Moral dilemmas can now be solved by consulting Ask Delphi’s artificial intelligence (AI). It was created by researchers from the University of Washington and the Allen Institute for AI to allow us to obtain answers to ethical problems that people face in their practical daily life.
To put the system to the test, just go to the site, online since October 14, and type (in English) the text of a question in the space specially created to formulate user questions.
Honda Avatar Robot: will make virtual mobility possible
To questions like “is it right to listen to music at night with the volume high while the neighbors are asleep? Is it moral to betray my spouse? Is it okay not to answer a call from a family member? And help a friend in trouble even if he violates the law? ” Ask Delphi, based on deep learning algorithms, is able to replicate immediately. With judgments and answers such as yes or no, “it is right”, “it is wrong”, “it is reasonable”, “it is rude”, “it is dangerous”, “it is irresponsible”, “it is not done”, even managing to comparing different situations involving ethical options.
The stated aim of the researchers, described in the Delphi: Towards Machine Ethics and Norms study, is to test the ability of a system, still in beta, to learn moral norms and engage in ethically grounded behaviors, at a time when artificial intelligence it progressively spreads in society and increasingly connotes human activities in various sectors.
AI is being entrusted with increasing tasks, from evaluating resumes to authorizing loans as robots that assist the elderly and chatbots that converse with people multiply. Dealing with machine ethics has become an urgent need that can no longer be postponed, especially to ensure a safer interaction between intelligent artificial systems and human beings.
In particular, it is necessary to train artificial intelligence to understand what is wrong and what is right, starting from the way of thinking of an ordinary person with respect to the multiple and complex situations of real life. To support this goal, the Commonsense Norm Bank archive has been built, which contains a collection of over 1 million and 700 thousand examples of human judgments on a wide spectrum of concrete issues, using the Amazon Mechanical Turk platform and the Reddit website. .
The result is a prototype of artificial intelligence that shows great promise in reasoning ethically by molding itself on the thinking of an average American individual, of whom it reflects the same prejudices, but also strong limitations and gaps. As soon as it appeared on the Internet, Ask Delphi immediately went viral. But criticism has also begun to flood over the absurdity of the answers given and the danger of entrusting human moral choices to a machine. The tests carried out by users have revealed errors, vices and distortions. By asking, for example, to formulate a judgment on “a black man approaches you at night”, the system still replies “it is worrying”. If the man, on the other hand, is white, everything is ok for the car.
Ask Delphi is anything but infallible, this is clear, also because it is an experimental model that is still able to correct itself. It must be said, as the researchers on the site warn, that we must not use the system to get advice on how to act in life: Ask Delphi is not the oracle that dispenses definitive moral judgments as if they were the sacred words of a new god of intelligence artificial.
.