bias

Neural networks

A learnable parameterparameter, n.: One of the adjustable numbers inside an AI model that are tuned during training. Large language models have billions of them, and together they store what the model has learned. added to a unit's weighted sum of inputs before the activation functionactivation function, n.: The simple rule each artificial neuron applies to its total input to decide its output, such as 'keep positive values, turn negatives into zero'. Without it, a deep network would be no smarter than a single layer., so that each artificial neuronartificial neuron, n.: The basic building block of a neural network: a tiny calculator that multiplies each incoming number by a weight, adds them up, and passes the total through a simple function to decide its output. computes f(w·x + b). The bias shifts the unit's threshold, letting it activate even when all inputs are zero; in linear models the same term is called the intercept. Biases are learned along with the weightsweights, n.: The learned numbers inside a neural network that set how strongly each part influences the next. The finished set of weights is, in effect, the trained model itself. but are usually exempt from weight decayweight decay, n.: A training rule that shrinks every weight slightly at each step, like a small tax on large numbers. It keeps models simpler and usually helps them work better on new data., and many recent transformer designs omit them from most layers.

Statistics

For an estimator, the gap between its average value over repeated samples and the true value of the quantity it targets; an unbiased estimator is right on average. In the analysis of learning algorithms, bias is the part of prediction error that comes from a model class too rigid to capture the true pattern, traded off against variance, the part that comes from how much the fitted model would change with a different training sample (bias–variance tradeoffbias–variance tradeoff, n.: The balancing act between models too simple to capture a pattern and models so flexible that they also learn the noise. The best predictions usually come from somewhere in between.). Related notions include selection or sampling biassampling bias, n.: A distortion that arises when the data collected over-represents some kinds of people or situations and under-represents others, so conclusions drawn from it don't hold for everyone., when data are unrepresentative, and measurement bias.

Inductive bias

The assumptions that lead a learning algorithm to prefer some hypotheses over others when the data alone underdetermine the answer, such as smoothness, simplicity or the locality built into convolutional networksconvolutional neural network, n.: A kind of neural network built for images and other grid-like data. It slides small pattern detectors across the input, finding edges first, then shapes, then whole objects.. Without some inductive biasinductive bias, n.: The assumptions built into a learning method that decide how it fills the gaps between the examples it was shown, such as that similar inputs get similar answers. Every learner needs some to cope with new cases., generalizing beyond the training examples is impossible.

Fairness and society

Systematic, unfair differences in how an AI system treats or represents people, especially by race, gender, age, disability or other protected attributesprotected attribute, n.: A personal trait, such as race, sex, religion, age or disability, that the law says must not be used to treat people unfairly.: a résumé screener downgrading women, face analysis erring more on darker-skinned faces, a language model reproducing stereotypes. Sources include unrepresentative or historically skewed data, choices of labels and measurements, objectives and deployment context; see algorithmic biasalgorithmic bias, n.: Unfairness in software that judges or serves people: it consistently produces worse results for some groups, such as women, older people or ethnic minorities, often because it learned from skewed or incomplete records., algorithmic fairnessalgorithmic fairness, n.: The effort to make automated decisions fair: to say in numbers what fair treatment means, measure whether a system achieves it, and correct the system if it does not. and Gender ShadesGender Shades, prop. n.: A 2018 study showing that commercial face-analysis software guessed the gender of darker-skinned women far less accurately than that of lighter-skinned men..

The audit found bias in the loan model: equally qualified applicants from two postcodes got different scores.

Judgment and model behavior

Systematic tendencies in human or machine judgment, such as automation biasautomation bias, n.: The habit of trusting a computer's suggestion too much: following it even when it is wrong, or missing problems because the machine did not flag them., the human tendency to over-trust automated outputs, or the position, verbosity and self-preference biases of LLM judgesLLM-as-a-judge, n.: Using one AI language model to grade or compare the answers of another, as a cheaper and faster stand-in for human reviewers.. Unrelated is logit biaslogit bias, n.: A setting that lets developers make specific words more or less likely in an AI model's output — even ban a word outright — by adjusting the model's scores before it chooses., an API setting that raises or lowers the probability of particular tokens.