Generative People Transformers
Generative AI tools like ChatGPT are revolutionizing work - but threaten to widen talent gaps. Here are some thoughts on how individuals and organizations can adapt and take advantage.
The Third Information Revolution
In a recent podcast with Microsoft CTO Kevin Scott, Bill Gates recalled a demo of a word processer presented by Charles Simonyi at Xerox in 1979. During the demo he and Charles brainstormed a list of capabilities that would become the Microsoft Office suite – which of course has since completely transformed how work gets done. These advancements in personal computing were the first major computing revolution, with smartphones and mobile being the second.
On March 14 OpenAI announced their updated GPT-4 model, which boasts a variety of impressive upgrades from the previous version. GPT-4 can now pass the SAT (94th percentile), the LSAT (88th percentile), and the Bar Exam (90th percentile) with flying colors.
Two other announcements came out on the same day but flew under the radar a bit: Microsoft and Google each announced a suite of new features that directly embed generative AI within Office and Workspace. This will enable AI assistance on tasks ranging from writing, creating presentations, and even data analysis. I highly recommend checking out Microsoft’s demo to get a glimpse. Now, exactly how these tools will change the way work gets done remains unclear – but they are undeniably shifting the paradigm. We aren’t just humans augmented with software, now we are humans augmented with AI. The third computing revolution has arrived.
Will generative AI tools like GPTs (language) and Midjourney (images) automate your job? Probably not. Doomsayers who fear an AI takeover are likely to be wrong (or perhaps just early), but it is clear that most workers will experience an effect. A recent study from researchers at Open AI and Penn estimated that 80% of workers in the U.S. could have at least 10% of their tasks impacted by tools like ChatGPT. Knowing how automation happens in practice, I think this is a plausible estimate. As John Boudreau and Ravin Jesuthasan outline so eloquently in their book Work Without Jobs, work transforms on the margins and at the level of tasks, not entire jobs.
Despite the potential for such an enormous shift in how work happens, I don’t think the HR and people analytics space has been able to offer solid guidance on these questions (yet, and with notable exceptions, check out Josh Bersin’s recent podcast). In the rest of this article, I’ll share some thoughts on how individuals and organizations might pivot, which I hope can move the ball forward in some small way.
A Paradigm Shift in Skills
I’ve always appreciated the concept of the skill gap which comes from an unlikely source: competitive gaming. In this context the skill gap refers to the difference in overall performance between people with novice and advanced skills. This is a useful analogy for the talent impacts of generative AI, because the skill gap has seemingly widened overnight.
To illustrate, when comparing resumes in the past, few would perceive MS Office skills to be a talent differentiator. The variance in work performance between people with introductory and advanced Office skills was real, but unexceptional. Historically we could say Office has had a low skill floor (the applications are easy to use), but also a low skills ceiling (advanced PowerPoint users aren’t necessarily getting headhunted). Going forward, it stands to reason that MS Copilot could widen the skill gap. The difference between an advanced MS Office user and a beginner could now mean a 10x difference in productivity, a paradigm shift in basic skills for knowledge work.
Maybe there was a skill gap in MS Office all along
We can slice the anatomy of this paradigm shift into two categories: 1) the introduction of new skills, and 2) changes in how existing skills are valued.
1) New skills for generative AI:
Obviously the dust hasn’t settled yet, but here are a few skills that are already being talked about in this space, coming soon to a job description near you:
Prompt Engineering: The ability to design effective prompts that elicit the desired outputs from generative AI systems, such as specifying the topic, tone, style, length, format, etc. of the desired output.
Output Editing: The ability to evaluate and edit the outputs of generative AI systems, such as checking for accuracy, relevance, coherence, originality, ethics, etc. of the content.
Output Synthesis: The ability to integrate and synthesize the outputs of generative AI systems with other sources of information or content, such as adding context and human insights to the content or combining outputs from multiple generative AI systems.
Generative AI Literacy: The ability to interact with and understand generative AI systems at a conceptual level, such as how they work, what they can and cannot do, and how to evaluate and leverage their outputs.
2) Changes in how current skills are valued:
The researchers who conducted the study on the workforce impacts of ChatGPT also conducted an analysis on which skills were associated with occupations that had exposure to GPTs. This is kind of an indirect way of testing the question, but it might offer a starting point for speculating on which skills are likely to become more or less valuable in the new paradigm:
Critical Thinking (+): This is intuitive because while GPTs are powerful, they do not have access to all of the relevant real-life context and cannot actually reason through real problems (at least, by the technical definition of “reasoning”). Making decisions on what to do with a vast-er array of slide decks, reports, code, and art will could be a talent differentiator in the future. Managers, who are accustomed to doing this with humans already could have an advantage.
