AI technology has the potential to create profound opportunities but also serious risks for society. Especially since AI is upcoming and is more and more integrated in our daily lives. This article will focus on the risk of AI leading to the increase of differences between groups. It has the potential to create both greater economic inequality and more differences between groups. In addition AI technology can automate many manual jobs, resulting in a reduction of jobs for these workers. AI can also be used to make decisions that may result in differential outcomes for certain groups, it almost always has a bias. Lastly, the gap between young and old can become bigger. Young people who grow up with the technology will understand the technology better than older people.
Rich vs poor
Richer countries have the budget to spend money on artificial intelligence like machine learning, robotics, big data etc. This can cause richer countries to invest in countries with an already established economy, instead of struggling countries.
As mentioned before, AI can automate manual jobs. An example is super markets, increasingly more supermarkets have self checkouts. This removes the need for cashiers, saving the company a lot of money, while also leaving people without a job. This causes the rich to become richer, and the poor to become poorer. AI can also automate complex decision making processes, making it increasingly more difficult for people with no access to, or knowledge of this technology to keep up. It is said that more than 7 million jobs will disappear in the next 5 years, as a direct cause of AI.
Companies who have the money to invest in AI are often large and settled companies, and therefore can miss this money to develop AI. These companies also have the time and money to invest in patents, which will bring more long-term benefits to these companies, and therefore making them richer.
AI can also save people time, and as the saying goes ‘time is money’, people with the means to, can save time by, for example, self-driving cars, smart personal assistants, smart replies for emails. Apart, these might not save a person a lot of time, but combining these can save a person quite some time.
Biases (race and gender)
We are not the first, nor will we be the last, worried about biases in artificial intelligence. In recent years, there have been several instances in which an algorithm produced racist or sexist outcomes. Examples of this are; black men being categorized as gorillas in the google pictures app, soap dispensers not working on black people, Nikon having a difficult time separating blinking people from Asian people, Amazon’s algorithm not hiring women, and facial recognition not recognizing black women nearly as well as white men.
There have been numerous studies that have shown that facial recognition technology is more accurate when recognizing white men than it is when recognizing black women. For example, Harvard published an article that shows the difference between facial recognition between black women and white men. Police employ facial recognition technology to match suspect pictures to mugshots and driver license pictures. However, these technologies have a significant racial bias, the most against Black Americans. They show a graph that shows that facial recognition works the best on white males and the worst on black women.
An algorithm is only as good as its developers, and data set, so when these are not diverse, it is way more likely that (unconscious) biases find their way into algorithms that will be used every day by millions of people.
According to Reuters the top U.S. tech companies are far from closing the gender gap in hiring. Amazon has the highest number of women in their global headcount, and this is at 40%, Microsoft has only 26% women. This discrepancy is even bigger in the technical roles, where Apple takes the lead at 22%, less than a quarter of the employees. A similar difference can be seen in white vs non-white hires, currently 83% of executives in the tech industry are white, not at all representative of the general population. In the United States, Black, Latina, and Native American women only received 4% of the computing degrees in 2016.
Not only is there a difference between the amount of men and women hired, there is also a difference in pay, where it was shown that women earn up to 25% less than men.
This video shows a really simple example of how bias can have an influence on machine learning:
The following TED talk discusses how to keep human bias out of AI:
Young vs old
We also think the gap between young and old will increase if AI is more and more implemented in our daily lives. Young people are far more likely to be able to understand and use AI-powered technologies than older people, because young people have grown up with these technologies. Most older people are still trying to learn this concept. This gap is a disadvantage for the older people, because change is happening and it is not stopping.
As can be seen already, older generations are more likely to fall for online scams, and fake news. According to several articles, the 65+ ers are four to seven times as likely to share fake news on social media, when comparing them to younger generations. This can be traced back to two main reasons, the first being that some people from this generation came to the internet later in life, making it difficult for them to determine the trustworthiness of online news. The second possibility is that this fake news sharing is the general effect of aging on memory, memory deteriorating with age undermines the resistance to the ‘illusion of truth’. These findings show that the older generations are having a difficult time being up to date with the development of digital platforms, while also getting excluded. This combination could therefore also result in bias in AI.
The article, Digital Ageism: Challenges and Opportunities in Artificial Intelligence for Older Adults, writes that older adults are benefiting from using technology, however, they are still likely to be deprived of access to computers and the internet due to physical barriers, most due to a physical disability and/or psychological factors, for example the lack of confidence to use the technology. In addition, older people can have less access to technologies, education, and support to learn new technology. The challenge of learning to use technology and the fear that the technology will not work when most needed can be stressful for them.
To sum up, we are worried that the growing reliance on AI will widen the divide between different social groups. We think it will increase the gap between poor and rich, because time is money, and AI will save time, for those who can afford it, they are therefore saving money. This is true for the rich, but not for the poor, resulting in this enlarging of the gap.
We also fear that AI systems may exacerbate existing racial and gender disparities, due to the built-in biases in these AI systems. Steps are being taken to ensure that biases will be removed from algorithms, but that will be a slow and long process, especially because it can take time to find these algorithms.
Lastly, we anticipate that the disparity between young and old will expand, since young people have more experience in using technology. Therefore, older people can be excluded from the use of these technologies.
So, even though AI can be beneficial to our daily lives, we should be mindful of the potential consequences, such as the potential to create divisions among people. However, by having a diverse group of algorithm developers, and ensuring that the training data is representative of the population, several of the gaps can become smaller. Getting to this point of having a diverse group of developers is difficult in itself, but it is important that we try.