Li Yen Ho
Tesla’s self-driving cars and GPT-3’s natural language processing (NLP) have showcased the prowess of AI, bringing us that much closer to the dream (or nightmare) of computers helping the world (or running the world). Despite this, it’s important to remember the limits of this technology and therefore the precautions we must take to shield against a wide variety of dangers AI pose.
It is easy to give AI too much credit for what they do considering all we hear about their achievement. It is also easy to forget that no matter how complex the system is, the results are entirely dependent on their sample data. AI lacks creativity and can only mimic its source. For example, image recognition software does not know what is in the picture, only that some pictures with some similarities are regarded as the same. At best, AI’s misunderstandings can be the AI accidentally focusing on the referee’s bald head (video).
However, at worst, AI can prevent people from getting the medical treatment they need due to long term effects of systemic racism the AI quite obviously was not aware of (for more information). Another report details the AI’s decisions in the stock market that cost a person millions of dollars (for more information).
Despite these accounts, it is important to remember that artificial intelligence holds no ill intent. Simply put, the technology isn’t advanced nor intelligent enough to have any moral intent. AI optimizes its algorithm to find patterns in the sample data to return the result that matches the answers, no more no less, and this is where the ethical dilemma really begins.
AI is notorious for picking out small details that evade humans. From the angle of the image to the background of the shot to noise in the image itself, AI has been compared to a toddler that scientists must keep a close eye on, one that sees the world in a fundamentally different way than humans This is combined with the fact that some AIs are black boxes, meaning that it is literally impossible to find out what patterns they see and how they come to their decisions (think of the Youtube algorithm).
As a result, the first reason makes bias prevalent in AI while the second makes it difficult to solve. One famous example of this is Google Translate’s fight against gender bias, where Google’s natural languaging process associates words like “nurse,” “assistant,” and “beautiful” with feminine pronouns and words like “clever,” “engineer,” and “CEO” with masculine pronouns. Of course, Google’s made great strides to combat bias within its algorithms, but still faces challenges stemming from systemic bias within humans that it is responsible for solving within their technology.
All of this comes to the root of the issue: unpredictability. AI constantly manages to surprise us with results that are very much not what we want, considering factors people do not think of or even notice in their decisions.
Artificial intelligence continues to amaze us with every new tech article we read and advertisement video we watch, but it is vital that we consider all of its shortcomings to prepare carefully for the decisions it will make. AI technology is improving rapidly but will forever remain a party trick for entertainment if we cannot solve the issues plaguing it.