ChatGPT — or Chat Generative Pre-trained Transformer — has received major media attention for its human-like dialogue and “articulate” answers across a range of knowledge.
This chatbot, developed by OpenAI, was initially launched on November 30, with a stable release on February 13.
ChatGPT, built upon OpenAI’s GPT-3 “family” of large-scale language models (specifically, ChatGPT 3.5), has continued to be fine-tuned, employing both human supervised and reinforcement learning techniques, programming languages such as Python and bulletin board systems, and gathering data from ChatGPT users and internet phenomenon.
What are ChatGPT’s potential capabilities? It can develop computer programs; compose music and write poetry, song lyrics, and essays; summarize long documents; answer test questions; play certain games; and simulate an entire chat room. Moreover, ChatGPT remembers previous prompts given to it in the same conversation.
While these capabilities are promising, more fundamental questions need to be asked: Is ChatGPT a product or process innovation, and is it a radical or incremental innovation? And is it something that consumers will view as value-added to their experience?
A product innovation is embodied in the outputs of a business firm, both goods and services. In contrast, a process innovation is an innovation that tends to focus on improving the effectiveness or efficiency of manufacturing this product or producing this service.
ChatGPT — or, more accurately, the technology that powers it — is a product innovation. This ChatGPT technology is expected to power a wave of new technology products in 2023, changing the way consumers interact with customer service chatbots, voice-enabled virtual assistants like Alexa and Siri, search engines such as Bing (and Google is launching its own ChatGPT competitor called Google Bard, while Opera is set to incorporate ChatGPT into its web browser), and e-mail inboxes such as Outlook, as well as Word and PowerPoint. Will the further development of ChatGPT (and, more importantly, its underlying technology) generate increased speed and efficiency in the process of manufacturing new products? Quite possibly.
Is ChatGPT a radical or incremental innovation? A “radical” innovation is conceived of as a combination of newness and the degree of differentiation. An incremental innovation might not be new or exceptional to an industry or firm but involve only a minor change or adjustment to existing products or practices.
ChatGPT technology is both new and different in its application from existing product innovation available to the consumer. Is it radically new and different — or an incremental innovation — from existing technology available to consumers? The answer to this question remains opaque, as its technical performance limitations will define its commercial usefulness — or demand — in the marketplace.
For example, OpenAI has stated that ChatGPT will “write plausible-sounding but incorrect or nonsensical answers,” as well as warning that it has limited knowledge of events after 2021 and it should not be used for advice.
In December, the question-and-answer website Stack Overflow banned the use of ChatGPT for generating answers to questions, citing the factually ambiguous nature of its responses. In addition, ChatGPT can plagiarize and can even deceive plagiarism checkers. Importantly, ChatGPT does not inherently know right from wrong, and it must be trained to see the difference — a difficult, costly and challenging proposition.
As to AI chatbox training (and further product improvement), Microsoft recently granted access to more than a million people in 169 countries to the new version of Bing, powered by OpenAI chatbot technology. Nathan Lambert, a research scientist at AI startup Huggingface, would have recommended that Microsoft performed its million-subscriber start-up as an experiment rather than a startup.
“I think there are going to be a lot of harms from this kind of AI, and it’s better people know they are coming,” Lambert said.
Dr. Celeste Kidd, a psychology professor at the University of California at Berkeley, points out that research has shown that people learning about new things have a narrow window to form a lasting opinion. Thus, Kidd said, observing misinformation during this critical initial period of exposure — to chatbot AI — could do lasting harm to a product’s acceptance.
Interestingly, in 2019, OpenAI decided not to release an earlier version of the underlying technology powering ChatGPT and the new Bing because company management deemed it dangerous. The ChatGPT technology will have to overcome significant performance challenges — as to accuracy and bias — before consumers are willing to fully embrace (and subscribe to) this still unproven, evolving product technology.