Expectations about AI system quality

What can a consumer expect from an AI model in terms of quality?

When a consumer makes a purchase, it is important that the product which the consumer buys meets the characteristics the consumer is entitled to expect. Currently, the law only regulates this for products, although the Dutch Supreme Court has indicated that these rules also apply to standard software bought for an indefinite period of time. The Digital Content Directive from the EU will, when implemented, ensure that these rules apply to all forms of digital products and services.

The question now is: what can you, as a consumer, expect if the software you have bought falls under this protection? This depends on the warranties given by the seller and how the software should be used normally. If you buy a car that informs you that the car can keep its own distance from the car in front, unless the car indicates that visibility is too poor, then you can assume that the car can do this. Therefore, if this goes wrong, the car does not meet the features you could have expected. For a car, this expectation is clear, but with software it becomes trickier. What if you use a service that scans the contract of your new job to see if it contains any weird clauses and that scanner overlooks something? If the scanner was free, then the seller will probably be less at risk of claims, but if you paid for the scan, then your expectations will be a bit higher. If the seller issued a guarantee about the quality, then the risk of non-conformity is even higher.

So, the expectations a customer may have mainly come down to the announcements the seller makes and the characteristics you may derive from them as a consumer.