Narrow vs General AI

• Artificial intelligence (AI) is now a key part of everyday life, often unnoticed in applications like Google Maps and Gmail's smart replies.

Narrow versus General AI

• Two main categories of AI are narrow (ANI) and general (AGI).
• Artificial Narrow Intelligence (ANI), or weak AI, specializes in performing specific functions in controlled settings.
• Artificial General Intelligence (AGI), or strong AI, exhibits open-ended, flexible intelligence across various domains, similar to human intelligence.
• Current AI is domain-specific and lacks the flexibility of AGI.
• There is a consensus that AI will eventually surpass human intelligence.
• A survey of AI researchers suggests a 50% chance of algorithms reaching human-level intelligence by 2060.

Artificial Narrow Intelligence

• Artificial Narrow Intelligence (ANI) can solve complex tasks faster than humans but is limited to its designed use cases.
• ANI systems cannot generalize knowledge from one domain to another, unlike the human brain.
• Voice assistants like Siri and Alexa combine multiple weak AIs to translate natural language and complete tasks but are limited to their trained capabilities.
• ANI demonstrates intelligence in complex problem-solving and single-task performance.

Artificial General Intelligence

• Artificial General Intelligence (AGI) is measured against human cognitive abilities, aiming to imitate sensory input and emulate the full range of human cognitive skills, with domain-independent generalization.
• AGI includes abilities currently seen in ANI, such as generalizing knowledge from one task to another domain, including motivation and volition.
• Some philosophical views require AGI to possess consciousness or self-awareness.
• Developing an AGI system would require:
  • Cognitive ability to function and learn in multiple domains
  • Human-level intelligence across all domains
  • Independent problem-solving ability
  • Problem-solving abilities at an average human level over multiple domains
  • Abstract thinking abilities without past experience
  • Cognitive skill to form new ideas about hypothetical concepts
  • Perception of the whole environment
  • Self-motivation and self-awareness
• Superintelligence extends beyond AGI, envisioning an AI system exceeding human cognitive capabilities through recursive self-improvement and remains abstract.

Application Areas

• AI applications have grown due to advances in computation and data storage and AI adoption.
• AI adoption means using AI capabilities like machine learning in at least one business function.
• Research on AI has increased, with a 34.5% growth in AI journal publications from 2019 to 2020.
• Natural language processing and computer vision are popular research topics.
• Key industries adopting AI include High Tech/Telecom, Automotive and Assembly, Financial Services, Business, Legal and Professional Services, Healthcare/Pharma, and Consumer Goods/Retail.

High Tech and Telecommunication

• AI is used to optimize, automate, secure, and maintain networks.
• AI facilitates predictive maintenance to resolve network issues proactively.
• Self-optimizing networks can accurately predict network anomalies.
• Big data aids in detecting network anomalies to prevent fraudulent behavior.

Automotive and Assembly

• Autonomous driving is transforming the automotive industry.
• Cars now have intelligent sensors for safety features like lane keeping and emergency braking.
• Sensors can detect technical issues and driver risks, like fatigue or intoxication.
• AI is applied for predictive maintenance and inefficiency detection in assembly.
• Computer vision is improving defect detection in assembly.

Financial Services

• AI algorithms detect and prevent fraudulent transactions and money laundering.
• Computer vision identifies counterfeit signatures.
• Robo-advisors provide investment recommendations based on user profiles, and AI applications help in portfolio optimization.

Business, Legal and Professional Services

• AI enhances efficiency in industries with repetitive tasks and paperwork with robotic process automation (RPA), automating routine tasks and reducing administrative costs.
• Natural language processing and computer vision enhance RPA with intelligent business logic.
• Big data technologies help extract information, using predictive analytics to identify market trends.
• Intelligent agents identify potentially fraudulent patterns, reducing risk and fraud in legal, accounting, and consulting practices.

Healthcare and Pharma

• AI-based systems aid in disease detection using symptoms, such as COVID-19 detection from cough recordings.
• Intelligent agents monitor patients and suggest optimal medication combinations to avoid side effects.
• Wearable devices track vital signs, providing advice and initiating emergency calls upon detecting critical anomalies.

Consumer Goods and Retail

• The consumer goods and retail industry uses AI to predict customer behavior through website tracking, which personalizes shopping recommendations and optimizes the supply chain.
• Market segmentation is now done street-by-street and improved natural language processing is used for chatbots and conversational interfaces.
• Well-developed artificial agents are used for customer retention and service.

Evaluation of AI Systems

• Evaluating AI systems is crucial as more businesses use AI to support or create new business models.
• Crucial to evaluate new systems, ensuring all datasets are independent with a similar probability distribution.
• Data is split into three sets for developing AI models:
  • Training data set: Used to fit algorithm parameters during training.
  • Development set: Used to evaluate and optimize the model, containing data not in the training set.
  • Test set: Used for final model validation, ensuring no overfitting, with previously unused data.
• Metrics are important for evaluating algorithm performance.
• Common metrics for binary classification tasks include accuracy, precision, recall, and F-score.
• Financial services use binary classification for fraud detection: transactions are classified as either fraudulent or not.
  • True positives (TP): Correctly classified fraudulent transactions.
  • False positives (FP): Non-fraudulent transactions incorrectly classified as fraudulent..
  • True negatives: Correctly classified non-fraudulent transactions.
  • False negatives: Fraudulent transactions incorrectly classified as non-fraudulent.
• Classification results are displayed in a confusion matrix.
• Accuracy is calculated as (TP + TN) / (TP + TN + FP + FN).
• Precision is calculated as TP / (TP + FP).
• Recall is calculated as TP / (TP + FN).
• F-score is calculated as 2 * (precision * recall) / (precision + recall).
• For classification tasks with more than two classes, metrics are calculated for each class and then averaged.

Explore the differences between Narrow (ANI) and General (AGI) Artificial Intelligence. ANI excels in specific tasks, while AGI aims for human-like, versatile intelligence. Discover current AI limitations and future possibilities.