Overview
Artificial Intelligence has both scientific and engineering goals. Its scientific goal is to understand intelligence through computational models of learning, reasoning, perception, language, and decision-making. Its engineering goal is to develop systems that use these capabilities to solve practical problems. These objectives range from automating narrow tasks to investigating more general forms of intelligent behavior. The success of an AI system must be evaluated according to its intended purpose, operating environment, available information, constraints, and consequences.
Primary Goals
| Goal | Description | Objectives |
|---|---|---|
Develop Intelligent Machines | Create computational systems that can perform tasks requiring capabilities such as learning, reasoning, perception, communication, and decision-making. |
|
Understand Human and General Intelligence | Investigate the computational principles underlying intelligent behavior and develop models that explain selected cognitive processes. |
|
Enable Learning from Experience | Develop systems that can identify patterns and improve their performance using data, feedback, or interaction. |
|
Automate Complex Tasks | Use intelligent algorithms to perform repetitive, computationally intensive, or difficult tasks with appropriate levels of accuracy and reliability. |
|
Improve Problem-Solving | Develop methods for identifying, representing, analyzing, and solving problems with large search spaces, uncertain information, or complex constraints. |
|
Support Rational Decision-Making | Enable systems to select actions based on objectives, available evidence, estimated outcomes, and operating constraints. |
|
Develop Autonomous and Adaptive Systems | Build systems that can perceive their environments, select actions, and adjust their behavior within defined operational boundaries. |
|
Core Functional Objectives
| Objective | Explanation | Examples |
|---|---|---|
Perception | Enable machines to interpret information from cameras, microphones, sensors, and other input sources. |
|
Knowledge Representation | Represent facts, concepts, relationships, rules, and other information in computational forms that can be stored, queried, and used for reasoning. |
|
Reasoning | Use available information and formal or probabilistic methods to infer conclusions, evaluate hypotheses, and identify possible solutions. |
|
Learning | Acquire patterns, representations, or strategies from data and experience. |
|
Natural Language Communication | Enable systems to process and generate human language for communication and information exchange. |
|
Planning | Construct sequences of actions that can achieve a specified goal while considering available resources, dependencies, and constraints. |
|
Prediction | Estimate unknown or future outcomes using learned patterns, statistical relationships, and available information. |
|
Content Generation | Produce new content based on learned representations and input conditions. |
|
Scientific Objectives
Develop testable computational theories of intelligent behavior.
Understand the mechanisms and limitations of learning algorithms.
Investigate how knowledge can be represented and used for inference.
Study the relationship between perception, reasoning, memory, and action.
Develop formal methods for analyzing intelligent-agent behavior.
Explore the limits of machine learning, reasoning, and computational problem-solving.
Investigate whether different forms of intelligence can be explained through common computational principles.
Engineering Objectives
Design AI systems that solve clearly specified real-world problems.
Improve accuracy, robustness, computational efficiency, and scalability.
Develop models that can be integrated into existing software and hardware systems.
Create reproducible training, testing, deployment, and monitoring processes.
Reduce operational risks through validation, testing, and human oversight.
Ensure that systems satisfy relevant privacy, security, legal, and ethical requirements.
Human Centered Objectives
| Objective | Explanation |
|---|---|
Human Assistance | Develop systems that help people access information, analyze evidence, and perform tasks more effectively. |
Human-AI Collaboration | Combine human expertise with computational capabilities so that people and AI systems can contribute complementary strengths. |
Accessibility | Develop technologies such as speech interfaces, translation, captioning, and assistive systems that can improve access to information and services. |
Responsible Use | Ensure that AI systems are designed and used with attention to fairness, privacy, transparency, accountability, and potential social consequences. |
Performance Objectives
| Dimension | Explanation |
|---|---|
Accuracy | The system should produce correct predictions or outputs at a level appropriate for its intended application. |
Reliability | The system should operate consistently under expected conditions and handle failures appropriately. |
Efficiency | The system should use computational resources, time, memory, and energy appropriately. |
Robustness | The system should maintain acceptable performance under reasonable variations, noise, or unexpected inputs. |
Interpretability | Users and relevant stakeholders should be able to understand important aspects of system behavior when required by the application. |
Fairness | The system should be evaluated for unjustified differences in performance or outcomes across relevant groups. |
Safety | The system should avoid unacceptable risks and operate within clearly defined boundaries. |
Adaptability | The system should be able to respond appropriately to relevant changes in data, tasks, or operating environments. |
Goal Dependent Design
Description
AI objectives must be translated into measurable requirements before a system is developed. Different applications require different priorities, and optimizing one objective may create trade-offs with another.
Examples
| Application | Priorities |
|---|---|
Medical Decision Support |
|
Industrial Quality Inspection |
|
Educational AI |
|
Autonomous Navigation |
|
Important Tradeoffs
| Tradeoff | Explanation |
|---|---|
Accuracy versus Interpretability | A highly complex model may achieve strong predictive performance while being more difficult to interpret than a simpler model. |
Performance versus Computational Cost | Larger models may offer improved capabilities but require more memory, energy, processing power, and infrastructure. |
Automation versus Human Oversight | Greater automation can improve efficiency, but high-impact decisions may require human review and clearly defined accountability. |
Personalization versus Privacy | Using detailed personal information may improve customization while increasing privacy and data-governance risks. |
Adaptability versus Stability | A system that updates continuously may adapt to changing conditions but can also experience unexpected behavior or performance degradation. |