The dangers of bias in Artificial Intelligence systems

Attention!

To participate in the forum of this paper and to receive the notifications, it is compulsory to activate it beforehand for your user. To do so, click on the button "subscribe". a little further down. You will be able to choose whether you want to receive all notifications in your email or only the replies to your comments.

Enjoy the Congress!

0 0 votes

Rating

2 Comments

Más antiguo

Más nuevo Más votado

Karen Hernández Rueda

1 year Past

Thank you for the presentation. It is a topic of current interest. In particular, I think the use of Generative AI involves first learning how to give it instructions and then training it to do what we want it to do, which takes time. It is not clear to me what the biases of the study are. I would like to know which ones were analysed and what qualitative aspects were considered in the analysis with the use of chatbots, facial recognition, among others. Thank you

Author

María Sáez de Propios

1 year Past

Reply to Karen Hernández Rueda

Thanks for your comments, Karen! You are absolutely right, the use of the Generative (as models of generative artificial intelligence) involves first learning how to give it precise instructions and then training it to perform the desired tasks. This process certainly requires time and proper handling of data and algorithms.
As for the study biasesIt is essential that these are carefully analysed, especially when it comes to technologies such as chatbots, facial recognition and other AI applications. In general, biases can arise from several points, such as:

Data BiasesIf the data on which the models are trained are biased (for example, if there is a disproportionate representation of certain demographic groups), the AI will replicate those biases in its results. For example, in the case of chatbots, if they are trained with conversations mostly in English or with certain cultural terms, they may have difficulty understanding or generating appropriate responses in other cultural contexts.
Algorithmic biasesAlgorithms may be designed in such a way as to favour certain outcomes. This can occur if careful adjustments are not made in the model design to avoid favouring certain predetermined behaviours or responses.
Biases in test designIf the methods used to evaluate the performance of systems (such as the facial recognition or the chatbots) do not include a wide range of scenarios or populations, the results may be biased. For example, face recognition has been shown to be biased in identifying race or gender if not trained on sufficiently diverse data.

Best regards,

The dangers of bias in Artificial Intelligence systems

The force behind the conference

ORGANISED BY

COLLABORATE

CONTACT

informacion@congresodigitalizacionynegocio.es