Human VS AI Insights
Based on Goldstein’s (2019) definition, human insights are characterized as sudden, involving the reorganization of mental models, and creating original solutions. These can be represented by multiple examples of “Aha moments” that Goldstein (2019) mentioned in the text where humans gained new insights. Humans achieve insights and create original solutions using multiple aspects of cognition. An experiment done by Martinsen and Furnham (2016) investigated the relationship between information processing, achievement motivation, and cognitive style theories. It found that the performance of insight problems is determined by integrating various amounts of these aspects depending on the task. As discussed, information processing depends on information organization and the encoding processes. As a result, individuals utilize different backgrounds and past experiences to create insights and problem-solving strategies. The achievement motivation that Martinsen and Furnham (2016) mention relates to Newell and Simon’s information processing approach. As Goldstein states (2019), diving the task into smaller goals and looking ahead can increase problem-solving.
On the other hand, artificial intelligence has become popular in recent years and used by organizations to form insights into humans’ behaviors and reasoning behind human decisions. Unlike insights, these analyses are often deliberate and conscious. Although humans can use systematic approaches and insights to solve problems, machines can only use a systematic approach to conclude by analyzing large data sets and discovering trends as defined by the programs. These analyses are often deliberate and conscious, allowing humans to store relevant steps in the working memory. It is easy to track the “thought process” and explain them to others, just as computers present the analytics using big data. Humans can see how programs arrive at that decision. However, as Carpenter (2020) states, insights are unconscious and only generated after some incubation time that humans cannot explain to themselves what has occurred. This can be observed in the think-aloud experiment done by Kaplan and Simon, where participants suddenly arrived at the answer after realizing the meaning of butter and bread (Goldstein, 2019). When humans cannot explain how they form insights and the steps to arrive at a particular solution and the thought process behind it, it is expected that humans cannot write programs that train machines to do so yet.
Another shortcoming of machine created insights is how machines and humans interpret weakly associated information in the neural network. Weakly associated information suddenly becomes important as humans retrieve them when forming insights. Because insight formation does not follow rigid steps, humans can retrieve all kinds of knowledge from long-term memory, whether used often or not. Individual variables are difficult for computers to imitate. The weights humans put into each piece of semantic information in long-term memory and how humans utilize them is also challenging for computers to emulate. As Carpenter (2020) states, insights result from reconnection between weakly and strongly related information and bring them to consciousness. To computer programs, every dataset is conscious. Therefore they define each node in the neural network by weights learned from prior experiences. Lightly weighted notes, to machines, are less useful and less relevant. Without specific instructions from the human to connect which nodes under what scenarios, computers would have difficulty connecting with weakly associated information. Therefore, based on the definition and characteristics of insights, artificial intelligence cannot accomplish comparable insights. It would be best if humans and machines can cooperate and enhance their specialties to create better problem-solving skills.
References
Carpenter, W. (2020). The Aha! Moment: The Science Behind Creative Insights. Toward Super-Creativity - Improving Creativity in Humans, Machines, and Human - Machine Collaborations, 1–13. https://doi.org/10.5772/intechopen.84973
Goldstein, B. E. (2019). Cognitive Psychology: Connecting Mind, Research, and Everyday Experience (5th ed.). Cengage Learning.
Martinsen, Ø. L., Furnham, A., & Hærem, T. (2016). An integrated perspective on insight. Journal of Experimental Psychology: General, 145(10), 1319–1332. https://doi.org/10.1037/xge0000208