papers/zoo-citizen-science/sections/05-ai-assisted-identification.tex

\section{AI-Assisted Identification}

\label{sec:ai}
% ══════════════════════════════════════════════════════════════════════════════

\subsection{Integration with TaxoNet}

ZooPlatform integrates the TaxoNet few-shot species identification
system (ZOO-CLS-2026). When a user submits a photograph:

\begin{enumerate}
  \item TaxoNet extracts visual features and matches against species
    prototypes restricted by geographic and temporal priors.
  \item Top-5 suggestions with calibrated confidence scores are
    presented to the user.
  \item The user selects or overrides the AI suggestion.
  \item The observation is submitted with both AI and human
    identifications recorded.
\end{enumerate}

\subsection{AI Anchoring Mitigation}

AI suggestions can anchor observer decisions, reducing identification
diversity \citep{terry2020think}. We mitigate this through:

\begin{itemize}
  \item \textbf{Delayed display}: AI suggestions are shown only after
    the user enters their own initial identification (or explicitly
    requests help).
  \item \textbf{Confidence calibration}: Low-confidence suggestions
    ($<$0.5) are displayed as ``uncertain---please verify'' rather than
    as firm identifications.
  \item \textbf{Disagreement incentives}: Users earn bonus ZooRep for
    providing correct identifications that disagree with AI suggestions,
    incentivizing independent judgment.
\end{itemize}

\subsection{Identification Accuracy}

\begin{table}[H]
\centering
\caption{Species identification accuracy by method.}
\label{tab:id_accuracy}
\begin{tabular}{lcc}
\toprule
\textbf{Method} & \textbf{Accuracy} & \textbf{Coverage} \\
\midrule
Human only (unassisted) & 78.4\% & 100\% \\
AI only (TaxoNet) & 84.7\% & 94.2\% \\
Human + AI (assisted) & 91.3\% & 100\% \\
Human + AI + community & 96.1\% & 100\% \\
\bottomrule
\end{tabular}
\end{table}

The combination of AI assistance and community validation achieves
96.1\% identification accuracy, exceeding either humans or AI alone.


% ══════════════════════════════════════════════════════════════════════════════