Core Concepts of AI and Machine Learning
Artificial intelligence (AI) is the field of computer science focused on creating intelligent machines that can learn and solve problems. A key idea is that AI is about endowing computers with the ability to reason, discover and generalize meaning, or learn from past experience.1
Fiction versus Actual
For more than a century, we have created fictional depictions of artificial intelligence, be it violent like HAL in 2001: A Space Odyssey or Skynet in Terminator, logical and unemotional like Data from Star Trek the Next Generation, or more benevolent and playful like Ian Banks Spaceship AI.
Questions
- What are your favorite fictional AIs?
- How do these fictional depictions compare to real AI?
Narrow (or Weak) AI vs. General (or Strong) AI
Narrow AI refers to AI systems designed and trained for a specific task in a narrow domain. General AI, also referred to Artificial General Intelligence or AGI, refers to a hypothetical AI that would have the ability to understand, learn, and apply similar reasoning to situations as human intelligence. This type of AI would be able to perform or exceed any intellectual task that a human can.
Key differences:
- Scope: Narrow AI is limited to specific domains while General AI would have broad capabilities across multiple domains.
- Adaptability: Narrow AI has limited abilities to adapt to situations outside of its training while General AI would be able to adapt and learn in novel situations similar to humans.
- Consciousness: General AI could exhibit characteristics of machine consciousness while Narrow AI does not.
Later today we look at the existential risk that AGI could pose to humanity.
Importance of Data
The old computer science cliché GIGO (Garbage In, Garbage out) refers to a long-standing observation that the quality of software outputs is directly related to the quality of the incoming data or inputs.
For AI outputs, issues with training data include:
- Bias that reflect the prejudices from the data creators
- Insufficient amount of high-quality data
- Data privacy and security concerns
Andrew Ng, AI scientist and entrepreneur, is a strong advocate of "data-centric artificial intelligence", where more time is spent in curating, cleaning, and managing data sets for use in model training and validation, especially for workflows that may not have a large amount of data to begin with, in order to improve the outputs of these models.2
Machine Learning
Machine Learning is a subset of AI where algorithms learn from data without being explicitly programmed. Two major types of Machine Learning:
- Supervised - training a model with labeled datasets where examples are paired with labels or categories.
- Unsupervised - training a model without explicit labeled responses where the model tries to learn the underlying structure of the data.