The Age of Cognition
The Problem with Gen AI Disruption
Background
Last week, I had the pleasure of being invited to a HR conference panel to share my thoughts around the ethical considerations for Generative Artificial Intelligence (Gen AI for short). After the usual soundbites on AI ethics, the conversation quickly converged on whether AI, and Gen AI in particular, would replace or augment the workforce. And if so, in what ways. And so I dedicate my 13th article to the kind of disruption that I foresee Gen AI would usher in for the workforce.
(I write a weekly article on bad thinking and bad practices in data analytics / data science which you can find here.)
The Fear is Real
In the world of AI today, it may be useful to separate the solutions into ‘traditional’ AI and Gen AI. In traditional AI, the use cases are related to the reduction of uncertainties in predictions and decisioning. For example, the AI in your smartphone battery management application attempts to estimate how much to throttle your applications to last out until you next charge your handphone (prediction) based on how much you will use those applications (prediction). Overall, the battery management AI optimises the charging and discharging process, leading to significant improvements in battery life and shelf-life. These kinds of prediction and optimisation are easily welcomed by humans and organisation workforces because we immediately recognise that these are not our natural competencies. We welcome them because we can see how they will make our lives better.
In Gen AI, the use cases (at present) are all related to information foraging, information summarisation and information synthesise; search, summarise, synthesise for short. These activities are inputs that form the base foundation of business decisioning. Humans instinctively understand that these activities are indeed our natural competencies, and hence the clarion call on limiting and censuring AI (collectively) through legislation, many of which are ill-defined and frankly, intellectually brutish.
Employees are NOT Assets
Let’s get real. Organisations don’t think of employees as assets; they are represented as cost in the balance sheet. Only a small percentage of employees would qualify as assets, and even then, they are not necessarily irreplaceable. A good asset is something that either holds or appreciates in value over time, and employees do neither without further investments. Businesses, in partnership with HR, have been doing everything in their power to reduce employee ratios since the industrial revolution. With this present knowledge revolution, it is not going to be any different.
Many in HR believe (or want to believe) that AI will augment rather than replace the workforce; that employees will instead lose their jobs to AI-savvy employees instead of to machines. The job loss is already upon us. Early economic research is indicating that Gen AI is leading to job losses in knowledge work; copywriters, graphic designers and creative agencies have all taken a hit. It’s also unclear what AI-savviness is, and therefore how to increase it. Is AI-savviness about tool expertise or something more — e.g. knowing how to fact-check the outputs from AI?
We’ve spent the last 20+ years advocating for data and digital literacy; we’ve spent massive amounts of money on up-skilling and re-skilling the workforce. And yet, the level of data and digital literacy in the working population is still dismal. By some estimates, less than 25% of the workforce is data literate. And considering that over 90% of jobs require digital literacy skills, only 66% are digital literate.
A tsunami is coming. The integration of traditional AI with Gen AI will accelerate job losses. If the workforce is not prepared to put in the hard work to be data and digital literate, then there will be a demand for AI to work harder to fill that space of human-driven work activity, leading to even more job losses. Ultimately, it’s a case of demand and supply economics.
Nature of Cognitive Work
Knowledge work requires both cognitive and technical abilities. Gen AI is reducing the need for the latter; for example you don’t have to be an expert in drawing to create sufficiently good visuals for corporate use with tools like Midjourney and Dalle. But you do need to have a good appreciation of art to know that your visual is sufficiently good. From where will you learn to appreciate art?
This takes me to the heart of the matter regarding the disruption that Gen AI brings to cognitive work. There are gradients of cognitive work, but you can simplify it into 2 broad categories — high and low. Administrative work, searching and researching information would be considered as low cognitive work; they follow a set of procedures and can be ‘templatised’. Even in academia, writing a paper requires massive amounts of time to conduct literature review (to figure out who is saying what in your proposed research domain), and that would be low cognitive work. I’m not saying that low cognitive work is easy and doesn’t require effort, but it follows a known procedural pattern. This is exactly what Gen AI has learnt to do, and is only getting better and better at it with each version iteration.
High cognitive work, on the other hand, is distinguished by a lack of procedure. While it uses historic data to inform, it is trying to imagine a new situation by considering what assumptions need to be true, and then figuring if one has the power to make those assumptions valid (in the near future) or it will be valid by the logical forces of economic nature. High cognitive work is hard to ‘templatised’ and requires both deep domain and broad cross-domain knowledge. Strategic thinking and problem framing would qualify as high cognitive work. This is the space that Gen AI has yet to crack because it is knowledge-centric and not data-centric, and the current brute-force scaling on large language models may not be the right mechanism to get there.
HR argues that traditional AI + Gen AI will free us from ‘mundane’ work and give us time to focus on activities that generate higher economic value. The massively flawed assumption here is that the workforce is able to do higher value work. I really don’t think so. There is no evidence to support this assumption. My earlier point on the lack of data and digital literacy sadly only supports the opposite conclusion.
Raising the Bar
In an AI-productive world, critical thinking is going to be a precious commodity. Critical thinking is really just a catch-all category for a bunch of higher cognitive abilities like sensemaking, design thinking, computational thinking. Activating and cultivating critical thinking requires exposure to diversity and divergent thinking, both of which are in short supply due to the narrow and converging nature of our education systems. The love of reading (whether consumed physically, digitally or audibly) has been significantly declining year over year, not just in adults, but more worryingly, in children and teenagers (formative years of learning!). We are circling the drain hole.
Instead of promulgating for a AI-augmented workforce, those who are likely to thrive are going to be part of the human-augmented AI workforce. Those employees who can augment the output of AI-generated work will be most successful. Those who already have or are investing in improving their critical thinking competencies and higher cognitive abilities will have a leg up. We are already seeing that in the traditional AI space where data scientists who lack data sensemaking abilities (which many do!) don’t get very far, even though they may be uber-coders. The paradigm of AI-augmented work is about increasing efficiency, but the paradigm of human-augmented work is about increasing effectiveness.
Conclusion
The currency of the future will be high cognitive competencies such as critical thinking. This has a negative reinforcement feedback loop — the bar will be raised for entry work in an AI-dominated workforce. That intern whose job was to do some basic market research? It will be gone. That fresh graduate whose job was to learn to be a practical data scientist with real-life coding work? It too will be gone. High cognitive abilities are honed through deep and wide exposure, all of which are largely absent from the education learning curriculum, unless it radically transforms NOW. In the short term, Gen AI may enhance your current abilities and even give you new ones, but it also means the bar for the median employee just got a whole lot higher, and to truly distinguish ourselves will mean more than just being AI-savvy.
