Debate on AI’s Role in Society Intensifies After ChatGPT Tests 

The ChatGPT and other chatbots released this year have spawned a debate about the benefits and drawbacks of artificial intelligence, but also its impact on society.

The newest AI, ChatGPT, was introduced by OpenAI three weeks ago without any fanfare, but it took the world by storm. Users can ask simple queries and get answers in return.

The scope of the AI tool has generated a lot of buzz about the potential of the technology, which could alter the way work is done. Moreover, the idea of machines having human-like creative and technical intelligence has raised alarm bells on how AI could impact daily lives.

Companies and individuals have been trying ChatGPT and DALL-E 2 to find out how the chatbots will impact their business or careers. DALL-E 2, which is also from OpenAI, is an AI tool that generates artwork based on text descriptions.

There are worries that ChatGPT and DALL-E 2 may replace human jobs in areas such as content creation and programming. But testers are concluding these tools are excellent augments to help humans and AI work together.

Many have found that ChatGPT’s answers to queries are incorrect or verbose. StackOverflow, a forum where human experts answer technical questions, was among the first to ban responses generated using the AI.

"Overall, because the average rate of getting correct answers from ChatGPT is too low, the posting of answers created by ChatGPT is substantially harmful to the site and to users who are asking or looking for correct answers," the site said in a letter to the community.

Others have raised concerns about students using ChatGPT to write essays, which may be hard for professors to identify. But the story is different on applying AI to more creative endeavors.

Michelle Morris Petta, an art teacher in Fort Worth, Texas, tested DALL-E 2 by posting a simple description asking the AI to create artwork on iconography in Ukraine during the ongoing war.

Generated by DALL-E 2, December 2022

The AI generated high-quality artwork on flags and key religious symbols that are influencing the artistic narratives. But the faces on icons were distorted, and Morris Petta wasn’t too thrilled. She concluded the AI art was more “folk art – not exactly what they should be.”

“AI doesn’t doom human artistry. I think some students use something similar to get ideas, but ultimately still create their own work,” Morris Petta told EnterpriseAI.

Others concluded AI could be valuable additions to their workflows, but only with human gatekeepers.

Nginx, a company known best for its web serving software, tested ChatGPT on its ability to troubleshoot issues with its software. The results? It's fun and better than other chatbots, but responses were a mixed bag.

"In some instances, it appears to do a good job providing detailed responses to programming questions. But its coverage remains spotty and your mileage will certainly vary with regard to NGINX information," said Robert Haynes, technical marketing manager at F5, which owns Nginx, in a blog entry.

GitLab also put the ChatGPT to test to see if it could generate code that could be plugged in its DevOps pipeline. DevOps revolves around collaborative software development and generating code that could be tested, verified and deployed quickly.

ChatGPT was able to generate sample code quickly, which was put into the CI/CD pipeline, but it did not pass the testing phase after generating errors. The AI was able to provide workarounds to help a software engineer solve a problem.

"ChatGPT wasn’t able to provide a working solution, but would be able to provide solutions and ideas to an engineer with some experience with the codebase," Gitlab engineers said in a blog that detailed the tests.

Gitlab concluded that context or human touch may still be needed to solve particular queries sent.

"Can we use ChatGPT to contribute to GitLab? At this time, we’d say, 'yes,' and you will need some understanding of the code to complete your solution,” the Gitlab engineers wrote.

Network engineer David Bombal drew a similar conclusion after testing ChatGPT, saying it could automate coding. In tests posted on Youtube, the chatbot was able to create C code, generate a Python script, and establish a Cisco network configuration.

Bombal drew analogies of ChatGPT being somewhat similar to how Google simplified the way humans solved problems and searched for answers. The AI will be able to generate code or suggest ways to solve network problems, and augment human knowledge on the subject.

He also said ChatGPT will allow computer science students to stop the practice of memorizing coding and commands by rote, as AI would take over that job.

"In 2023, make it a priority to learn artificial intelligence," Bombal said, adding "become one of the people that changes the world, rides the next wave, do things the new way."

ChatGPT and DALL-E 2 are part of a larger trend called generative AI, which allows simple text-based queries to generate answers. At the core of ChatGPT is OpenAI’s GPT-3 model, which is also used in DALL-E 2.

The supervised training model behind both AIs involves a big data approach – acquiring data from different sources (including user inputs) with consistent labeling as a big part of the process.

The ChatGPT model, for example, was fine-tuned to GPT 3.5. ChatGPT uses a more refined and flexible AI technology to select the best answers to a user query. That interaction is then sent back to ChatGPT to strengthen the AI.

The race is now on to bring more context to bots like ChatGPT and DALL-E 2 to be able to provide detailed answers in verticals that include manufacturing and health care.

Nvidia's NeMo LLM service, which was introduced in September, adds the intelligence and interactivity tool for users to harmoniously interact with complex AI models in domains such as biotechnology and medicine.

Users can type simple medical questions, and the NeMo LLM service will be able to parse the language and provide an answer. The NeMo LLM service is built on pre-trained models that include the NeMo Megatron model, which has 530 billion parameters, and a tuned version of GPT-3, which has 5 billion and 20 billion parameter variants.

ChatGPT was trained on Microsoft’s Azure Cloud infrastructure, according to OpenAI’s website. The organization did not provide further details on the hardware configuration, and Nvidia did not respond to requests for comment on whether its GPUs are used in the underlying hardware that helps ChatGPT process answers.

We turned to ChatGPT to find answers on its underlying hardware. Like a seasoned PR person, ChatGPT expertly dodged the question with a boilerplate answer that the AI was part of GPT-3, and ran in the cloud.

“In general, language models like GPT-3 are trained and run on a combination of hardware and software that is optimized for machine learning tasks. This can include a variety of different types of hardware, such as graphical processing units (GPUs) and tensor processing units (TPUs), as well as specialized software for training and running the model,” the ChatGPT AI responded.

Others have called ChatGPT a more refined version of Google. Some tests on ChatGPT and DALL-E 2 set imaginations wild about AI's potential in areas like content creation. But others raised concerns about the technologies.

In its test, Gitlab acknowledged the ethical and legal issues surrounding AI generated code. For example, Oracle sued Google in 2010 for replicating copyrighted Java code in its Android OS, a fight that Google won in the U.S. Supreme Court last year.

"Generative models can produce data from their training sets, so it'll be hard to let ChatGPT loose inside of a company that must conform to data compliance rules," Naveen Rao, CEO of MosaicML, told EnterpriseAI.

The impact of ChatGPT is going to be lower than most think, Rao said. OpenAI opened the can of worms, but most of the use cases will be "fun" for now, he said.

"Also, companies need to build models that are experts at particular problems...that'll be based on their unique data," Rao said.