Artificial Intelligence Can We Control It
Artificial Intelligence: Can We Control It or Will It Take Over?
AI does not need to become conscious to change who holds power. This documentary separates today’s demonstrated harms from emerging technical risks and uncertain loss-of-control scenarios, then examines alignment, hallucination, bias, privacy, security, agents, misuse, manipulation, critical infrastructure, military systems, interpretability, evaluations and practical governance.
Forwards or Backwards — PAST ← | → FUTURE https://www.youtube.com/@forwardorbackwards
In these topics
Chapters
- Control begins before a machine becomes powerful
- What exactly are we trying to control?
- Why behaviour can surprise its creators
- Specification and reward miss the human meaning
- Helpful to whom, and under which values?
- A confident answer is not knowledge
- Data, labels and deployment shape unequal outcomes
- Useful context can become permanent exposure
- Attackers talk to the model too
- A test score is not the real world
- Looking inside a machine made of numbers
- Can a model behave differently when watched?
- Time, tools and persistence multiply capability
- Human intent can weaponise useful capability
- Control can be lost through attention
- A system can take over because people stop practising
- Who controls the systems that control decisions?
- Economic control is part of human control
- Errors become physical when software touches the world
- Speed can erase the space for judgment
- Capability can accelerate the next capability
- A possibility is not a prediction
- Training, constitutions, monitors and tripwires
- Rules must reach the organisations deploying AI
- What responsible deployment looks like
- Control is a capability societies must keep rebuilding