News and social media are buzzing about artificial intelligence. Since the introduction of “chatbots” that answer questions in the form of sentences or even entire blogs, everyone is talking about artificial intelligence. So, what exactly makes artificial intelligence, intelligent?
The first step is to understand the concept of A.I. According to Wikipedia, “artificial intelligence is based on the assumption that the process of human thought can be mechanized.” Google shares a similar explanation, stating “A.I is the broader concept of enabling a machine or system to sense, reason, act, or adapt like a human.”
Like many others, I started to wonder where the source material to educate and train these machine learning and A.I. algorithms comes from. After all, the output of any computing system, regardless of complexity is only going to be as good as the information going into it. . . . . right?
So what better place to start than to ask an A.I. where it gets its source materials from? For this experiment, I chose the popular “ChatGPT” bot to enlighten me. I asked my question in two specific forms and received the following responses:
The second time I rephrased my question a bit:
In the two responses I received from ChatGPT, both mentioned the use of “articles” as source materials. There’s nothing inherently wrong with using a variety of source materials to provide a broad range of perspectives. It does however call into question, whether the A.I. can be misled? More specifically, can the A.I. be biased?
To illustrate the bias concern, let’s take a brief look at a typical news cycle here in the United States. Anyone familiar with any news media will likely tell you that sources of news can be biased. Climate news is a great example of how media can show bias. Some media outlets are stating that there are no climate issues and no reason for alarm. Other media outlets are concerned stating there is indeed a climate issue.
Another example of a biased concern could be political. As everyone is familiar, there are multiple political parties with vastly different opinions on issues affecting their constituents. Depending on the news media reporting on the subject political issue, you are likely to get a different opinion as to the proper perspective of the issue, and the overall “who’s right and who’s wrong” on the subject political topic.
Another source shared by ChatGPT for the source material is “books”. Back in 2020, the New York Times published an article showing the differences between textbooks used in educational institutions in both California and in Texas and pointed out the differences with respect to the Second Amendment of the U.S. Constitution. According to the article, the differences in these educational materials are “shaded by political bias.”
The question now becomes “who is feeding these machines their source materials?”. Based on the New York Times article discussed above, an artificial intelligence designed in one geographic location can have a significantly different response to a query than one built in a different geographical location. The subject article was simply looking at two states. What happens if the A.I. is built in different countries with different social and political structures? Different laws and beliefs?
As of this writing, I don’t have any concrete answers. I just find myself with a growing list of questions. Artificial Intelligence will continue to grow rapidly over the next decade, and the answers to these important questions will significantly shape the engagement, guidance and responses we will see from these artificial intelligence platforms. We will continue to keep an eye on the evolution of A.I. technology and update this blog as new iterations are presented.
N.Y. Times | Texas vs California History Textbooks https://www.nytimes.com/interactive/2020/01/12/us/texas-vs-california-history-textbooks.html
SAS | What is Artificial Intelligence https://www.sas.com/en_in/insights/analytics/what-is-artificial-intelligence.html
Orient Software | A.I. in Cybersecurity https://www.orientsoftware.com/blog/ai-in-cybersecurity
