Source:
Artificial intelligence is currently one of the hottest technologies that is believed to hold immense potential to benefit the human society. That said, there are inherent issues that raise concerns about the potential dangers of the technology.
Some of the most discussed concerns around artificial intelligence include:
Bias: Probably one of the biggest concerns, bias in artificial intelligence creeps in when the training data for mathematical models is skewed. AI being a data-driven technology reflects the bias in the output. Many instances have been reported where translation systems produced biased results in terms of gender, race, and religion.
If left unchecked, the bias in data can have dangerous consequences. This prompted the European Commission to publish ethical guidelines to help address bias and errors in data before using it for training purposes.
Trash talking robots: In a research, Carnegie Mellon University has established that robots trained to utter discouraging words can perturb the users. Although the trash talk for the purpose of the study was mild, Pepper–the robot uttering these words–successfully discouraged its human opponents in the games they were made to play together.
In instances where humans and robots compete, rather than collaborate, trash-talking robots can significantly lower the morale of the humans which may impact their productivity and mental well-being.
Self-learning went wrong: The incident where Facebook’s robots Alice and Bob went out of control and began interacting in a language of their own is still fresh in memory. The claptrap that the two robots developed appeared to be gibberish to the scientists who entrusted the robots with the task of learning by communicating among themselves.
Later, it was discovered that the faux pas was a result of a programming error that failed to instruct the robots to communicate in comprehensible English according to the rules of the language. As a result, the robots made their own changes to English to create a language only the two of them could comprehend.
Inadequate or incomplete training can, therefore, have dangerous consequences, however unintentional they may be.
Addressing concerns is not easy
As mentioned before, artificial intelligence is a data-driven technology that provides outputs based on the data used to train the machine learning models. Therefore, the chances of machines replicating human mistakes, prejudices, and behaviors are high. Addressing these concerns is by no means easy and will continue to place hurdles in the wide-spread adoption of the technology.
