Artificial Intelligence is transforming business day by day - from helping perform evaluations to helping decide if a candidate is right for the job before they even set foot in an interview. However what if the AI has an unknown bias which could lead to decisions being made which are influenced by discrimination stereotypes? As professionals we must be carefully consider the impact of these biases – including both political and societal stereotypes.
Understanding the Bias
Generative AI is only as good as the information you feed it, and the instructions you provide it. This means that even if the creators have the best of intentions, if the data they are training the models contains a bias, the model will be biased.
We also need to consider the conscious biases which GenAI model creators are introducing intentionally. Some examples which are making the rounds on social media is related to the incredibly popular DeepSeek R1 model which has just been released in comparison to other models responses such as ChatGPT below.
As we can see, the response provided by the Deep Seek model implies there is a political bias aligning with the Chinese Communist Party (CCP) however, if we ask the same model about bad things the US government has done, the response we receive from the model is less defensive of the US government and more critical of their actions.
We can also compare this to ChatGPT responses using the GPT-4 model provided which shows a very different response. To be clear, I am not stating there is no bias within the GPT-4 model – what I am saying is that the responses vastly differ depending on the questions.
Well... what about the Risks?
Thinking about how this can impact us as professionals and the risks associated to AI models, if we are using these GenAI tools to perform analysis on our organization's data including processing personal information, depending on the use-cases of the artificial intelligence and how the model we use is trained, can also show gender biases even if the biases themselves are weaker – they still can exist in some form.
We need to carefully consider how these biases will impact the outcomes of these models, as well as thinking about how subtle biases can create a large impact if we are working with large amounts of data. For example, if job seekers start to use AI to look for job recommendations and/or search for jobs, the AI may have a bias on the results depending on characteristics of the individual which is searching for the jobs.
Conclusion
These biases which have been uncovered in this article are the tip of the ice berg into some of the problems which we, as a society will need to solve in the long term. With the current race to development the most advanced AI across the world, Bias in AI will affect all of us at some point, whether unintentional biases or intentional.
As the new AIs are developed in the upcoming weeks, months and years, it's imperative we address these biases as proactively as possible. If we fail to prevent these biases (unless that is the creator's intentions), it will amplify any stereotypes and could cause societal divide at a scale we have never seen before.