelligence comprises the technologies and processes used to collect, integrate, analyze, and visualize data in support of decision-making. It is broader and more technical than the scorecard, covering dashboards, queries, and analytics of every kind. It does not by design tie measures to strategic objectives or serve as the framework for allocating resources against those objectives.
The scorecard's explicit linkage between strategy, measurement, and resource allocation distinguishes it from the other two options and marks it as the correct answer.
Before adopting AI technology, which of the following should be the very first consideration?
Identifying the problem AI can solve
Ensuring AI training is available
Defining key performance indicators (KPIs)
The first consideration before adopting AI is identifying the problem the technology is meant to solve. Successful projects begin with a clearly articulated business need, for example reducing churn, speeding claims processing, or improving defect detection, because that definition determines everything downstream. It guides whether AI is appropriate at all, which type of model or tool fits, what data must be collected, and how the effort will be justified. Organizations that start from the technology and look for a use afterward tend to produce expensive pilots that never reach production.
Defining KPIs is an essential step, but it follows problem identification. Indicators measure progress toward a goal, so they can be specified only once the goal itself is understood. Choosing metrics first would risk measuring the wrong outcome.
Ensuring AI training is available relates to workforce readiness. Staff need appropriate skills to operate and oversee AI systems, but the type and amount of training depend on which solution is selected, which in turn depends on the problem being addressed.
The sequence is therefore logical: define the problem, set measurable success criteria, assess data and skills, and then select and implement a solution. Placing problem identification first keeps AI investments aligned with real organizational value.
Which of the following illustrates the use of AI to help consumers find new or former relationship connections?
Apps that use facial recognition
Apps that use a global positioning system (GPS)
Apps that look at possible connection proximity
Apps that use facial recognition illustrate how AI can help consumers locate new or former relationship connections. Computer vision models detect a face in an image and convert it into a numerical representation, then compare that representation with others to find matches. A user can upload a photograph and receive suggestions for the same person elsewhere or for people who resemble known contacts, which allows reconnection with an old acquaintance or discovery of a potential new one. The matching process depends on trained neural networks, so the example qualifies as direct use of AI for this purpose.
GPS apps determine location through satellite signals and trilateration. They calculate position and routes using established geometry and do not rely on AI to identify people or judge relationships. Location data may be consumed by other services, but the positioning technology itself is not the AI element.
Apps that look at possible connection proximity use distance or shared-venue information to suggest nearby people. Proximity is a simple measurement derived from location and does not identify or match individuals through learned pattern recognition, so it does not illustrate AI as clearly as the recognition option does.
Recognition of a specific person's face is the capability that links an individual to a possible relationship connection, which makes it the best illustration. Privacy and consent concerns accompany its use.
Which of the following is an example of video synthesis?
Object resolution
Face-swapping
Image classification
Video synthesis refers to using machine learning, typically generative models, to produce or alter video so that the result looks authentic. Face-swapping is the clearest instance: a model learns the facial characteristics of one person and renders them onto another person's movements across successive frames. Autoencoders and generative adversarial networks power most implementations, and the output is the basis of what is commonly called a deepfake.
Object resolution is not a synthesis task. The term relates to the level of detail at which an object can be distinguished in an image or sensor reading, and improving it is a matter of enhancement or upscaling rather than creating new footage of events that never occurred.
Image classification is an analytic task in computer vision. A trained model assigns a label such as "cat" or "stop sign" to an existing image. It interprets content that already exists and does not generate or manipulate video.
The distinguishing test is direction of flow. Analytic tasks take media in and return a label or measurement, while synthesis tasks return new or altered media. Face-swapping fits the second category, and its risks, including impersonation and disinformation, explain why synthetic media features prominently in AI ethics and security discussions.
An organization wants to develop measurements to assess the efficiency levels of its organizational processes. What serves as the starting point for describing the current state of the item being measured?
Lead indicator
Benchmark
Internal process perspective
A benchmark supplies the baseline, a documented reference describing where a process stands today. Before an organization can say that efficiency has improved or declined, it needs a starting measurement against which later results are compared. The benchmark may derive from the organization's own historical performance, an industry standard, or a peer comparison, but in every case it anchors the current state and gives subsequent metrics meaning. Without it, a figure such as "processing takes six days" cannot be interpreted as good or bad.
A lead indicator is a predictive measure that signals future performance, for example the number of sales calls made as a precursor of closed deals. It looks forward, so it does not describe the present condition of the item being measured.
Internal process perspective is one of the four perspectives of the Balanced Scorecard, alongside financial, customer, and learning and growth. It is a lens for deciding what to measure, specifically the efficiency and quality of operations, and not a record of the current state of any individual measure.
The question asks what serves as the starting point for describing the current state, and this is the defining role of a benchmark. Establishing it comes first when developing performance measurements.
Which of the following is an example of probability sampling?
Sampling fraction
Cluster
Judgment
Probability sampling gives every member of the population a known, non-zero chance of selection, and the selection is made by random process. Cluster sampling qualifies. The population is divided into naturally occurring groups, such as schools, city districts, or store locations, and a random selection of whole groups is drawn. Everyone within the chosen clusters is then included, or a random subset of them is taken. Because chance governs which clusters appear, the technique supports statistical inference about the wider population and is particularly economical when members are geographically dispersed.
