Want to Increase Productivity With AI? Be a Centaur or a Cyborg (Or Use AI Infrastructure Software)! 

Does AI actually increase productivity? A study by Harvard Business School and the MIT Sloan School of Management says, “Yes, but…”¹  

When used appropriately, AI users finished relatively straightforward tasks 25% faster and saw a 40% improvement in the quality of their results compared to a control group that did not use AI. Although AI-enhanced results were generally of higher quality, they were also less diverse—the AI was more likely than a person to suggest generic or common solutions. For more ambiguous tasks, failure to understand and use AI appropriately led to an 18% decrease in correct responses compared to the non-AI control group. 

Although AI is a powerful tool that dramatically increases productivity, like all powerful tools successful results depend on understanding its strengths and weaknesses. The trick to using AI effectively is to use it in a fashion that maximizes human and AI capabilities, enabling the strengths of one to compensate for the weaknesses of the other. This turns out to be surprisingly difficult. The study found two approaches, although these are hardly the only possibilities: 

• Centaurs strategically split tasks between humans and AI based on the user’s evaluation of whether the tasks better fit the strengths of the person or the AI.  

• Cyborgs, by contrast, worked closely with the AI, integrating it into their workflow to the point where it became unclear whether the output was human or AI generated.  

Both of these approaches allowed users to benefit from AI. The challenge is getting there. For all its vaunted “intelligence,” AI is not as simple to use as the hype makes it seem. The question, therefore, becomes how do we successfully integrate AI into business operations while maximizing the benefits of AI and minimizing the downside?  

Lights Are On, Nobody’s Home 

A major challenge to using AI effectively lies in the nature of the tool. Artificial intelligence could more accurately be called illusory intelligence—AI Large Language Models (LLMs) put on a good show and, just as a skilled magician makes their illusions look like magic, LLMs create the illusion of intelligence. Although LLMs appear to carry on a conversation and do a credible job of seeming intelligent, there is no consciousness or awareness in there: they are really probability machines. Given an input, they calculate the most probable response based on a statistical analysis of everything they’ve been trained on. Thus, a vague or generic input gets a generic response because that generic response is seen as most probably correct. Conversely, a well-constructed prompt gets a more specific (and hopefully better quality!) response because it focuses the LLM. The AI is still making probability-based choices, but by limiting the search space it is choosing amongst more useful options. The results are more helpful and the AI seems very smart. 

It is still an illusion of intelligence. But that illusion can be very convincing.  

Coaching Is A Good Start 

The HBS-MIT study found that users who were coached on the proper use of AI did better—to a point. Because AI does such a good job creating the illusion of intelligence, users become complacent. When you’re deep into solving a complex problem, it’s easy to fall into the trap of treating the AI as if it were a colleague and not a computer. Without getting into the cognitive psychology weeds here, magicians distract us during their performances so that we’ll be so busy watching and listening to the act that we won’t see through the illusion. Similarly, when we’re busy with a complex task, we’re less likely to see through the AI’s illusion of intelligence, and therefore we do not give the AI-generated results the scrutiny they deserve.  

Thus, while teaching people the proper ways to formulate a prompt and other best practices for AI use is a good start, it is not sufficient to obtain consistent improvements in performance. Nor does coaching solve the sameness problem—even Centaurs and Cyborgs produce results that lack variety. 

The Power of Infrastructure 

One powerful approach that sharply increases the benefits of AI is using Infrastructure Software. AI Infrastructure Software is an interactive framework that makes it easier for users to take advantage of AI strengths while being shielded from their weaknesses.  

Effective AI Infrastructure Software provides several key benefits and capabilities: 

• AI Freedom—AI Infrastructure Software lets you choose which AI LLMs you want to use. You can use multiple models and swap models in and out according to your needs. 

• Guided interaction—AI Infrastructure Software works with the user to help develop and refine the prompt, increasing the quality of results. 

• Enhanced Prompt Engineering—AI Infrastructure Software takes the refined prompt and further enhances it (including using RAG). This dramatically reduces the chances of inaccurate or hallucinatory responses.  

• Highly Relevant Results—AI Infrastructure Software conducts relevancy analysis of AI responses, helping to make sure that what you get is appropriate to the problem you are solving. 

And quite a bit more, depending on specific situations and use cases. 

A Council of AI Advisors 

The HBS-MIT study consistently found that AI use produced a “diminished diversity of ideas.” In other words, far too much sameness in results. This is not surprising: AI only presents the illusion of creativity. It cannot think outside the box because it is the box.  

However, because Infrastructure Software can simultaneously support multiple AI LLMs, it can increase the diversity of responses by automatically prompting more than one model at a time. AI Infrastructure Software enables dynamic examination of results and removes duplicates, reducing the cognitive load on users. 

While even the Council of AI Advisors won’t give you truly creative results, it will provide you with more diverse results that can spark human creativity. That’s something that both Cyborgs and Centaurs can benefit from.  

Infrastructure Supports AI Optimism 

The HBS-MIT study presents an optimistic picture of the benefits of using AI to significantly improve performance on important, challenging knowledge work tasks. However, it is also clear that by itself AI only appears to be easy to use, can actually decrease performance if users become overwhelmed or complacent, and tends to produce results lacking in variety. 

Although there are strategies for effective AI use—Centaurs and Cyborgs being the two identified in the study—even they can produce wrong results and suffer from the sameness problem.  

However, Infrastructure Software justifies the optimism around AI. Infrastructure Software provides necessary support for new users of AI while also improving the performance of Centaurs, Cyborgs, and other experienced users. As businesses integrate AI into their day-to-day operations, AI Infrastructure Software is the vital component that turns AI hype into reality. 

To be a leader in using AI Infrastructure Software, or simply to learn more about AI Infrastructure Software, contact SWIRL today!