AI-900 Practice Question: Describe Artificial Intelligence workloads and considerations

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✗

  When an AI model is licensed for use by two different organisations simultaneously

  Why it's wrong here

  A scenario where two organisations separately license the same AI model is strictly a commercial and intellectual property matter, involving agreements, pricing, and contractual terms. It does not introduce a new ethical dimension about the capabilities themselves; whether one or one thousand organisations use the model, the underlying risk of misuse remains identical. Dual-use is not about the number of licensees or the breadth of deployment, but about the intrinsic potential of the model's functionality to be applied in both constructive and destructive contexts.

• ✓

  The risk that AI capabilities designed for good can also be used for harmful purposes

  Why this is correct

  This is the correct definition: dual-use in the context of responsible AI refers to the reality that a single AI capability, developed for legitimate purposes, can also be weaponised or exploited for harm. Classic examples include generation of realistic images for art versus using the same technique to create non-consensual deepfakes, or language models that assist with coding but are also used to craft phishing or malware. Such risk mandates proactive safeguards like usage restrictions, input/output filtering, content watermarking, and threat monitoring to balance innovation against safety.

• ✗

  Deploying the same AI model for both training and inference to reduce costs

  Why it's wrong here

  Deploying the same model for both training and inference is not technically standard, since training requires iteratively updating weights while inference requires a fixed, stable model; in practice, one may reuse a pre-trained model for a different task via transfer learning or fine-tuning, which is a cost-saving efficiency technique. This relates to resource optimisation and model reuse, not to the potential for a beneficial capability to be turned to malicious ends. Dual-use is about the inherent misuse potential of a model's function, not about reusing the model across operational phases.

• ✗

  Combining two AI models to achieve better results than either model alone

  Why it's wrong here

  Combining two AI models, such as via bagging, boosting, or stacking, is a well-known model ensembling technique used to enhance predictive performance by reducing variance or bias (e.g., random forests, gradient boosting). This is a purely machine-learning-centric optimization strategy, not a characteristic of how the technology might be repurposed for harm. The notion of dual-use specifically refers to the same fundamental AI capability being simultaneously beneficial and exploitable, whereas ensembling addresses accuracy, not ethical risk.

Go deeper

Related to this question