Opinion formation on social networks with algorithmic bias: Dynamics and bias imbalance

Algorithmic bias can override natural opinion diversity: The research demonstrates that content filtering strong enough can determine final global opinions regardless of initial conditions, meaning platform algorithms may have more influence than individual preferences or genuine disagreement.

Bias strength matters critically: The study identifies a threshold effect—when algorithmic bias favoring one opinion is "strong enough," it fundamentally changes population outcomes, suggesting that even moderate filtering could have outsized effects on societal consensus.

Multiple problematic outcomes are possible: Rather than just polarization, the model shows algorithms can produce coexistence (fragmented groups), consensus (artificial agreement), or polarization—meaning the harms of biased filtering extend beyond the commonly discussed polarization concern.

Convergence to opinions can be unpredictably slow: The research finds that populations sometimes take vastly different amounts of time to settle on final opinions, with occasional "critical slowing down" effects, suggesting that algorithmic bias effects may be hidden or delayed before becoming apparent.

Network structure amplifies algorithmic effects: The interaction between community structure in social networks and algorithmic filtering creates complex dynamics, indicating that platforms' filtering algorithms don't operate in isolation but interact with how people naturally cluster into groups online.