The following is an excerpt from RE-HUMANIZE: How to Build Human-Centric Organizations in the Age of Algorithms by Phanish Puranam.
Engineers talk about the “design period” of a project. This is the time over which the formulated design for a project must be effective. The design period for the ideas in this book is not measured in months or years but lasts as long as we continue to have bionic organizations (or conversely, till we get to zero-human organizing). But given the rapid pace of developments in AI, you might well ask, why is it reasonable to assume the bionic age of organizations will last long enough to be even worth planning for? In the longer term, will humans have any advantages left (over AI) that will make it necessary for organizations to still include them?
To answer these questions, I need to ask you one of my own. Do you think the human mind does anything more than information processing? In other words, do you believe that what our brains do is more than just extremely sophisticated manipulation of data and information? If you answer ‘Yes’, you probably see the difference between AI and humans as a chasm—one which can never be bridged, and which implies our design period is quite long.
As it happens, my own answer to my question is ‘No’. In the longer term, I simply don’t feel confident that we can rule out technologies that can replicate and surpass everything humans currently do. If it’s all information processing, there is no reason to believe that it is physically impossible to create better information processing systems than what natural selection has made out of us. However, I do believe our design period for bionic organizing is still at least decades long, if not more. This is because time is on the side of homo sapiens. I mean both individual lifetimes, as well as the evolutionary time that has brought our species to where it is.
Over our individual lifetimes, the volume of data each one of us is exposed to in the form of sound, sight, taste, touch, and smell—and only much later, text—is so large that even the largest large language model looks like a toy in comparison. As computer scientist Yann LeCun, who led AI at Meta, recently observed, human babies absorb about fifty times more visual data alone by the time they are four years old than the text data that went into training an LLM like GPT3.5. A human would take multiple lifetimes to read all that text data, so that is clearly not where our intelligence (primarily) comes from. Further, it is also likely that the sequence in which one receives and processes this enormous quantity of data matters, not just being able to receive a single one-time data dump, even if that were possible (currently it is not).
This comparison of data access advantages that humans have over machines implicitly assumes the quality of processing architecture is comparable between humans and machines.
But even that is not true. In evolutionary time, we have existed as a distinct species for at least 200,000 years. I estimate that gives us more than 100 billion distinct individuals. Every child born into this world comes with slightly different neuronal wiring and over the course of its life will acquire very different data. Natural selection operates on these variations and selects for fitness. This is what human engineers are competing against when they conduct experiments on different model architectures to find the kind of improvements that natural selection has found through blind variation, selection, and retention. Ingenious as engineers are, at this point, natural selection has a large ‘head’ start (if you will pardon the pun).
How Artificial Intelligence is Shaping the Future of the Workplace
This is manifested in the far wider set of functionalities that our minds display compared to even the most cutting-edge AI today (we are after all the original—and natural—general intelligences!). We not only remember and reason, we also do so in ways that involve affect, empathy, abstraction, logic, and analogy. These capabilities are all, at best, nascent in AI technologies today. It’s not surprising that these are the very capabilities in humans that are forecast to be in high demand soon.
Our advantage is also manifest in the energy efficiency of our brains. By the age of twenty-five, I estimate that our brain consumes about 2,500 kWh; GPT3 is believed to have used about 1 million kWh for training. AI engineers have a long way to go to optimize energy consumption in training and deployment of their models before they can begin to approach human efficiency levels. Even if machines surpass human capabilities through extraordinary increases in data and processing power (and the magic of quantum computing, as some enthusiasts argue), it may not be economical to deploy them for a long time yet. In Re-Humanize, I give more reasons why humans can be useful in bionic organizations, even if they underperform algorithms, as long as they are different from algorithms in what they know. That diversity seems secure because of the unique data we possess, as I argued above.
Note that I have not felt the need to invoke the most important reason I can think of for continued human involvement in organizations: we might just like it that way since we are a group-living species. Researchers studying guaranteed basic income schemes are finding that people want to belong to and work in organizations even if they do not need the money. Rather, I am saying that purely goal-centric reasons alone are sufficient for us to expect a bionic (near) future.
That said, none of this is a case for complacency about either employment opportunities for humans (a problem for policymakers), or the working conditions of humans in organizations (which is what I focus on). We do not need AI technologies to match or exceed human capabilities for them to play a significant role in our organizational life, for worse and for better. We already live in bionic organizations and the way we develop them further can either create a larger and widening gap between goal and human centricity or help bridge that gap. Technologies for monitoring, control, hyper-specialization, and the atomization of work do not need to be as intelligent as us to make our lives miserable. Only their deployers—other humans—do.
We are already beginning to see serious questions raised about the organizational contexts that digital technologies create in bionic organizations. For instance, what does it mean for our performance to be constantly measured and even predicted? For our behaviour to be directed, shaped, and nudged by algorithms, with or without our awareness? What does it mean to work alongside an AI that is basically opaque to you about its inner workings? That can see complex patterns in data that you cannot? That can learn from you far more rapidly than you can learn from it? That is controlled by your employer in a way that no co-worker can be?
Excerpted from RE-HUMANIZE: How to Build Human-Centric Organizations in the Age of Algorithms by Phanish Puranam. Copyright 2025 Penguin Business. All rights reserved.
