ChatGPT can do what many researchers take pride in, it can write. What does this mean, is it a eureka moment or a very big mistake?
While ChatGPT may seem like a very tempting alternative to writing. It does have its flaws. The following are my impressions about its accuracy, ethics, use in plagiarism, and what are the future possibility with this tool. My impressions are based solely on academia.
For those of us who are unaware of the capabilities of ChatGPT. Look it up for yourself at https://chat.openai.com/chat. ChatGPT has been published by OpenAI. The company behind DallE and many other impressive AI research. DallE is the artificial intelligence image generator that has been making a lot of waves in social media circles lately. Read more about DallE here.
Accuracy of ChatGPT
ChatGPT is really good at what it does i.e. communication. It is, in its structure a chatbot. Ask anything from it and it will answer as a real person. Much like, Morgan Freeman answering a boy’s general knowledge questions in the opening scene of the bucket list.
While ChatGPT may seem perfect. It is only when you ask it what you already know that you start to find slight irregularities in its answers. Its answers are believable and authoritative but they are sometimes not accurate or miss quality. This is very much accepted by an AI language model. Also, this slight inaccuracy is only for very few instances where a question is out of its range (or in this case training data). Many times ChatGPT will tell you that it doesn’t know the answers but in one instance. It did share believable results that were completely made up.
As a Ph.D. student looking for a quality research paper can be hard. So I gave ChatGPT a try on this, asking it to tell me about research on a very particular new domain. While Google Scholar and other platforms failed to give very accurate results. ChatGPT gave me what I was looking for. It told me that yes, there are papers available on this topic. As ChatGPT cannot share online links. I asked if it can share the Digital Object Identifier (DOI) for these papers. It replied yes. Once I searched for these DOIs on multiple platforms. For what should have been direct access to papers quoted by ChatGPT. All returned the papers that were not remotely related to the topics. I concluded that the DOIs were just copied and pasted from other papers without any relevance. Okay, that could be a mistake. Let’s search for the paper titles along with the author’s name. This usually returns the paper as the first result on Google scholar. Again, there was no evidence that the paper ChatGPT was talking about even existed.
ChatGPT is good. Sorry, let me rephrase. ChatGPT is really good. But it is a communication bot. It cannot generate information that has not already been published. So as long as the research domain is concerned, that is cutting-edge research. ChatGPT cannot help. You can ask it for understanding a concept that is in the public domain and it will give your accurate results but not otherwise. Also, if a bot can answer your questions about the research you want to do. Do you really think you are doing anything new?
Ethics of ChatGPT on Academic and Research
Most of the ongoing research in many domains only reproduces work by applying existing technology to a new area. While this is in no way bad or unethical. In fact, technologies that solve and can be applied in multiple areas are needed to accelerate research. ChatGPT will affect this type of research as now progress will not depend on who has the knowledge but on who can ask the right questions and gather information fast.
ChatGPT can very accurately write articles, essays, letters, programming codes, and homework. This will affect a lot of schools and universities and create an extra burden on them. But hear me out, we don’t need to fight it. Long back when calculators were introduced it was prohibited in our school and house. We were to do all the calculations with pen and paper or in our heads and it felt like cheating when people were allowed to use calculators. Now, as someone who can use calculators in exams. We know, having a calculator does not make it easy to answer all questions. Similarly, as technology advances, we have to accept these tools and learn to live with them.
Plagiarism with ChatGPT
Yes, simply using ChatGPT as a tool to do your work is ethically wrong and OpenAI understands that. They are already working on a plagiarism tool to detect content produced by ChatGPT. Some third-party tools with doubtful accuracies are already available online. ChatGPT is so good that finding if the content is produced by it is difficult. Moving forward, our children will not be judged on how they produce work or homework. They will be judged on what they know.
Quality of ChatGPT responses
I am echoing what many have already pointed out. If you have the knowledge on topics you ask ChatGPT about. You will soon realize that its response is very mediocre or to say in other words, too structured. Sooner or later, we are going to generate a pattern recognition system in our brains for these types of responses. And then what about literature? I can’t imagine a chatbot generating literary masterpieces as we have today. Or is it just a matter of time?
Whatever it is, I am happy ChatGPT is here. It is only going to improve. I am excited to see how we acclimatize to it and how it advances us as engineers.
PS: Unlike many stories on ChatGPT available on Medium. This one was not written by ChatGPT. Every single word was written by me, or was it?
