Reinforcement-learning lets algorithms reshape themselves based on how they influence humans.

Evidence

Using neural networks to process gargantuan amounts of data, algorithms featuring reinforcement-learning could do things beyond Don Draper’s imagination. By surveying the reaction of millions of people to their prompts billions of times every hour, they could train themselves at lightning speeds not only to influence us but, also, to pull off the fascinating new trick that Alexa and her ilk, as we saw earlier, are now capable of: to be influenced by the way they influence us; to affect themselves in light of the way they affect humans. — Yanis Varoufakis, Technofeudalism, 312

Categorization

• XREF: This mechanism — algorithms that adapt in response to their influence on humans — underpins Varoufakis’s claim elsewhere that cloud capital ‘learns’ and thereby extracts rent on human behavior; connects to platform-feudalism core thesis and to earlier Alexa discussion. — connects to: Reinforcement-learning lets algorithms improve themselves faster than humans can.; Reinforcement-learning lets algorithms reprogram themselves using internal performance and reward subroutines. — Yanis Varoufakis, Technofeudalism: What Killed Capitalism, 311; AI interactions can influence human emotional responses and behaviors. — Joanna Stern, I Am Not a Robot- My Year Using AI to…, 583 (+3 more)

Broader Connections

• Reinforcement-learning lets algorithms improve themselves faster than humans can. #expands (0.85)
• Reinforcement-learning lets algorithms reprogram themselves using internal performance and reward subroutines. #expands (0.85) — Yanis Varoufakis, Technofeudalism, 311
• AI interactions can influence human emotional responses and behaviors. #expands (0.80) — Joanna Stern, I Am Not a Robot- My Year Using AI to…, 583
• Algorithmic exposure shapes the very choices that appear voluntary #expands (0.80) — Sander van der Linden, Foolproof, 422
• Machine learning divides into unsupervised, supervised, and reinforcement learning #expands (0.75) — Brian Christian, The Alignment Problem, 58
• Algorithms reinforce existing tastes rather than challenging or expanding them #qualifies (0.75) — W. David Marx, Blank Space, 709