Cloud computing and AI are two of the most important and impactful technologies in our era. However, it is not necessarily better to combine these two if we don’t know how to do it right. In fact, some IT experts are warning that cloud availability may worsen AI bias issues.
Research firm Gartner estimated that by 2022, 85 percent of AI projects will produce incorrect outcomes because of bias issues. The biased AI can originate from biased data, algorithms, or its creators and management teams. According to a survey by Gallup, 85 percent of Americans are using at least one AI-enabled device, program or service in their everyday life.
Biased AI and machine learning models have been a major concern in the field for a while. These technologies have wonderful computing ability; nonetheless, what they know is taught by humans. And as we all know, humans are biased and carry a lot of stereotypes. Therefore, it is not difficult to understand why AI can be so biased that it makes erroneous decisions that affect human life tremendously.
When the world is moving to the cloud, AI systems also have to follow the flow. Hosting AI systems in the cloud is cheaper and more efficiently for the owners. More companies can afford AI and they all want a piece of the hot technological advancement. The increase of AI systems may sound great; however, it creates another problem.
The growth of AI adoption is high but the number of qualified AI professionals can’t keep pace with it. As a result, besides the intrinsic bias we already have, we can now expect more mistakes caused by the unqualified AI recruits.
Of course, we can’t blame cloud computing for extravagating the bias problems in AI systems. After all, cloud computing only helps to make AI cheaper and available to a broader user base. The underlying problems are the skills gap, quantity and quality of the workforce, as well as the principles and ethical issues in the development and management of AI systems.
Take Google Images for example. A study in 2015 recorded when the image results for keyword “CEO” included only 11 percent of images were women. The ratio results from the deeply rooted social stereotype that men tend to hold higher positions than women in the corporate setting. There are various examples proving that AI systems learn popular biases from its creators and owners.
If organizations leverage biased AI systems, they will end up with biased decisions. That can hurt the service targets or the organizations themselves. Therefore, it is vital that companies understand the risk of using biased AI and have solutions to avoid falling into this trap. If everyone in the AI ecosystem is conscious of the bias issue in AI, we can soon resolve the problem of biased AI or at least minimize the damage of it.