Back to blogs

AI, The reality behind the hype

Published on
Hero for AI, The reality behind the hype

AI: The reality behind the hype at Howest Tech & Meet

Following up on the previous session about Baserow, I recently attended another interesting Howest Tech & Meet. This time, the topic was one that’s impossible to ignore these days: Artificial Intelligence. The speaker started with a title that sounded a bit skeptical (“AI, de realiteit achter de hype” or “AI, the reality behind the hype”), but quickly clarified they are actually a big fan of AI’s potential. The main point? There’s a lot of “bullshit” being sold, and it’s important to understand what’s really going on.

What Do We Mean By AI?

The session started by grounding us in what AI systems aim to do:

  • Observe the world (through data)
  • Learn from that data
  • Reason based on learnings
  • Solve problems
  • Understand and use language

But how they do this can differ significantly. The speaker highlighted two main approaches:

This is the approach most people associate with AI today, powering things like ChatGPT. It’s inspired by how our own brains work with interconnected neurons.

  • How it works: It takes input data (like text, photos, videos) and processes it through layers of simulated neurons to produce an output.
  • The Catch: The speaker raised a critical question: What is the system actually learning? Often, we humans project understanding and intelligence onto what are essentially complex mathematical formulas. ChatGPT, for instance, doesn’t understand context in the human sense; it’s incredibly good at predicting the next word in a sequence based on the vast data it was trained on.
  • Prompt Engineering: This highlights why “prompt engineering” is so important, it’s about carefully crafting inputs to guide the system and critically interpreting its outputs.

2. Top-Down: The Knowledge Bank

This approach is less talked about in mainstream hype but is another valid way to build AI systems.

  • How it works: Instead of learning patterns from raw data, this method uses a pre-defined knowledge base, much like a database holds data. This base contains facts and rules about a specific domain.
  • Logic: These facts and rules are often described using logical programming languages, allowing the system to reason and draw conclusions based on established knowledge.

AI’s Place in Society: Hype vs. Reality

AI is often compared to previous technological shifts, like the calculator or the internet. The speaker suggested a key difference:

  • Internet: Helped us find information.
  • AI: Can help us analyze and interpret information, making it potentially more actionable.

However, this potential comes with significant dangers that fuel the hype cycle:

  • Overhyping: Attributing magical capabilities to AI leads to unrealistic expectations and silly applications (an AI-powered coffee machine was mentioned… why?!).
  • The 80/20 Trap: We often see demonstrations where AI works brilliantly (the 80%), but the failures or limitations (the 20%) are hidden. This gives a skewed perspective.
  • “Blinding” Output: AI tools like ChatGPT can produce text that looks incredibly convincing and well-written, even if it’s subtly wrong or nonsensical.
  • Human Laziness: We are biologically wired to take shortcuts. It’s easy to accept AI output without applying critical thinking or common sense.
  • Misuse: The power of AI can easily be weaponized for disinformation, scams, or other malicious purposes.

Because of these risks, regulation is becoming a hot topic, with efforts focusing on banning harmful AI applications and applying rules based on the level of risk different AI systems pose.

AI in Software Development: The Next Step?

The speaker also touched upon how AI fits into the evolution of programming:

  1. Microcode: Directly programming hardware.
  2. Machine Language: Programming the CPU using hexadecimal codes.
  3. Assembler: Using mnemonics (like SUB, MOV ) that translate to machine code.
  4. Higher-Level Languages: (C, Java, Python, etc.) Abstracting away the hardware details.
  5. Object-Oriented Programming: Structuring complex code using objects and classes.
  6. AI / Prompt Engineering: The newest step? Using natural language prompts to have AI generate code or entire program snippets.

AI tools like GitHub Copilot are already changing how developers work, acting as powerful assistants.

Taking notes during the talk.

Personal Reflection: Skepticism Meets Potential

Like the speaker, I find myself both excited and wary about AI. The session was a great reminder that while tools like neural networks are incredibly powerful pattern-matchers, we shouldn’t mistake correlation for causation or prediction for genuine understanding.

The discussion about the 80/20 rule and the “blinding” nature of AI outputs really struck me. As future developers and tech professionals, it’s crucial we don’t fall into the trap of blindly trusting AI output. We need to maintain critical thinking, understand the limitations, and use these tools responsibly.

Seeing AI positioned as the next step in programming’s evolution was also thought-provoking. It’s clearly becoming a powerful tool in our toolbox, but like any tool, knowing how and when to use it (and when not to) is key. It seems understanding the fundamentals, whether it’s bottom-up neural networks or top-down knowledge systems, is more important than ever to navigate the hype effectively.

Overall, a very grounding session that encouraged a balanced view: appreciate the power, but stay critical and aware of the limitations and risks.