Scientific Ability (+): Similar to critical thinking, scientific ability requires a lot of specific training and context, but also involves steps of hypothesis generation, study design, and analysis of real-life data, which GPTs cannot perform. Sam Altman, the CEO of OpenAI recently said on a podcast that he would not consider AI to be “superintelligent” until it could considerably advance humanity’s scientific knowledge on its own.
Learning Strategies (+): This skill refers to “Selecting and using training / instructional methods and procedures appropriate for the situation when learning or teaching new things.” I’m guessing this has to do more with content curation in learning & development / educational roles, but in a more general way, learning strategies might also become crucial for individuals to adapt to the new paradigm.
Writing (-): This skill has been essential for so long and I’ve often argued that it’s severely underrated. That said, this is unsurprising – generating first-draft content of any kind is now infinitely easier with generative AI, so writing skills might matter less going forward. On the other hand, prompt engineering is still dependent on writing, so maybe this skill will be looked at differently, not less often.
Programming (-): This is similar to writing as GPT can generate code with impressive speed and quality. However, to be skilled at output-editing code, there is still quite a bit of programming knowledge required. I would expect engineers and data scientists to spend a lot more time editing and reconfiguring code to optimize rather than create from scratch.
Making decisions on what to do with a vast-er array of slide decks, reports, code, and art will could be a talent differentiator in the future.
Just for fun, I’ll throw in a couple of competencies that I think will be re-valued, although these are highly speculative and should be treated as mere guesses:
Openness to Experience (+): This is the big-5 trait that I’m thinking will be more important in the short-term, for the simple reason that high-openness people are more likely to experiment with and adopt generative AI in their workflows.
Execution (+): It’s long been an adage in the startup world that good ideas are not scarce, it’s all about the execution. Although brainstorming ideas and generating content are easier than ever with generative AI, humans will continue to address “last mile problems”. The key differentiator for innovation will be synthesizing outputs and executing.
Workers who have the right skills, but especially those who acquire the new generative AI skills, could enjoy a slew of advantages. The most obvious would be higher productivity from time-saved on many tasks. Skeptics may point out that outputs generated by AI are not high quality, but a recent study from MIT found that workers who used ChatGPT assistance not only saved time, but generated higher quality outputs as well.
Still, it shouldn’t be solely up to individual workers to figure out how to adapt to technology shocks. For organizations that successfully encourage their workforce to adopt generative AI, it’s not unreasonable to think that task-level benefits will roll up to improved bottom-line outcomes too. As with all transformations, HR plays a vital role shepherding change, re-designing jobs, selecting and developing the right people, and reinforcing the right behaviors.
How Individuals Can Pivot
Leaning into change: Given the speed at which generative AI is improving and being implemented in new ways, getting comfortable being uncomfortable is more important than ever. Being willing to just dive in and try things out will be a key differentiator of talent.
Spend time intentionally practicing the new skills: It might sound silly, but there really is a learning curve to skills like prompt engineering and output editing, and repetition can help you become skilled in getting what you want out of generative AI tools. Economics blogger Tyler Cowen recently said he spends about 2 hours a day practicing with GPT-4! Reading tips and guidelines from people at the cutting edge can also give you some knowledge so that you spend less time in trial-and-error mode.
Make using generative AI habit: In The Power of Habit, Charles Duhigg consolidates a wealth of psychological literature into a simple 3-step model of habits - 1) The Cue, 2) The Routine, and 3) The Reward. Old habits can be hard to change, so identifying cues like “my manager asked me to create a presentation” can lead you to develop a generative-AI routine that eventually becomes habit. Don’t forget the reward- it’s key to recognize success after implementing the new behavior, even if it’s a self-pat on the back.
How HR Can Pivot:
Guiding Change: HR teams should create and communicate a stance on generative AI that encourages workers to experiment and innovate, but also to use new solutions responsibly. Ultimately, organizations that are able to consciously balance new opportunities and use cases with concerns about the ethical implications of generative AI systems will be best positioned to make and sustain productivity gains.
Work Redesign: Most jobs will not be automated away, but most jobs will have some pieces automated. This will create new capacity and demand for different skills, so working with business leaders to formalize this is a quick win for HR. Understanding for which jobs using generative AI should be an expectation, not just an extracurricular activity going forward will set the stage for HR support.
Selection and training: Once the foundation is laid for generative AI as a legitimate tool at work, a wave of new selection assessments and trainings need to be created to support the selection of workers with updated skills and/or the training and development of current workforce on new skills. Simply asking behavioral or situational interview questions to assess generative AI literacy may be a solution in the near-term.
Re-examine your tech stack: It might sound cumbersome to reevaluate your HR systems (“we just finished our implementation!”), but a variety of established vendors and new entrants are starting to use the API and plug-in tools offered by OpenAI to expand on the capabilities of the HRIS and ATS.
Resources
It shouldn’t surprise you that this article was written with the help of AI. If you haven’t gotten started on your generative AI journey, here are some places to start: