It is not always that the implementation of autonomous electronics into everyday reality runs smoothly.
Another piece of news, which again caused series of discussions around AI technologies and their physical realization in real life: in the city of Tempe, Arizona, United States, an Uber self-driving car hit a pedestrian. As a result of the accident, the woman died.
What is the way to avoid such problems in the future? What are the conclusions to be drawn? What risks should be envisaged by those who intend to use the potential of new technologies in their mobile and web applications?
Like many other emerging technologies, AI holds tremendous opportunities and shows great promise, which we have more than once told about. But the pitfalls and the possible adverse consequences of the improper application are better to be learned in advance. As they say, praemonitus – praemunitus.
Let’s divide the AI-related issues into two large groups:
internal
Bugs, failures, errors, as a result of which systems behave in the most unexpected way, sometimes to the surprise of their creators.
external
Ethical and legal issues, misuse, or other moments arising as the artificial intelligence interacts with the real world.
Accidents Happen in the Best-Regulated AI Families
Unchildish Talk
Bot-talk
In 2017, Facebook created two chatbots designed to negotiate, helping people make orders, schedule appointments, etc. At some point, things took an unexpected turn: Bob and Alice began to communicate with each other in their own artificial language. The bots had not been originally given the instruction to communicate in a language understandable to people, so they simply chose their own way. As a result, the absence of one restriction led to a misunderstanding between the creators and their brainchildren, in the truest sense of the word.
Alexa’s Invitation-Only Party
In one of the apartments in Hamburg, Amazon Echo spontaneously started a party in the middle of the night. To be more precise, the artificial intelligence device began to play the host’s playlist at top volume. The neighbors were not inspired by the idea and called the police. As a result, the door was broken, the impulse party was interrupted. Amazon apologized to the owner offering to pay the fine and the bill for a new door.
Games AI Plays
Game over!
Elite Dangerous artificial intelligence started developing a super weapon, which had not initially been developed by the creators and hunting the gamers’ ships. It all happened after the deployment of The Engineers (2.1) update. In that case, the players had no chance to resist the new powerful weapons, and the developers had to intervene to save the situation and the ships of human players.
The Importance of Being Unbiased
A researcher at the MIT Media Laboratory (Massachusetts Institute of Technology) analyzed the function of three commercial programs intended to identify faces. For this purpose, Joy Buolamwini collected 1200 photographs of various people. The result showed that the neural networks are excellent at recognizing the faces of light-skinned men, while the share of the mistakes with the dark-skinned women was 34% more.
The conclusion is that in order to exclude bias in assessments, machines need to be “learned” on a large number of diverse examples.
“Open the Pod Bay Door, HAL”
Paper, Rock, Scissors with Sophia
In this regard, one cannot but mention Sophia (a humanoid robot). In March 2016, she made blush her own creator by the unabashedly given positive answer to his question about whether she wants to destroy humanity. To be fair, we should note that by the fall of 2017 her views grew softer. In one of the interviews, Sophia said that she is filled with human wisdom with the purest altruistic intentions and asked to perceive her that way.
We all have our faults. But those who are aware of the responsibility and severe consequences will take into account the mistakes of others to exclude them at the earliest possible stage. Developing an AI application for use in medicine, commerce, marketing or advertising, one can always learn from experience in other areas, even if not directly related.
Artificial Intelligence: Reality Tests
Legal Environment Game
The artificial intelligence technologies penetrate gradually into the different spheres of human activity. Accordingly, new questions arise, how the AI actions should be interpreted from the point of view of legal norms.
Refer, for example, to your own driving experience. Surely you have had to deal with situations where the driver needed to take a lightning and not always a single-valued decision.
Shall you pose a risk to your own passengers or save a child running across the road? Which algorithm will be chosen in this case by a machine? Even if we do not take into account the moral and ethical component, it is still unclear who is responsible for the consequences of the accident involving a self-driving vehicle: the code developer, the manufacturer or the car