Quartz 4

week1-was-ist-ki

8 min read

Type: lecturenote

Up: 025_introduction_to_ai Next: week2-ki-systeme

Was ist KI?

Lecture Notes:

Goal of this lecture: Help you understand AI marketing hype vs. real AI. Spot what’s real and what’s not.

What is Artificial Intelligence?

  • Artificial Intelligence is a field of computer science that studies how to make machines act intelligently.
    • It asks: How can we make computers or systems behave in ways that seems intelligent to humans?
  • AI is not only about robots or ChatGPT. It’s about systems that can act, and make decisions in complex environments.
    • Also: AI is not just Deep Learning. Deep Learning is one part of AI, but AI also includes reasoning, understanding, explaining, and sometimes creativity.
  • AI can be seen as the avant-garde of computer science — the most forward-looking part of computing — because it connects to many other disciplines:
    • Cognitive Science 🧠 (studying human thinking)
    • Psychology 🧩 (understanding human behavior)
    • Philosophy 💭 (asking what intelligence and consciousness mean)
    • Engineering ⚙️ (building systems that act in the real world)
  • 💡 Key Idea AI ≠ only Neural Networks.
    AI includes logic, reasoning, problem solving, learning, and perception.

The Problem of Defining Intelligence

  • There is no single definition of intelligence — and that’s one of the biggest problems in AI.
    • Even in psychology, philosophy, and everyday life, people disagree on what “being intelligent” means.
      • Some say it’s about thinking logically.
      • Others say it’s about learning from experience, or adapting to new situations.
  • So, if humans cannot define intelligence clearly, how can we define artificial intelligence?
    • Some researchers even claim that “there is no real definition of AI”.
    • But this is lazy thinking. We do have working definitions — maybe not perfect, but useful. We don’t stop studying biology just because we can’t define “life” perfectly.
  • Also: We should not think about intelligence only in human terms. A system doesn’t have to think like a person to be considered intelligent.
  • Chinese Room Argument: This philosophical idea by John Searle asks whether computers really understand or just manipulate symbols.
    • You sit in a room with a rulebook for Chinese. You don’t know Chinese, but you can look up how to reply correctly to messages. To people outside, it looks like you understand — but you don’t.
    • The argument says AI might work the same way — it produces correct answers without true understanding.
    • But maybe the whole system — not each part — has understanding. That means we might be asking the wrong question.

Classic Definitions of AI

Here are two important early definitions from AI pioneers:

  • John McCarthy (1950s): “The science and engineering of making intelligent machines, especially intelligent computer programs. It is related to the similar task of using computer to understand human intelligence, but AI does not have to confine itself to methods that are biologically observable.”
    • McCarthy focused on creating intelligent behavior, not just studying human minds.
    • For him, AI is not about copying humans — it’s about understanding intelligent behavior in general.
  • Marvin Minsky: “The science of making machines do things that would require intelligence if done by men.”
    • This definition uses human as a reference. It means: if a human needs intelligence to do something, then a machine doing the same thing can be called AI.
  • It is OK to use humans as a model, but AI doesn’t have to behave exactly like humans.
    • Example: Autonomous driving should not imitate all human mistakes.
  • 💡 Key Idea AI is not about copying humans — it’s about achieving intelligent behavior by any method.

General AI vs Narrow AI

AI can be divided into two main categories:

  • 🤖 Narrow AI (Weak AI)
    • Designed for one specific task (playing chess, recognizing races, translating text).
    • It doesn’t understand or learn beyond that task.
    • Most of today’s AI systems — including ChatGPT, image generators, and self-driving algorithms — are narrow AI.
  • 🌍 General AI (Strong AI / AGI)
    • A system that can understand, learn, and act in many different areas.
    • It could solve any problem, not just one type.
    • This would be like a machine with broad, human-level intelligence.
  • We are still far from true General AI — maybe decades, maybe never.
    • However, tools like ChatGPT make us feel closer to it, because they can handle many types of tasks with one system.
  • General AI would need to be
    • Adaptive → able to learn from new experiences.
    • Autonomous → able to make its own decisions.
    • Reflective → able to think about what it has learned.
  • Discussion: If a system becomes too autonomous, should it be allowed to make decisions on its own? Who is responsible then — the machine or its creators?
  • 💭 Reflection:
    A “smart” AI is not only about data and algorithms — it’s also about autonomy, responsibility, and ethics.

Machine Learning, Deep Learning, and the Internet

  • Machine learning means teaching computer to learn from examples instead of giving them step-by-step instructions.
    • You don’t tell the computer how to do the task — you show it many examples and let it find patterns.
  • Example: Show a system thousand of photos labeled “cat” and “dog”. Overtime, it learns to tell which is which — without you explaining what “ears” or “tails” are.
    • This approach is powerful but limited. It works well only when there are many examples. It can’t easily handle new or unexpected situations.
  • That’s where Deep Learning (DL) comes in.
    • DL uses neural networks — computer structures inspired by the brain — to learn more complex patterns.
      • However, DL is not all of AI. AI also includes reasoning, explaining, planning, and understanding, not only recognizing patterns.
  • The key reason why Deep Learning succeeded is the Internet 🌐.
    • Because the internet created huge amounts of data, AI could finally train on billions of examples.
    • So AI’s progress is not magic — it’s the result of data, computing power, and connectivity.
  • 💡 Key Idea Without the Internet, we would not have modern AI.

Rationality and Human Behavior

  • Rationality — the idea of acting logically or making decisions that make sense. “If AI should act intelligently, should it also act rationally?”
  • 🤔 What is Rationality?
    • To be rational means to make the best possible decision based on the available information.
      • For example: You have two paths. Path A is safe but slow. Path B is risky but faster. A rational choice depends on what you value more — safety or speed.
    • AI systems also make decisions like this, but often using mathematical models. Economists and computer scientists both use these models to simulate rational behavior.
  • However, humans are not always rational. We are emotional, biased, and unpredictable. So, if AI tries to copy us, it might also copy our mistakes.
    • Humans are the model for AI — but not always a good one.
  • Think about COVID-19 crisis: Governments made rational decisions based on what they knew at the time, even if those decisions later turned out imperfect. That shows the difference between rational and optimal.
    • Rational: Best choice with the information you currently have.
    • Optimal: The absolute best choice (requires knowing everything)

Ethical and Social Questions

  • AI is not only a technical problem — it’s also a social and ethical challenge.
    • AI influences people’s lives, jobs, and emotions.
      • A woman lost her job as an illustrator because of AI image generators like Stable Diffusion.
      • Some researchers claim AI systems might someday have feelings or consciousness, like Google’s “LaMDA” model — but that raises serious moral questions.
  • AI cannot be developed without ethics. Every AI researcher has a social responsibility — to understand how technology affects humans, culture, and the environment.
  • We have to be careful about the idea of Effective Altruism which tries to measure moral value with numbers and decide whose life matters more. Such thinking can be dangerous when combined with AI decision system.

The History of AI: From Hype to Reality

  • AI didn’t start with ChatGPT — it began in 1956 at the Dartmouth Workshop, where researchers first used the term Artificial Intelligence.
    • They were very optimistic: “With the right team, we can solve all problems of language ,vision, and learning in one summer” — they couldn’t.
  • After that, AI went through periods of disappointment, called AI Winters. Funding stopped because progress was slow and expectations were too high.
    • In the 1970s, people lost interest after discovering that neural networks had serious mathematical limits.
    • Later, in the 1980s and 90s, interest came back through logic-based AI, then declined again.
    • Finally, in the 2010s, AI became strong again through machine learning and deep learning.
  • The main reason:
    1. More data (the internet)
    2. Faster computers (GPUs)
    3. Better algorithms
  • Hype cycle: Excitement → Disappointment → Real progress → Productivity”

Graph View

Backlinks