Once you explore the realm of generative artificial intelligence (AI), you quickly realize its remarkable capabilities. However, for professionals, this remarkable aspect also presents a challenge. When a groundbreaking technology like AI has the potential to impact various fields, its adoption involves not only harnessing its machine capabilities but also managing the human element, along with addressing any unforeseen shortcomings, making decisions about its utilization more complex.
The efficacy of large language models (LLMs), such as those fueling AI systems like ChatGPT, in enhancing diverse tasks is evident in numerous studies. Whether it’s generating meeting summaries, analyzing data, or crafting press releases, LLMs offer a time-saving advantage. They have the potential to enhance customer service and even assemble IKEA bookcases, albeit without the human touch.
AI has the capacity to foster innovation. Research by Karan Girotra from Cornell University and his colleagues compared the idea generation prowess of the latest ChatGPT model with that of individuals from esteemed American universities. While a solitary individual can generate five ideas in 15 minutes, employing GPT-4 boosts this number to 200. Importantly, surveys indicate that the quality of ideas generated by AI surpasses those originating from humans. The abundance of choices enabled by such capabilities can sometimes lead to decision paralysis among decision-makers.
The utilization of LLMs comes with its share of pros and cons. On the positive side, wider adoption of conceptual AI can spur the development of numerous applications. Those well-versed in LLMs can leverage their capabilities more effectively. As noted by Reid Hoffman, a prominent AI investor and host of the management podcast “Boss Class,” experimentation is key. If you dabbled in asking ChatGPT to compose a haiku in the past but haven’t revisited it since, there’s untapped potential waiting to be explored.
Familiarity with technology can help mitigate the typical human apprehension towards it. A study by Siliang Tong from Nanyang Technological University and collaborators highlighted this phenomenon before the surge in conceptual AI popularity in 2021. It demonstrated that AI-generated feedback outperformed human managers in terms of enhancing employee performance. However, the revelation that the feedback originated from a machine led to a decline in respect, heightened job insecurity fears, and diminished performance. Exposure to LLMs can help alleviate these concerns.
Yet, there are inherent challenges, including technological glitches. Due to the propensity of LLMs to propagate misinformation, the Cambridge Dictionary crowned “hallucinate” as the expression of the year. While advancements are underway, a recent report from R. Thomas McCoy and Princeton University researchers suggests that certain issues are ingrained in the technology.
Off-the-shelf models can stumble upon unforeseen obstacles due to their training on probabilistic digital data for predictive purposes. While ChatGPT’s underlying LLM, GPT-4, can adeptly handle tasks like adding 9 to 5 and then adding 32, it may falter when confronted with less common queries like adding 31. These discrepancies arise from the variance in training data availability. Custom models may also exhibit similar errors.
Moreover, there’s a pertinent concern regarding the monitoring of employee AI usage within businesses. Unauthorized sharing of personal data through AI interactions poses a risk, as evidenced by Samsung’s recent restriction on ChatGPT usage following allegations of source code dissemination among professionals.
Navigating this intricate landscape of AI superpowers, conveniences, and pitfalls necessitates a nuanced approach from organizational leaders. Embracing targeted initiatives, establishing clear guidelines on data inputs for LLMs, understanding the technology’s intricacies, and leading by example are crucial strategies. While the allure of AI creation is enticing, mastering its implementation demands diligence and expertise.
