I know it is striking that I am announcing the death of a profession that appeared so recently, but from the standpoint of human-artificial intelligence interaction, that is precisely the goal.
I have been emphasizing this point practically since the term was coined. Perhaps this is due to my aversion to using the word “engineering” too loosely. In any case, right now it is a statement that can be understood by anyone who has interacted even minimally with an AI.
What is prompt engineering?
- DefinitionIn plain terms, it consists of engineering the instructions given to an artificial intelligence to obtain the best possible results.
- The scope of resultsOf course, “better results” can involve many different aspects. There are plenty of examples of proud chatbot users announcing they have managed to “trick” an artificial intelligence into revealing data it shouldn’t have provided in the first place.
- The core essenceNumerous prompt engineering techniques exist, but to sum it up briefly, the main objective is to place the artificial intelligence into the most precise context possible so it does exactly what you want. Hence formulas such as asking a chatbot to adopt the role of an expert in the specific field you are inquiring about.
Usability, User Experience, and Prompt Engineering
Different definitions for usability exist, but the clearest one to me is provided by ISO/IEC 9241:
Usability is the effectiveness, efficiency, and satisfaction with which a product allows specific users to achieve specific goals in a specific context of use.
- Evolution of interactionFrom this perspective, our interaction with artificial intelligence should evolve to a point where it is capable of understanding what we want to do without us having to complicate things by thinking about how to phrase it.
- The synchronization goalDo you know those people who are so perfectly synchronized that when faced with a vague request like “Hand me that thingy over there,” they immediately grab the drill with an 8mm bit and pass it to their partner without hesitation? Well, that is the level of trust we should reach with artificial intelligence. That is the direction our interaction with artificial intelligence ought to take.
- Context detectionTechnical advancement should allow us to obtain the best possible answers, even when we provide an incomplete context or specify it poorly. Artificial intelligence may eventually become capable of detecting aspects like a sense of urgency or other circumstances that might extend beyond normal human perception.
- Current realityThis might seem highly futuristic, but when there is a clear interest or need, humans have already proven that we can advance technology rapidly. Going beyond human perception is surely something many of us have felt when we see shopping suggestions for certain products appear on our phones. It is truly something that is already happening.
- Environmental and psychological factorsOn the other hand, we face the need to minimize energy impact and environmental damage. Should we place this burden on the user, forcing them to phrase queries in a specific way?What about use cases where a chatbot needs to be empathetic, such as assisting individuals with depression? Are we not going to be allowed to get angry with the artificial intelligence when it makes a mistake, for fear of causing irreparable damage to the environment?
- Human communication preferencesI highly doubt humans will choose to institutionalize speaking to machines in an inhuman manner, unless we are somehow forced to do so. Perhaps like when SMS cost a fortune and we had to communicate efficiently, much like with telegrams.
- The uncanny valleySimilarly, I generally do not think we would prefer artificial intelligences to address us without expressing humanity. Furthermore, in many cases, we will want them to do so in a personalized manner. If AIs dehumanize the conversation, we would surely enter the uncanny valley, a phenomenon I hope to discuss at another time.
Interaction with Artificial Intelligence Must Flow
Those familiar with the famous Dune saga should recognize the line “the spice must flow.” The universe presented in this saga is shaped by events that occurred due to the emergence of artificial intelligence, so I take this opportunity to reuse the expression and recommend another literary work that, back in 1965, already reflected on the consequences of AI for humanity.
If we want our interaction with AI to be fluid, it makes no sense for the discipline of prompt engineering to remain with us for much longer. The challenge for interaction experts is to ensure that robots and any system using artificial intelligence reach this point, achieving natural, effective, and efficient communication in every sense.