“The intelligence exhibited by machines or software”, artificial intelligence holds a lot of promise in making machines smarter using tools of natural language processing, reasoning, computational intelligence, robotics etc. The commercial potential includes customer service, self driving cars and personal care.
Microsoft launched an experimental chatbot based on AI last week. On Wednesday Tay was born, an artificially intelligent chatbot with the personality of a 19-year-old female American, with the aim of “conducting research on conversational understanding”.
But it quickly went wrong, within hours Tay’s twitter account was supporting conspiracy theories around 9/11, espousing right wing views to vie with Hitler or Donald Trump. Tay’s life was short, Microsoft took her offline by the evening, and the worst of her tweets started disappearing. (Not before loads of people took screen grabs).
In the same week FastCompany reported on Whisper’s Arbiter software which ensures that nothing untoward is published by the company’s user-base. Whisper is anonymous and combines text with images, a virtual version of PostSecret. Their filter software is build on masses of data, but they still use human moderators to make sure that their user generated content stays on the right side of the company’s policies. This is particular challenging in an environment of anonymous accounts and sneaky attempts to subvert the algorithm.
So why didn’t Microsoft do the same? This “troll” phenomenon is well-known and well documented, and Microsoft has significant experience using social platforms, certainly enough to predict this. WIRED report that Microsoft advised them in an email “We have taken Tay offline and are making adjustments” so perhaps when Tay comes back online she’ll have learnt from the first experiment.
There’s an oft quoted saying “artificial intelligence is no match for natural stupidity”, in this case Tay learnt from us, she copied patterns of speech and opinions from the humans who interacted with her. She’s a human creation in more ways than one; to make artificial intelligence better, we need to be better humans.
Image: digitization via pixabay