IB Digital Society 3.6D Evolution of AI Notes

3.6D Evolution of AI

Fiction, Philosophy and the Birth of AI

1. The dream of a thinking machine predates the computer: myths of living statues, Frankenstein's creature, and the word robot explored human hopes and fears about creating minds we cannot control.
2. Philosophy asked whether a machine could actually think and how we would know, a question running to the modern Turing test. These old hopes and fears still shape reactions to AI today (saviour or monster headlines).
3. Real AI research grew from the mid-twentieth century. In the 1940s, cybernetics studied how animals and machines use feedback to control behaviour, treating thinking as information processing, which made building intelligence imaginable.
4. The field was named at the 1956 Dartmouth workshop, coining the term artificial intelligence: the founding claim that any feature of intelligence could be described precisely enough for a machine to simulate.
   1. Early programs like the Logic Theorist and the 1960s chatbot ELIZA convinced some users they understood, and researchers predicted human-level machines within a generation.
   2. Reality was far narrower, opening the first gap between promise and delivery. Founders, funders, and press all had reasons to overstate progress (change concept: is a claim a real leap or another wave of enthusiasm?).

Booms and AI Winters

1. AI has advanced in fits and starts. It moves in booms (bold promises attract funding), then AI winters (promises fall short, money collapses).
2. Winters hit in the 1970s and late 1980s. Each followed a stretch where reality fell short of inflated expectations.
   1. Concrete case: 1980s expert systems encoded specialists' rules by hand but were costly and brittle outside narrow cases, and the market collapsed into the second AI winter.
3. Lesson: because AI moves in cycles, a dramatic breakthrough claim is not always a lasting transformation. Watch the incentives: companies and the press both gain from hype, so claims arrive pre-inflated.

The Deep Learning Revolution

1. From around 2012, deep learning (neural networks with many layers) dramatically outperformed older methods, first in image recognition then across speech, language, and games. This era put AI into everyday life.
2. Three things combined to make it possible: enormous digital data, far cheaper computing power (Moore's Law), and better network designs. No single factor was enough alone.
   1. Turning point: in 2012 a deep network called AlexNet won a major image-recognition contest, roughly halving the error rate overnight; the approach soon spread to speech, translation, and games.
3. Contested: whether this is just another boom or a genuine turning point is still debated. The concentration of data and compute places outsized power with the few organisations that can afford it, and the rest depend on their models.

The Singularity Debate

1. The singularity is a proposed moment when AI improves itself so fast that progress races beyond human understanding or control. It is a striking but entirely speculative idea.
   1. Believers argue accelerating progress could produce a super AI reshaping civilisation, and that we should prepare now.
   2. Sceptics counter that today's systems are still narrow, that predictions of imminent superintelligence have a poor track record, and that the fixation distracts from present harms.
2. It is a power question about attention: framing AI's central issue as a distant singularity can serve those building today's systems, pulling focus away from present harms like bias and job loss.

Reading the Evolution: Change and Power

1. AI's evolution has come in surges, disappointments, and revivals, driven as much by hype, funding, and data as by invention itself.
2. The change concept gives a sharp tool: is an advance an evolution (a faster version of before) or a transformation (a genuine break)? Whoever controls the data and compute holds real power over its direction.