The latest webinar hosted by the University of Johannesburg lacked neither quality nor quantity on 27 May.
The third webinar on Data and Delusion after Covid-19 brought three guest speakers of different backgrounds to the fore to analyse what went wrong with data predictions and how data can be used going forward, with regard to the Coronavirus and future pandemics.
Dr Shakir Mohamed, an expert in machine learning regarding global challenges, and senior researcher at DeepMind in London, said the global pandemic involved a complex social problem which could not be solved by machine learning alone.
One of the reasons it did not play much of a role in predicting and addressing the pandemic was because society still has a premature approach to machine learning.
Rather, society should be cautious with data taken from machine learning and use it in conjunction with data taken from other methods.UJ’s Prof Charis Harley, who has worked as a data scientist, said another problem at the moment is that scientific data about Covid-19 is held by a number of organisations and research centres around the world and they are not sharing this data effectively.
As a result, the public are listening to other voices in the public dialogue. This is because a good scientist, she said, does research and is quiet at first, only speaking when they have results from data gathered. Thus the first voices spoken in the public are often not scientific or if they are, they are not wholly accurate.
Prof Olaf Dammann, a professor at universities in America, Germany and Norway and an epidemiologist working in the United States, agreed that a good scientist should do research before adding their voice to the public debate.
He added that the situation is tricky because with global pandemics the models researchers can build on the data should only be applied to specific places, or countries.
Thus a working model constructed from data from the catching of the virus in say, Europe, cannot be applied effectively to stop the spread of the virus in South Africa.
For this reason models should be applied locally and they should also be as simple as possible.