Lowering the temperature: when not to use AI jargon
Temperature, RAG, context window. Chances are you have encountered at least one of these words in a conversation about artificial intelligence. Jargon is normal and necessary. It forms through the simple process of people adopting new expressions with a particular meaning to avoid awkward acronyms or unreadable sentences. The more complex the idea you need to communicate, the more you depend on jargon. Yet at some point you most likely felt the sting of not being able to contribute to a conversation because of the words others were using. What was supposed to make the conversation more efficient felt much less so for you personally.
Jargon immediately forms an in-group and an out-group. Whether you understand exactly what RAG describes becomes a question of whether you are proficient in this new technology that you might already be late to. The language that was intended to facilitate rapid scaling of AI systems becomes a barrier to the very people that adoption depends on. The solution is not a simplification of AI-specific language. The mistake would be insisting on the use of a single vocabulary and register when it is not only unnecessary but actively counterproductive. By looking at different cross-sections of an organization or society at large, we can fine-tune our vocabulary for the appropriate context.
Most people: the machine feels nothing
Most people need to understand how communicating with Large Language Models (LLMs) differs from communicating with a fellow human. For agentic AI, they need to recognize how programmed workflows might behave differently from a person. The key message should be that these models feel nothing. More importantly, they do not feel shame. And we depend on shame more than we think.
Consider a colleague saying they have completed a report. Why do you believe them? If they then promise you that they won’t just copy last year’s numbers, why do you trust them? It has to do with the shame with which society punishes liars who are caught. You implicitly extend trust to people around you because you would feel humiliated if you were caught not having met a deadline or having used made-up numbers.
None of this reaches an AI system. AI systems present falsehoods as confidently as truths, because a sentence does not need to be true for it to be statistically likely in a given context.
To properly explain this, we should not reach for the jargon that describes the specific nuanced behaviour of AI systems. Instead we should validate the similarity between communicating with these systems and day-to-day speech. In fact, we should encourage the use of normal human language. But we should be clear and explicit on the differences. A comparison between behaviours should prevent users from carrying human assumptions into interactions with machines.
Leadership: cutting decision latency
The aim of explaining AI concepts to leadership, on the other hand, should be to decrease the decision latency, the time between a decision becoming necessary and the actual decision being made. This requires an understanding of the trade-offs that these systems introduce to the decision-making process. For that, a clear view of what these systems can and cannot be trusted with matters far more than knowing what to call specific mechanisms.
Practitioners: where jargon earns its place
The people who need to configure, control, and maintain AI systems. This is where jargon is necessary. We should not confuse the need for jargon with inventing vocabulary for its own sake. The standard ISO/IEC 22989:2022 is actively being updated with an amendment to standardize terms like “hallucination” and “prompt engineering”. By contributing to the creation of such standards and by consistently applying them, we can harness the full power of AI systems. Or at least not hamper them with inefficient communication among humans.
The contradiction we have to name
The biggest advantage of LLMs over previous AI systems is the ability to communicate in natural language. It is easy, therefore, to present a contradictory message: use your own words but avoid communicating with them like you would with a human. We need to make this contradiction clear to everybody involved. But that can only be done if we ourselves use natural language and are clear about what role AI systems play in the work of different groups in an organization as well as the broader society.
Further reading
If you only read one:
Brown, Z. C., Anicich, E. M., & Galinsky, A. D. (2021, March). Does Your Office Have a Jargon Problem? Harvard Business Review.
Deeper research:
Shulman, H. C., Dixon, G. N., Bullock, O. M., & Colón Amill, D. (2020). The effects of jargon on processing fluency, self-perceptions, and scientific engagement. Journal of Language and Social Psychology, 39(5–6), 579–597.
Brown, Z. C., Anicich, E. M., & Galinsky, A. D. (2020). Compensatory conspicuous communication: Low status increases jargon use. Organizational Behavior and Human Decision Processes, 161, 274–290.