(The title was pilferedinspired from a comment by a work colleague, who agreed to be henceforth referred to as [Your Name], as ChatGPT offered this placeholder for their signature.)
Artificial Intelligence or, more accurately, Machine Learning is an amazing tool for sifting through large amounts of data and discovering insightful patterns. A task where a human operator would generally get bored and become sloppy — or simply die of old age in the process — can be very effectively performed by a machine, and a result returned, sometimes in a matter of seconds.
Rather than exhibit true intelligence, however, those systems only learn as much as is present in the data they are given. This is also what they regurgitate. It is no wonder that outputs from those algorithms replicate the biases present in their input data.
Much research work has gone into identifying and reducing biases in training data, or actively de-biasing responses, but the final decision of what to do with the result of an ML process is entirely in the hands of a human being operating it.
tl;dr: Rather than focusing solely on painstakingly fixing each ML system separately, we should also leverage generative AI chatbots to help train humans to recognise, and critically think, when dealing with any ML system, de-biased or not.
Continue reading