Centaurs vs. Cyborgs

Hi there,

If you're like most people, you're probably equal parts excited and confused by the rise of Generative Al.

On one hand, tools like ChatGPT are capable of mind-blowing feats, writing top-notch strategy pitch decks or mastering the art of persuasion. On the other, they can barely count to 50.

The line between these two realms is tricky and difficult for most users to understand, which can lead to costly mistakes even with humans in the loop. But two species are consistently crushing it: Centaurs and Cyborgs.

Let's find out what they are - and which type you are!

Preorder Now!

Augmented Analytics is now available as an early release and for pre-order at all major bookstores.

Generative AI: Genius and Idiot at the same time

In a recent, very popular study by Harvard Business School and other top-notch organizations, researchers confirmed what we already suspected - GenAI is incredibly good at some areas and really poor at others. A phenomenon the researchers call "the jagged frontier” - which led to the title of the study and the illustration of this phenomena as shown below:

What this picture says is that there's a line of seemingly equally difficult tasks (from a human perspective), and sometimes the task is clearly within the capabilities of modern generative AI systems, and sometimes it's clearly beyond.

The problem is that it's hard to tell which areas those are. An example?

If you ask a modern Large Language model, to give you a list of 10 short sentences, it will very likely perform this task correctly. However, if these short sentences should always end with a particular word, then things go south:

The reason for this behavior lies in the inner workings of how Large Language models "tick" (a.k.a. being fancy word predictors). But for most users (including, very often, AI experts), these limitations are not entirely clear.

The Problem and the Need for Human-AI Collaboration

This unpredictability can lead to problems such as reduced quality of work, especially when users become disengaged or "fall asleep at the wheel," as researchers call it, when using AI. To mitigate these risks, we need a tight integration of human and AI efforts, but an integration that stands the test.

After observing hundreds of AI users in a real-world scenario and comparing their performance, the researchers identified two key patterns of human-AI collaboration that stood out: Centaurs and Cyborgs.

Introduction to Centaurs and Cyborgs

Centaurs and Cyborgs represent two distinct models of human-AI collaboration that differentiate by how they interact with the AI.

Let's dive deeper into each.

Deep Dive: Centaurs

Centaurs, a term coined by chess grandmaster Gary Kasparov and picked up by the HBS researchers, refer to a clear division of labor between human and AI - similar to how a centaur is clearly half human, half horse.

In Centaur workflows, humans strategically assign tasks based on the strengths of each party. This requires a good understanding of which areas are relatively "safe" within the capabilities of the AI.

Here are some examples for Centaur-style tasks:

• Human writes text, AI translates it

• Human drafts text, AI edits it

• Human provides text, AI summarizes

• AI suggests framework, Human applies it

• AI suggest outline, Human fills it with content

• etc.

In any case, in a Centaur-style workflow, in the end it's pretty clear which part was done by the human and which part was done by the AI. There's a clear division of labor.

Once humans "understand" which tasks the AI is good at, they can hand off tasks (such as translating a text) to the AI relatively easily and safely, and be fairly confident that the results will be correct.

So the typical interaction between AI and human is pretty "short". I give some input, and I get some output that I'm going to keep working with.

For example, if I gave the above text to the "world's best AI-powered translator" DeepL and had it translate it into German, I would be pretty confident that the out-of-the-box result is pretty correct because I know that translation is generally within the frontier of what the AI can do pretty reliably:

Deep Dive: Cyborgs

Cyborgs, on the other hand, involve a deep, continuous, and typically longer back-and-forth interaction the between human and the AI.

In retrospect, it's often difficult, if not impossible, to tell which part was done by the AI and which part was done by the human.

A good example of this is the workflow I recently described in my article "3 High-Quality Prompts To Use ChatGPT For Better Analytics".

In particular, one prompt helps the user write a SMART problem statement step-by-step by asking specific questions and making recommendations. Who wrote the final problem statement? The AI or the human? It's hard to say.

Rather than a clear handoff of tasks, Cyborgs intertwine their efforts with AI, moving back and forth over what the researchers call the "jagged frontier" of AI capabilities.

Examples of Cyborg-typical workflows include:

• Creative writing

• Sentence completion

• Character emulation and chat

• etc.

Personally, I’m a cyborg in most use cases. I go back and forth, interrogate the AI and let it help me create better outcomes in less time.

For example, I gave this newsletter to ChatGPT and asked it to come up with more workflow examples for Centaurs and Cyborgs. When it generated the list, I followed up to emulate a typical newsletter reader of mine to see if the examples would make sense to them.

(In the end, I didn't include the additional suggestions verbatim because - well, you got them in the screenshot.)

Which type are you?

Are you a Centaur or a Cyborg? The good news is you don’t have to choose. In fact, a single high-level business process often involves both Centaur and Cyborg-preferred approaches, depending on the task at hand.

For example take a look at this hypothetical candidate screening workflow as part of an HR process as we would map it out in an AI Design Sprint™:

Depending on the step of this workflow to focus on, we would rather choose Centaur or Cyborg tactics to augment and improve the steps.

In step 2, for example, the HR employee skim-reads through the short-listed candidates and decides which ones to focus on. Here it would be helpful to use a Centaur-like workflow where the employee reads summaries generated by AI instead of doing the skim-reading themselves.

On the other hand, in step 6, where the hiring manager schedules and prepares for the video interview, they could use a Cyborg-style workflow to help structure the interview in advance by drafting relevant questions, identifying important areas for follow-up, and overall tailoring the interview process to the candidate.

Conclusion

Centaurs and Cyborgs are both proven user patterns for navigating the tricky frontier of AI capabilities. While Centaurs carve out clear swim lanes, Cyborgs thrive on the continuous interplay of human and machine intelligence, using judgment and creativity to steer AI in productive directions.

Ultimately, most business workflows require a mix of both concepts, depending on the sub-task and the individual.

Being aware of Centaur- and Cyborg-style workflows and using them strategically when needed is your biggest advantage. And don't be afraid to experiment with different modes of AI augmentation.

The key is to gain enough experience with AI to start seeing the contours of its capabilities. Develop a sense for the “jagged frontier”.

So saddle up your Centaur and charge up your Cyborg.

It's time to explore the new frontier of work in the age of AI.

Because:

No AI? – No chance!

In that sense, see you next Friday!

Tobias

P.S. If you found this newsletter useful, why not leave a testimonial here? It would mean so much to me and make my day - if not my whole weekend! Thank you so much!