Judgment sampling is a non-probability method. The researcher decides, based on expertise, which elements are most representative or informative. Since selection is subjective and the inclusion chance of each element is unknown, results cannot be generalized with calculable error, and the approach is vulnerable to the researcher's own bias.
Sampling fraction is not a sampling method at all. It is the ratio of the sample size to the population size, for example 200 out of 10,000, or two percent. It describes how large a sample is compared with the population and tells nothing about how the sample was chosen.
The deciding characteristic is random selection with known probabilities. Of the options provided, only cluster sampling meets it, alongside simple random, systematic, and stratified sampling.
How would you define machine learning?
a predetermined set of rules.
a type of automation.
a predictive model.
Machine learning is best defined as a predictive model: a mathematical representation learned from data that maps inputs to likely outputs. Rather than following instructions a programmer wrote for every case, the algorithm infers patterns from examples and encodes them in model parameters. When new data arrives, the model applies those parameters to produce a prediction, such as a price estimate, a category label, or a probability of default.
The first alternative, a predetermined set of rules, describes conventional programming and rule-based expert systems. There the logic is authored by people ahead of time and stays fixed until someone edits it. That is the very approach machine learning was developed to move beyond.
The second alternative, a type of automation, is too broad to count as a definition. Automation covers any technology that performs a task without continuous human effort, including a scheduled batch script or a factory conveyor with no learning component. Some machine learning systems do automate decisions, but automation does not capture what makes the technique distinctive.
The defining feature is that the system's behavior emerges from training data and can be refreshed by retraining as data changes. A definition built on prediction from data therefore separates machine learning from both fixed rules and generic automation.
Expert systems rely on a knowledge base that is built through several stages, including elicitation of knowledge, coding, testing, and which of the following?
Call centers
Knowledge fusion
Ethical modules
Constructing an expert system's knowledge base proceeds through several stages. Knowledge is first elicited from human specialists through interviews, observation, and review of documentation. It is then coded into a machine-usable form such as if-then rules, frames, or decision trees, and tested to confirm that the resulting rules produce correct conclusions. Because expertise is frequently drawn from more than one specialist and more than one source, a further stage is needed to reconcile overlaps and conflicts and merge the material into a single consistent body of knowledge. That stage is knowledge fusion. It resolves contradictory advice, removes redundancy, and ensures the combined knowledge base operates coherently when the inference engine uses it.
Call centers are customer-contact operations. An organization may use an expert system within a call center to support agents, but a call center is a business setting and not a step in building a knowledge base.
Ethical modules do not form part of the established knowledge engineering process. Ethical considerations should certainly inform design and governance, yet no recognized stage of knowledge base construction carries that name.
Knowledge fusion is the stage that completes the sequence described, which is why it is the correct answer. Quality at this stage strongly affects the reliability of an expert system's advice.
As an AI system gains greater autonomy and expands its capacity to learn without human supervision, which of the following represents the GREATEST concern?
Its treatment of humans
Its impact on the job market
The intrusion on individuals’ privacy
As an AI system becomes more autonomous and learns without human supervision, the greatest concern is how it treats humans. Removing people from the loop means nobody reviews, corrects, or constrains the behaviors the system develops. Unsupervised learning identifies patterns on its own and can reinforce unintended, biased, or unsafe ones. The system might then discriminate against groups, make harmful decisions about individuals, or pursue objectives in ways that conflict with human welfare, and no check would exist to intervene before consequences materialized. The more autonomy and capability the system has, the more severe and less reversible such outcomes may become, which raises questions of safety, alignment with human values, and control.
Impact on the job market is a real and significant social issue, but it is a broader economic effect of AI adoption in general. It does not depend on the system's autonomy or its ability to learn unsupervised, and it concerns aggregate employment, not direct conduct toward people.
Intrusion on individuals' privacy is also a serious risk, since autonomous systems may collect or infer personal information. It is, however, one particular form of harm among the many that flow from an unchecked system's treatment of the people it affects.
Because unsupervised autonomy threatens how the system behaves toward humans at the most fundamental level, this represents the greatest concern.
Using AI to tailor pricing and personalize advertisements in real time is defined as which of the following?
behavioral AI.
superintelligence.
targeted marketing.
Targeted marketing is the use of data analysis and AI to tailor offers, prices, and advertisements to individuals or defined audience segments according to their behavior, preferences, and characteristics. Real-time personalization is a core illustration. As a consumer browses, algorithms evaluate signals such as location, purchase history, device type, and time of day, and then adjust the price shown or select the advertisement most likely to convert. Dynamic pricing in retail, travel, and ride services and individually curated ad feeds are common deployments. The description in the question, tailoring pricing and personalizing advertisements in real time, is a direct definition of the practice.
Behavioral AI refers to the use of machine learning to model patterns of human activity, as in detecting anomalies in how users act on a network. It is a method that can feed many applications, and it is not the name for the commercial practice of customizing prices and ads. The phrase is not a standard term for this practice.
Superintelligence is a hypothetical stage at which AI would surpass human cognitive ability across virtually all domains. It has no connection to personalized marketing and does not exist today.
Targeted marketing is therefore the precise term. Because it depends on extensive personal data, it also raises privacy and fairness considerations.
TESTED 22 Nov 2026