Ever since OpenAI launched ChatGPT late last year, the Internet has been abuzz with debate about whether continuously improving AI tools can or should replace humans in a variety of jobs. But according to Alan Fern, professor of computer science and executive director of AI research at Oregon State University’s College of Engineering, AI chatbots still mostly work best as tools for programmers rather than as programmers themselves. He believes that when it comes to the more thoughtful design decisions, humans are not going anywhere anytime soon.
“There is already a ChatGPT-style system for coding called Copilot, and it’s basically a GPT model whose training was focused on code, GitHub code. I’ve heard many very good programmers say that tool has improved their productivity, but it’s just a tool and is good at the mundane things that take programmers time to look up or learn,” he said in an email to Government Technology. “Copilot still will produce erroneous code, just like ChatGPT produces incorrect statements, so humans must still be in the loop. These models don’t really reason at a deep level and there isn’t a clear path to getting them there. It is for that reason that I think programmers will be employed for a long time, but the efficiency will improve dramatically.”
“The types of jobs that might become obsolete or much reduced [by AI advancements] could be those that are mainly about eloquence but do not require deep thinking. Some customer service jobs are like that,” he added. “The difficult thing to predict is what jobs, companies, industries, will be created.”
Dakota State University computer science professor Austin O’Brien agreed that while AI has made dramatic leaps in its ability to copy human writing, for example, it still has a long way to go before it can be trusted to do coding. He said the technology is still prone to making mistakes like AI hallucinations, which happen when an AI model generates output to an inquiry that makes little to no sense.
“ChatGPT was trained on human language with the goal of producing human-like text. It’s clear that code repositories were also used for training, and I’ve seen some very impressive output when asked to produce code similar to assignments I have given to students. That said, I’ve also asked it to produce a few things that aren’t possible in code, and it would give its best shot, although it was quite incorrect. This occurred when ChatGPT was first released, but trying it again recently, it now lets me know that it’s not possible, so it appears that they are continually updating it with new information to make it better,” he wrote in an email. “Since it’s based on natural language models, it’s mimicking what it has seen before in that context and doesn’t have a deeper understanding of the algorithms, data structures, or possess general problem-solving skills. It can’t truly extrapolate new solutions to unknown problems and will likely struggle when new ones are presented.”
While using current AI technology to replace coding professionals may be years and years down the line, and especially for more advanced software development functions, O’Brien expects some jobs more generally to become obsolete due to advancements in AI. He said this is already happening in careers such as data entry and customer service, slowly but surely.
“With the loss of these jobs, there is typically an increase in job creation in other areas, usually in the technology industry itself, like AI, cybersecurity and data analytics. I don’t, however, really think it’s fair or reasonable to tell someone who may lose their job to simply learn a new technology skill,” he said. “I think it’s important for transition programs to be in place to help these people procure new jobs in the changing market. … History is full of examples where workers have been displaced by new technology and the job market had to adapt. It’s not necessarily a new problem, but one that must be addressed again soon.”
Saurabh Bagchi, a professor of electrical and computer engineering and computer science at Purdue University, said ChatGPT-like AI tools appear to be getting better at generating “snippets” of code, but agreed that the technology is still not completely reliable by any means.
He added that when ChatGPT puts together a piece of code, there is no way of tracing it back for attribution to see whether it comes from licensed software packages, which could present intellectual property concerns for those using it in its current form for software development.
“It’s a quantum leap over where AI code assistants were even two years back,” he said. “But a lot of the industry colleagues that I work and collaborate with that I hear from are a little cagey about using code generated by ChatGPT. It’s not clear how secure or reliable ChatGPT-generated code is. This is under active investigation in academic labs, including ours, and we hope to get a better idea within six months or so.”
While the technology is still not capable of replacing human programmers responsible for updating and maintaining large-scale software involving efficient algorithms, legacy systems and languages, computer science professor Amanda Fernandez of the University of Texas at San Antonio said in an email that AI chatbots could prove helpful in jobs like journalism for preliminary topic research, or for helping teachers create lesson plans, for example.
Still, she said, “a human will always need to be in the loop to verify accuracy” with the AI’s output.
“There have been many innovations in programming meant to ‘remove the programmer’ from needing to write code over the decades. For example, COBOL programming language was meant to make it easier for anyone to write code [through] a programming language which reads more like English, as opposed to assembly language or binary,” she said. “These tools and concepts have changed the way programmers complete tasks, and technologies like ChatGPT will similarly impact these jobs as a useful resource.”
But O’Brien said that advances within the field of AI outside of programs using natural language processing could eventually be a different story when it comes to whether humans will be replaced in any given job field, or for more advanced software development roles.
“One day, a new technology that possesses these traits may come along, but I don’t think ChatGPT is it,” he said.
