AI for Social Good
I recently attended a wonderful symposium exploring the recent advances in AI and how we can achieve positive outcomes from society.
As AI applications are now becoming a part of everyday life from Alexa’s news skills, Netflix’s movie recommendations to Amazons product recommendations.
These applications now raise new challenges around ethics and robust design. This talk bought about some interesting challenges of how we train AI systems to make decisions without the biases we as humans carry.
The talks were held at the Royal society, I can say the people in this room, are driving more value than some of our senior Politian’s negotiating Brexit. The value and knowledge openly shared by these scholars, will directly impact the prosperity of our nation. Our senior AI guru’s, again are leading the discussion on the ethics of how this 4th industrial revolution will be shaped, with equality, fairness and ethics by design, being at centre of our research and thinking.
Max Welling (Research Chair in Machine Learning, University of Amsterdam, Netherlands) kicked off with some very interesting points on AI per kilowatt hour. AI is great, but we can’t ignore the huge demand on TPU’s, and the energy consumption required to power our autonomous cars and so forth. Data centres currently utilise 4% of our global energy consumption, so AI is far from free. We learned to produce a Facebook timeline, cost 30 Micro Dollars, the question remains, should we shorten our timelines and save some trees?
Moving AI to the cloud and to Edge devices cannot sensibly sustain this demand and performance without more efficient architectures. Max raises an important point, the use of low bit computation, compression and clustering algorithms is a must, if these solutions are to become truly portable. Again developers and scientist are now tasked with developing efficient coding and architectures, as a key performance metric if these algorithms are to live on the net.
Matt Kusner (Research Fellow, The Alan Turing Institute) presented an interesting discussion on Counterfactual Fairness, how do we code our algorithms to produce fair and accurate results which do not carry the current biases and prejudices we as humans all suffer from.
Given the previous disasters with Microsoft Tay (Bot) we can see how, without this kind of thinking, we can easily get into some serious litigation. Are we then take up issues with the bot or the data scientist challenges still at bay.
As the Data Scientist domain grows, there is ever more responsibility and immense power and influence that sits at their fingertips. As new algorithms are developed to crunch numbers and make predictions, strict responsibility and thoughts must be extended, to ensure policy making is upheld and respected so that fair outcomes are reached.
Closing notes, Singapore’s investment and budget in AI, is making the UK budget look like once again we are playing with crumbs. This is an area which we can clearly dominate, we have grown the best intellectuals and we live in service sector economy. Toyota sponsored this event, we need more Toyota’s to stand up funding to put UK AI on the map, before we lose our scientist to better funded institutions.
Many thanks to the Turing Institute for a very informative day, looking forward to attending more.