The productivity promise (and perils) of generative AI


There’s a lot of tech industry excitement these days about tools like ChatGPT and DALL-E that fall into a class called generative AI. Generative AI looks at existing work and generates a result that seems unique,  but is actually derived from what it’s viewed. (I’d argue this is how most people produce work — by taking what they’ve learned from others to create a skill set that can be used independently of those earlier references.)

Like many others, I really enjoy working out the concept of something I want to create, but I get bored quickly when executing that concept. This reminds me of an old friend who bought a knock-off Ferrari kit car. Coming up with the dream of driving it was easy and fun, but actually putting in the thousands of hours to build it was too much. It sat in his garage, unbuilt. With these tools, you can focus on the fun part of creation (coming up with an idea), and then let the tools step in to do the tedious part of bringing everything to fruition. 


This new class of tools has a lot to offer, but there are some initial problems we’ll need to overcome.

The promise of generative AI

We are clearly at the very early stages of this technology; even so, some of the results are amazing. There’s art work that has won contests and music that is very impressive. While the writing isn’t top tier, it is readable and often interesting.

That’s one heck of an initial performance baseline.

The clear promise of this technology is that it can free users from the tedium left by other tools that automatically edit images, check spelling and grammar, and provide advice and help like Siri. But much like searching on the web, you need to develop skills for these tools so you can word a command or query that quickly delivers the result you want. Without that skill, you’ll struggle — either because you haven’t taken the time to fully detail your query or you don’t know how to properly word it.

A big promise of these tools is that they won’t replace you (unless you don’t learn how to properly use them). But if you can master the syntax needed, they’ll significantly improve your productivity and contribution to your company.  

The problem with generative AI

Much as it was when the web became publicly available and we discovered the need for Boolean Logic (and that few had learned it), people haven’t yet developed the skills to use generative AI efficiently. Worse, rather than moving to embrace these tools, educators treat them like cheating, suggesting that even though the skill of being an “AI Whisperer” will be highly marketable, most students will treat the technology as if it is somehow illicit. This seems to happen repeatedly. Instead of realizing a new tech skill is important to students’ long-term careers, schools train the students for a world that will no longer exist after they graduate. 


And, of course, because of valid concerns that these tools could one day eliminate jobs, litigation is starting to crop up — even though they pretty much learn the same way we do, through observation. They simply do so at machine, not human, speeds. Artists, writers, engineers and even litigators largely learn by observing others. If that somehow becomes illegal, we’ll have a serious problem not only training AIs but training people. My thinking is that this behavior should fall under fair use standards, with the main difference between how generative AI “learns” and how a person learns is fewer textbooks and magnitudes more observation at machine speeds. 

Where we go from here

To sum it up: generative AI will be a game changer. A lot of jobs will become obsolete once it reaches critical mass, which often happens when new technology arrives. The sustaining skill will be learning how to properly direct this new class of AI to reduce the number of iterations needed to align what a user wants with what the AI creates.

Learning how to be an “AI Whisperer” will separate those who flourish during this time from those that do not. In addition, the level of productivity gained from these kinds of tools should ensure the future of firms that embrace this technology over those that do not. While there is, indeed, intellectual property litigation about how these tools learn, given that process is similar to how we learn, I don’t believe it will be successful.

That said, it could slow adoption in the meantime. 

Regardless, we are at the beginning of what promises to be a massive technology change. Making sure  we (and our companies) ride that wave instead of being drowned by it should be one of our highest priorities.