The Ethical Conundrum of AI
By Laurie Lumish
ChatGPT took the internet by storm on November 30, 2022. In a matter of hours, tweets, blog posts, and Instagram stories were being posted purportedly written by the Artificial Intelligence (AI) model created by San Francisco-based OpenAI.
Within a few weeks, people began to question how far this program could go. Could it write a great novel, a masterful poem, a proposal worthy of submitting to a client? But what if the question isn’t whether ChatGPT could write those things, but rather should it?
Any A/E/C marketer knows the Request for Proposal (RFP) cliché: “do not include boilerplate.” What is the text created by AI models other than boilerplate, and potentially inaccurate boilerplate at that?
The ethical limitations and challenges of AI-enabled systems in A/E/C marketing include inaccuracy, bias, and lack of contextual understanding. Additional ethical questions about AI systems in industries beyond A/E/C involve concerns around data privacy and misuse.
Inaccuracy and Bias are Inherent to Any Aggregator
AI models are built on large data sets. The information that the models provide is an aggregation of material that they find in their data set. For example:
Founded in 1936, Degenkolb Engineers has a rich history of providing innovative engineering services to some of the world’s most iconic structures.
This is a sentence from a ChatGPT-authored firm profile that I generated on the system. With no training or other resources to investigate, one might say that is a fair opening to any firm profile. However, a little bit of fact-checking will tell you that:
1. Degenkolb Engineers was not founded in 1936.
2. The firm is not generally known for “the world’s most iconic structures,” being a firmprimarily located in the Western United States and Mexico.
Trying to “work smarter, not harder,” a marketer presenting this to a firm Principal would immediately receive criticism of the marketer’s competency—or, worse yet, the material could find its way into a proposal or qualifications document undetected.
The AI model can only relay information found and learned from its data set. In the case of conversational AI models, the data sets are built off information found from a wide range of sources with little to no ability to account for accuracy.
The limitations of the data set also prevent the AI model from identifying or self-correcting implicit bias. Whatever information is provided as the learning tool will carry bias, so unless the data sets are completely balanced in representational data, bias will be implicit in the results. When prompted about its own bias, ChatGPT provided the following response:
As an artificial intelligence language model, I do not have personal beliefs, opinions, or biases. My responses are generated based on the input provided to me and the patterns I have learned from the vast amount of text I have been trained on. However, it's important to note that the quality and accuracy of my responses may depend on the quality and diversity of the data I was trained on. Additionally, my responses may reflect the biases present in the language used in the training data. However, efforts have been made to mitigate these biases during my training to ensure that I am as unbiased and impartial as possible.
Context Matters for A/E/C Marketers, and AI Models Have Not Mastered This Skill
By its own admission, ChatGPT “does not have a deep understanding of the context or meaning of the text it generates. This can result in responses that are not appropriate or relevant to the user’s query.”
For A/E/C marketers, the contextual understanding of a potential project comes from many sources; for instance, the project’s political, social, and geographic settings. Frequently, this information is not written in an RFP/RFQ document and may not be readily available on the internet.
Writing a well-thought-out project approach or value proposition requires understanding the spoken, written, and unwritten clues of the project’s various environments. Current AI models, even very strong conversational ones, cannot achieve this level of contextual understanding in the A/E/C environment.
The Role of AI-Enabled Systems for Content Creators in A/E/C Marketing is Murky at Best
There is not yet an AI system that can write an accurate, fully customized, and contextually rich RFP response. Eventually, more sophisticated AI models may create competitive divides between large, small, and mid-sized firms.
When a large A/E/C firm develops its own conversational AI based on a strong data set of successful proposals for specific project types, then we may start to see the value of AI in our work. But when that happens, will our clients be wondering: “Is this AI-generated boilerplate, or did an actual human consider our project?”
Conclusion
Given the current questions about the future of AI, there may be few marketers who are on the leading edge of utilizing this technology for A/E/C. Even with all the caution to be had, marketers who identify innovative ways to overcome these barriers will find themselves at the forefront of our profession.
Laurie Lumish, CPSM
Director of Marketing & Business Development, Degenkolb Engineers
[email protected]
President, SMPS SF Bay Area Chapter
