Datacrash: A dyslexic dropout’s look into the future of biased AI
In October of 2019, the European Forum for Geography and Statistics (EFGS) invited Alexis Hannah Smith to be a keynote speaker at their conference in Manchester, alongside Robert Cuffe, Head of Statistics for BBC. The event was hosted by the Office for National Statistics (ONS), and gave Alexis the opportunity to talk about the importance of “diversity of thought” in AI, our technology future in general, and the risk that we’re heading towards a “datacrash.”
In her keynote address, entitled “A dyslexic dropout’s look into the future of biased AI,” Alexis started with the current state of data:
As the world’s demand for data grows, we are all using AI to digest, distil and disseminate data efficiently; whether that’s processing and visualising earth observed data or turning petabytes of vehicle telematics into useful insights. However, traditional AI still requires humans to be involved in the creation of the training data or the rules to allow these tools to learn by themselves.
As CEO of a female-founded business and with a team that has been created intentionally to include truly diverse thinkers of all types, from different religions, genders, sexualities, education levels, ages, and ethnic backgrounds. Alexis brings a unique perspective to the implementation of AI and automated data pipelines, recognising and addressing their potential for good as well as their potential for weaknesses in the future.
Yet, as a society, we are walking into a future in which important, life defining decisions are based on AI-derived products. If we then imagine that these derived products are in turn based on other AI derived products, which are in turn based on hundreds or thousands of other AI derived products, this could result in a situation where these decisions are based on data that has inherent dormant flaws. The compound of which could result in erroneous decisions being outputted without any discernible pattern and without any prior warning. In essence, a datacrash.