Jonathan Prunty

AI systems increasingly incorporate continual learning mechanisms allowing their behaviour to adapt after deployment, from (1) in-context learning and (2) memory features already in wide use to (3) post-deployment weight modification under research. We argue that, by treating AI systems as frozen artefacts whose performance and safety are assessed at release, current evaluation practices structurally ignore the behavioural trajectory of a system that continues to learn from experience. Our position is that evaluation of continual learning systems should be centred on behavioural trajectories, with the complementary goals of characterising the landscape of possible behaviours and forecasting how behaviour will evolve from a given set of experiences. This can be operationalised through trajectory elicitation sandboxes and predictive monitors that forecast behavioural evolution, but may face fundamental obstacles analogous to those seen in dynamical systems. These are best addressed by (1) applying trajectory-centred evaluation to today's continual learning systems and (2) relying on the resulting evidence to design systems amenable to it, yielding a virtuous cycle in which systems and their evaluations co-evolve.

As AI is integrated into the workplace, organisations increasingly face allocation decisions between human and machine workers. These decisions are increasingly made or assisted by algorithms, creating a Reverse Turing Test dynamic wherein the machine is now the judge. In addition, human and machine workers may ``compete'' for a given task, reproducing aspects of adversarial games. This raises new methodological questions about assessing task suitability between humans and machines. The criteria often used to assess people (e.g., education, experience, references) cannot feasibly scale to AI systems; conversely, AI evaluation methods (benchmarks, red teaming, leaderboards) cannot be easily applied to human workers or yield comparable metrics. In this position paper, we argue that suitability evaluations for task-assignment should be profile-driven -- that is, based on assessments that infer latent constructs such as capabilities and propensities from observed performance. This approach places humans and AI systems on shared scales, supporting comparisons that are predictive of novel-task performance, explanatory of why agents succeed or fail, and auditable. We outline the core features of this approach, discuss its practical implications, and compare it with alternative frameworks for human-machine workplace allocation.