Artificial intelligence (AI) is one of the biggest technological trends of our time. Inspired by the way in which the brain learns, it uses unfathomably large volumes of data to unearth insights and even respond to those insights with specific actions. Yet, AI has remained controversial. Elon Musk has referred to AI as an existential threat to humanity ; Bill Gates has called it both promising and dangerous , and many other thinkers and business leaders have expressed concern about how humanity will live in a world of AI. Over the last few years, concerns have arisen that AI is biased. An article in Search Enterprise AI explains, the government is starting to act on those concerns.
Over the last decade or so, there has been a rise in concern over the dangers posed by AI. Gates located the gravest danger posed by AI in its applications to warfare. AI can be used in idigital, political and physical attacks, to impersonate public figures using speech synthesis; the manipulation of the public thanks to the analysis of the beliefs, behaviors and moods of people; automated hacking; and in physical weapons such as swarms of micro-drones, among many other applications. For Gates, AI is unique for bringing both promise and danger.
Musk believes that AI is potentially more dangerous than nuclear weapons. Musk worries that we are calling forth a demon , and that demon may be beyond our control. Musk has, however, invested in AI companies and OpenAI, the artificial intelligence research laboratory, because he believes that his input could help prevent AI turning into something deadly. Musk has been calling for greater regulatory oversight on the development of AI, believing that the current free-for-all risks placing humanity in danger without any discussion over its dangers. Other entrepreneurs and tech leaders, such as Bezos, have expressed similar weariness about the dangers of AI.
Yet, all this missed a danger that exists, not in the future, but in the here and now. For many people, the idea of AI brings up notions of hyper-objective intelligence. It does not strike most people that a machine could be as biased as a human being. AI has been seen as a tool that can super-charge the social sciences, and help us discern injustice and inequity across the gamut of human existence, and unearth patterns that could help close the gaps in inequality. It has been praised as a tool that can help deliver greater access to healthcare, nutritious food, water and other basic needs, regardless of age, race, class or gender. However, research has shown that AI is often biased.
In 2014, Amazon discovered that a program its software engineers was building to review job applicants’ resumes discriminated against women when assessing their competence for certain technical roles. The program was scrapped.
When the National Institute of Standards and Technology (NIST) reviewed facial-recognition algorithms from almost 100 developers from 189 organizations, including Intel, Microsoft and Toshiba, they found that there were demographic differentials in the bulk of the algorithms the project studied. In other words, the algorithms were biased. They still are.
The reason is quite simple. The insights that AI gains are derived from the data it is fed and parsed through the analytical framework it was built around. As the old computer science saying goes, Garbage in, garbage out . If AI is fed biased data, it follows that it will spit out biased insights. Objectivity is an illusion. Ai may be able to perform specific tasks faster than any human being could, but it is not objective. Subjectivity is built into it through the data and assumptions that drive it.
The concerns around AI have led to efforts to regulate it. The Federal Trade Commission (FTC) has instituted regulations around the use of AI in lending, with the Fair Credit Reporting Act (FCRA) at the heart of its efforts. It has also included AI regulations in the Equal Credit Opportunity Act (ACOA) and the FTC Act.
These laws do not specifically address AI, but the FTC has issued guidance in the last two years in which it has stipulated that lenders cannot use biased algorithms or algorithms that they do not understand or cannot explain, when making decisions about consumer credit, housing, employment, insurance, housing and other benefits.
Under the FTC Act, racially biased algorithms cannot be sold or used.
These efforts are the first federal level efforts to regulate AI. The United States is still behind the European Union, which proposed a framework for AI regulation in 2018. The precautionary principle tells us that if something could cause widespread harm to humanity, then development must be halted until more is known about that thing and an informed decision made about it. This idea flies counter to the tech industry’s impulse to innovate at full-speed. So what we have is a situation in which technology is evolving faster than our understanding of it, and as our naivety about the promise of AI is replaced by a more mature understanding of its promise and danger, we realize that we are unprepared for the dangers.
We have to balance the good that AI can deliver with regulations to contain the bad, such as biased algorithms.
