{"id":51460,"date":"2018-11-15T19:14:00","date_gmt":"2022-10-18T19:14:23","guid":{"rendered":"https:\/\/www.salesforce.com\/?p=51460"},"modified":"2022-11-01T20:12:15","modified_gmt":"2022-11-02T03:12:15","slug":"ai-are-we-creating-a-superhero-or-supervillain","status":"publish","type":"post","link":"https:\/\/www.salesforce.com\/au\/blog\/ai-are-we-creating-a-superhero-or-supervillain\/","title":{"rendered":"AI: Are we creating a superhero or supervillain?"},"content":{"rendered":"\n

As AI matures and its use becomes widespread, the technology learns what we teach and uses it to do as we instruct, so ethics is a vital consideration in both creation and consumption.<\/p>\n\n\n\n

As an infant, Superman was sent in a rocket from Krypton and crash landed on Earth. He was found and adopted by the kind-hearted Kent family, who taught him strong values. As every parent knows, a large part of the person a child grows up to become results from the values parents role model.<\/p>\n\n\n\n

But this was no ordinary child, he had super powers. The values the Kent family infused into the boy as he grew allowed him to become a superhero rather than a super villain.<\/p>\n\n\n\n

We\u2019ve now seen the birth of artificial intelligence, another powerful force. It\u2019s in its formative years and, as we bring it into our industries, our workplaces and our homes<\/a>, as we shape what it will become, what sets of values are we going to embed into it?<\/p>\n\n\n\n

People are trying to work through the implications of how AI is developed in its infancy, of how and what it is fed in order for it to grow into a responsible adult. I think there are two important lenses we must look through. The first is about creation and the second is consumption.<\/p>\n\n\n\n

Creating responsible AI<\/h2>\n\n\n\n

The discussion around creation centres on the inherent biases we might be hard-wiring into AI as we build it. AI algorithms feed on data, under a form of supervised learning \u2013 our Chief Scientist Richard Socher recently said about training AI, \u201cIf you put rubbish in, you get rubbish back out\u201d.<\/p>\n\n\n\n

Fundamentally, it is a sophisticated pattern recognition framework that is based on the data it is fed. But what if it is not fed the right mix of data?<\/p>\n\n\n\n

I remember a wonderful TED talk by an MIT researcher who is African American. She was working on facial recognition technology which, despite very easily recognising the faces of her Caucasian colleagues, couldn\u2019t recognise her face as a face. When she began to unpick this, she found the training data contained only Caucasian faces. As a result, it errored out when it saw anything that did not match that set. This is a good example of AI having the potential to amplify our own biases.<\/p>\n\n\n\n

It can cause errors, and in more serious instances it can cause deaths. A report in The Economist<\/a> about the death of a pedestrian hit by a self-driving Uber vehicle said the car spotted the pedestrian but, because she was pushing a bicycle, didn\u2019t recognise her as a pedestrian. Many other factors were involved in this terrible accident, but once again it demonstrates the importance of fully representative data in the set from which the AI is learning.<\/p>\n\n\n