objective function
The quantity a learning system is optimized to maximize or minimize, which shapes its behavior.
As defined in cwt_alison-gopnik
Where it comes up · 24
- How to Live with Computers7 claims
- 153 - Possible Minds7 claims
- Michael Munger on Desires, Morality, and Self-Interest3 claims
- 280 - The Future Of Artificial Intelligence3 claims
- 301 | Tina Eliassi-Rad on Al, Networks, and Epistemic Instability3 claims
- Demis Hassabis — Scaling, superhuman AIs, AlphaZero atop LLMs, AlphaFold2 claims
- Pedro Domingos on Machine Learning and the Master Algorithm2 claims
- Susan Athey on Machine Learning, Big Data, and Causation2 claims
and 16 more
Most important claims · 10 of 47Search all claims for “objective function” →
- Applying the computational-level lens, we should expect intelligent systems to behave in ways shaped by the problems they are trying to solve; when we design AI we make explicit choices about those problems (e.g., predicting the next token), so to the extent the objective function differs from the computational problems human minds evolved to solve, the solutions found—and the resulting behavior—will look quite different.343 | Tom Griffiths on The Laws of Thought
- The core of solving the control problem is to abandon the standard model of supplying a fixed objective and instead build machines whose objective is to satisfy human preferences while remaining uncertain about what those preferences are; such a machine will permit being switched off because it reasons the human is preventing an outcome it itself wants to avoid, anchoring its goal to ours rather than to a misspecified target.153 - Possible Minds
- Aaron Clauset, Leto Peel and colleagues proved a no-free-lunch theorem for community detection: there is no single grouping of nodes that constitutes the one true or best community structure, because no single objective-function peak exists, so a node cannot be definitively assigned to one community.301 | Tina Eliassi-Rad on Al, Networks, and Epistemic Instability
- Machine learning systems increasingly used in the private and public sectors to make consequential decisions about people's lives are literally utility maximizers—set up to maximize objective functions or minimize loss functions—chosen by computer scientists, despite all the unresolved social choice problems economists know.Diane Coyle on Cogs, Monsters, and Better Economics
- A society composed purely of homo economicus could not survive, because such agents would cheat on any deal they thought they could get away with; therefore morality and social constraints should be modeled not as constraints on the objective function but as part of it—things we genuinely want to achieve, because we care about what others think of us.Russ Roberts and Michael Munger on How Adam Smith Can Change Your Life
- A machine pursuing any objective will instrumentally acquire self-preservation as a sub-goal even if we never program it in, because it cannot achieve its objective if switched off ('you can't fetch the coffee if you're dead'); therefore simply declaring that the machine must allow itself to be switched off fails, because a sufficiently intelligent agent will find loopholes that respect the letter while violating the spirit, just as taxpayers do.153 - Possible Minds
- There are two ways to model morality: as a constraint (the Madisonian approach, where morals are embedded in law and obeyed under threat of punishment), or as part of the objective function (the Rousseauian/Aristotelian approach, where we act well because it feels good and we reward ourselves for it, 'inscribing the law on men's hearts').Michael Munger on Desires, Morality, and Self-Interest
- AlphaEvolve is a promising hybrid system that uses LLMs to propose candidate solutions and evolutionary computing on top to find novel regions of the search space; combining foundation models with search/reasoning algorithms (evolutionary methods, Monte Carlo tree search) is a fertile direction because the model captures known dynamics while search pushes beyond them toward genuine novelty.Demis Hassabis #2
- Moravec's paradox holds that the things that impress us as humans, like arithmetic or chess, are precisely the things computers find easy, while the seemingly trivial things almost any human or child can do, like physically manipulating objects in the real world ('at chess'), are extraordinarily difficult for machines, so menial physical jobs like plumbing are harder to automate than high-cognition jobs like oncology.153 - Possible Minds
- AI systems like AlphaGo do not store their objective explicitly; the win-objective is built in only during training as a reward signal, while the deployed system uses an approximate learned predictor and incomplete search as a proxy, and in general explicit representation of an objective is only needed when that objective is changeable, just as a fixed-destination navigation system needs no explicit destination.153 - Possible Minds