By Miro Cafolla, Postgraduate and International Tutor.
Artificial Intelligence (AI) plays a fundamental role in our daily life. AI now finds applications in a wide number of areas, from healthcare to the military. On one hand, this technology has been able to reduce human efforts and provide us with more accurate and faster results. However, on the other hand, AI comes with a number of problems and questions we need to address in order to make sure that the technology is used for the good.
On Monday 18th November 2019, Professor Alexandra I. Cristea provided some potential solutions to these problems, during a stimulating talk for our very own Café Scientifique held in the LRC lecture room.
A great audience including members of all College Rooms attended – JCR, MCR, SCR, CCR and even visiting fellows included!
Professor Cristea is one of the top scientists in the field of Artificial Intelligence. She is currently the Head of the Innovative Computing research group and Deputy Head in the Computer Science Department of Durham University. Professor Cristea has also published more than 250 articles on the topics of web science, learning analytics, user modelling and personalisation, semantic web, social web, and authoring. Additionally, she has won a number of prizes and awards for her scientific contributions, and has been included in the top 50 researchers in the world in the area of educational computer-based research according to Microsoft Research.
Professor Cristea explained that recent AI-based technology, namely Machine Learning (ML), uses large data sets in order to predict a possible outcome or mimic the human behaviour. Unfortunately, AI algorithms may come with a number of social issues.
As an example, she discussed chatbots. Chatbots are a form of AI which conducts a conversation via auditory or textual methods. The Microsoft twitter chatbot, TayA, mimicked the language patterns of a 19-year-old American girl and produced racist outputs. This led to its shutdown 16 hours after launch. AI-based algorithms are also used in courts (e.g. in the U.S.) in order to predict the likeliness of criminals to reoffend. Such algorithms have been found to misclassify black people, especially men, as more likely to commit a crime again in comparison to white people.
Examples of social bias include also algorithms used in the financial sector and in advertising. White young men are more likely to be classified as reliable customers and potential applicants for highly qualified jobs. However, women and individuals from minority groups tend to be given lower credit limits, and are advertised jobs for low-skilled positions more regularly on their social media pages.
Social bias is not a new topic when considering algorithms that process an extremely large number of data sets. This is due to having skewed input data. However, this problem is becoming more problematic and widespread today. Algorithms use data available on the Internet, where a small percentage of the world population is responsible for the vast majority of the image and verbal content. 45% of all the image data comes from the USA (4% world population), whereas China and India contribute about 3% of image data but represent 36% of the world’s population!
Is then AI social bias inevitable today?
Professor Cristea showed a number of ways researchers are trying to and succeeding in mitigating these effects. Bias mitigation may take the form of accurately selecting balanced data, and assigning a larger weight to underrepresented categories. Algorithms should also use different users when training. This will help AI to develop an approach unbiased with respect to the type of user.
Professor Cristea concluded her fascinating talk suggesting that, as IBM says, ‘AI bias will explode. But only the unbiased AI will survive.’. Professor Cristea added that the AI revolution may not be a linear process, but it may be as hard and as difficult as the industrial revolution in the 18th century, with a number of people having to abandon their classic jobs and re-invent themselves. Professor Cristea is however optimistic that, in the long term, technology will further improve our lives.
The talk was a stimulating opportunity to help participants “see” AI for what it actually is – a great tool which is up to us to use for good.
If you want to find more on AI, have a look at the videos Professor Cristea showed us!: