What are humans good for?

A Surviving Facts Blog

This week, I read an article on Bloomberg News about AI detectors falsely accusing students of plagiarism. The article attracted my attention because a similar tool had flagged one of my daughter’s essays for cheating. The professor asked no questions, gave her an F on the paper and was unwilling to discuss the situation.

I’m no helicopter mom, but I do have an academic background and understand the rules of various guides (Associated Press, New York Times, MLA, APA and other style guides). I offered to review my daughter’s paper so that I could teach her accurate use of sources. In other words, I assumed she had done as the teacher claimed- lifted work from other sources without adequate paraphrasing or attribution.

Instead, I found astonishing inaccuracies in a system a university uses to assess the authenticity of ALL student papers. Similarly, Bloomberg Businessweek identified that AI inaccurately detected cheating in a small but significant sampling of students. Most AI plagiarism detection systems falsely flagged non-native English speakers, neurodivergent students and students who write in a plainer and simplified manner. Ironically, as a former university writing instructor, I encouraged simplicity.

Catina is neurodivergent. When I read her essay and studied the original sources, I did not see actual plagiarism. There was no lift and shift overlap. Catina writes with eloquence and is a keen critical thinker. What was going on?

I asked to see the flagged passages again, as well as the annotated explanations of plagiarism. Here’s what I found: the plagiarism detection system was unable to distinguish between common use language, certain grammatical forms and commonly used couplings of nouns and verbs.

Here are a few examples:

• Catina’s essay was on Cannabis. Every time Catina used, “Cannabis is,” the system claimed plagiarism. Seeking a first usage of “Cannabis is” is impossible. So to provide attribution, we googled “Cannabis is” and cited a handful of sources followed by this notation: “Google cites over 10 million sources using this noun-verb combination. We have only provided a sampling.”
• The word “the” is an article that really can’t be avoided in writing. The system had flagged “the” as plagiarism at least 50 times.
• The system claimed random words used in other sources as plagiarism. Some examples: “scientists have found,” “laws have” or “laws have not.”
• Transitions also were called out. We changed “however,” to almost every synonym available,
• One paraphrase with identified source was flagged as plagiarism. Apparently, a student at a university in some southern state had used similar language in a college essay. We googled the student’s essay and trolled various student paper databases. We never found the other student’s paper. Catina had definitely not used it as a source, AND she had identified the actual source she did use.

Because of my academic background, I helped Catina write a response to her teacher highlighting these observations. Catina also made sure to note that the system had not identified a single instance of directly lifted original material. The extent of the claimed plagiarism was around the generic instances noted above.

Catina revised the flags anyway and sent her revised essay and note to the teacher. The professor never responded. I, therefore, encouraged Catina to speak directly with her professor. It took some effort since the professor did not appear during her published office hours, but, finally, Catina was able to speak to the professor directly. The professor had decided to accept Catina’s essay but noted she would have to receive a “C” due to the original flagging. Although Catina raised the issue with her advisor, no challenge process existed. And although Catina had given permission for me to speak, the professor refused to. She probably thought I was a helicopter parent when all I wanted was to speak one academic to another. I doubted she would reconsider the grade.

The examples used in Bloomberg’s article seemed to elicit the same response from professors. Have we become so jaded that professors refuse any relationship with students? It seemed to me that once Catina was “labeled” by this incident, she never was able to gain standing with the professor again. Ultimately, Catina chose to transfer schools, a decision that has worked well for her.

As AI systems are beginning to take over many functions and activities, we are losing an attribute humans have: judgment. AI isn’t ready to weigh one word usage against another or to assess whether one paraphrase is more effective than another. And most of all, AI understanding of common usage language and phrases is poor.

What does this mean for the corporate world seeking to reduce costs? Two recommendations: caution and closed systems. Elements of good writing such as an individual’s tone, vocabulary usage and sentence structures are still human and not large language models.

In the corporate environment, AI has a place, though to avoid plagiarism and inherent and unconscious bias (which AI has been shown to have), a closed and proprietary system is critical. Brands have unique features, intellectual property and voice. So far, AI hasn’t replaced these. Brands also attribute written work as original to a company, leader or author. Ethical issues arise if we credit non-original work generated from AI. I am happy to see the trend on social media and other places of needing to note if a piece of writing is AI generated.

AI can help achieve efficiencies. It can speed onerous tasks. It can wade through huge amounts of content quickly. It does learn over time. My advice to the writing teams I used to oversee was to use it for research, for quickly summarizing numerous sources, for providing insights on keywords. It also can help with organization by providing several outline options. The corporate writer doesn’t have to follow any one exactly, but can see the options without having to lose hours in a day. At times, AI can take the lead- such as for generic marketing or sales language. With a proprietary system, it can generate things like product descriptions. However, humans are still needed for review. Human judgment isn’t perfect but it can understand usages of legal or standard language that cannot be changed, no matter how the wording sounds.

I am supportive of AI. Even more, I am supportive of human beings. I have yet had AI provide that frisson in the head that ignites when writing is beautiful. Originality (we can debate its meaning) exists. I see it everyday.

So, go ahead, explore AI. But manage the “human replacement” expectation. As science steamrolls ahead, we have to ask an important question: in the face of all these advancements and how far they can go in the not too distant future, what are humans good for? We better answer that question- as a society, as nations- lest future generations wake to a world where machines work with the help of selected few genius human beings. The rest of humanity will no longer have work or societal purpose, leading to depression, despair, substance abuse and crime. Humans actually need work- as much as we fantasize about winning the lotto. What humans need even more is purpose and contribution, as well as the belief of being able to effect progress and change. Without that, we will continue to cut off Americans from productivity. Without retraining, we could see similar trends as in the rural south, Michigan, coal-mining towns and other places built around manufacturing. Many of these communities now live in poverty, without education and healthcare and with drug abuse, crime and aimlessness. What parameters we set with AI today will determine this future.

Be sure to subscribe to this blog, like and share!