Brainpool is back with another ‘Top Takeaways’ post on the latest AI Summit 2020 Episode! This week, we’re highlighting some stand-out sessions on equality in data and the importance of data governance.
Equality in Data:
Preventing biases within datasets has been one of the prominent and overarching themes of this year's summit. Previously, we have heard speakers such as Rediet Abebe, Junior Fellow at the Harvard Society of Fellows, suggest that businesses have a responsibility to embrace technologies that help dismantle discrimination. In a similar vein, two sessions stood out to our team this week - both of which explored the practical means of ensuring fair Artificial Intelligence. Dr Evert Haasdijk, Senior Manager at Deloitte, offered examples of how technology should be implemented in his talk: The Age of With – Human’s working with AI. Alongside this, the fantastic roundtable discussion on Making diversity & inclusion a business imperative, featuring Nazareen Ebrahim, Renee Cummings, Dr. Natalie R and Lavina Ramkissoon, saw clear reasons and means of utilizing unbiased data.
The panel outlined two major points that all organisations should consider when attempting to leverage data. Firstly, as is becoming increasingly apparent in today's climate, companies that are more diverse will benefit from a range of perspectives on their data. Varied approaches accordingly create further economic incentives to embrace a truly diverse workforce. Secondly, Renee Cummings of the University of Virginia highlighted the importance of being aware of inherent historical data biases. Taking the time to consider the ways in which data is collected and the potentially problematic factors in its creation, such as inherent discrimination, is essential for gaining a full and fair portrait of your company's environment.
Dr Evert Hassdijk’s talk earlier in the episode effectively outlined guidelines for creating fair and unbiased AI, centred on humans and machines working together, Dr Hassdijk stressed the need to prepare for inevitable mistakes in AI systems. The fallibility of the technology may be down to a variety of reasons, from accidental data entry to inherent bias - however, building machines which overcome this prevents the knock on effect of mistakes. He suggested that Artificial Intelligence should be built in a way that allows human intervention and full visibility over data. Not only does this provide the ability to correct mistakes that might have been made, but illustrates why conclusions have been decided upon. Running through an example of anomaly analysis, the session clarified why certain pieces of data had been presented as anomalies. This is a fantastic example of how to practically implement AI whilst considering the panels recommendations for unbiased datasets.
The importance of Data Governance
Wendell Wallach (Senior Advisor, The Hastings Centre & Interdisciplinary Center for Bioethics, Yale University), Anja Kaspersen (Former Head of Geopolitics and International Security, World Economic Forum, Wendy Hall (Regius Professor of Computer Science, Director, Web Science Institute, University of Southampton) and Ana Alecsandru (International Security Programme, Chatham House) delivered an excellent session on the complexity of data governance systems in today’s society and its increasing importance.
The speakers highlighted the chasm between the speed of digital development and innovation (accelerated by COVID-19) and existing forms of data governance and oversight. Given the constant change brought about in the field, governance models cannot keep up. As such, it is of vital importance that steps are taken to establish 21st century models that regulate new technologies. There needs to be agreement at not just a national level, but at an international one.
Global industry leaders must ascertain two things in particular. Firstly, commonality between models should be established in order to clearly identify where the disparity between governance systems lies. Secondly, leaders should determine how they wish to deal with these disparities in order to move forward in a viable way.
Wendy Hall also highlighted the necessity of thinking practically rather than theoretically in this regard. Fortunately, technical experts today have safety questions in mind and recognise the importance of safe and ethical deployment. Consequently, we’re on the right path to ensuring that the relevant conversations take place in regards to the fitting together of existing frameworks and models.
As part of our company ethos, Brainpool AI is committed to the development of fair and unbiased AI.
When considering innovation presented by AI opportunities, the impact on humans cannot take a backseat to efficiency, cost reduction and profits. Companies need a formal way of assessing this impact as part of their business case.
Next time, we’re really looking forward to Episode 5 - Applied AI’s sessions on the likes of ‘Bringing new insights to asset managers through visual data and AI’ by Camiel R. Verschoor (CEO, Birds.ai), ‘Explainable AI for fair and inclusive outcomes in financial services’ by Janet Adams (Head of Strategic Projects & Performance - TSB) and ‘Tech for social good: from research to practice’ by Ana-Andreea Stocia (Co-organiser - MS4SG).
Written by Anjali Kapila & Dominic Richmond