Generative AI and Your Job
Open AI’s ChatGPT, as well as the several companies offering or announcing competing solutions these past several weeks (Google Bard, Athropic Claude, Baidu Earnie), has heightened the awareness of the state-of-the-art/science in generative natural language-based artificial intelligence.
Generative AI can receive as input plain text written by humans – known as a “natural language prompt” or just a “prompt” – and respond by generating various types of content: long-form text, short-form text, photorealistic images, artistic illustrations, and completely unique royalty free music. All these forms are written or created with the appearance and an aesthetic formerly believed to only be possible by actual humans. It’s genuinely impressive.
It’s the case with any technology that is well marketed as having “human-like” capabilities that these innovations cause concern, questions, and uncertainty – some of which are intentionally promoted by those who have their own fear-enhanced agenda – that somehow this technology might replace you, your job, or your livelihood. Don’t misunderstand, I think it’s a worthwhile question, but as you’ll see in this article, it’s my belief that we’re more than ten years away from any risk to your current livelihood. By then, your job would have likely evolved to adapt to the new capabilities – so that even then, you’ll still be gainfully employed.
There are three reasons that I believe this is true.
It’s Not as Good as You Might Think
ChatGPT has created such a stir because of the apparent knowledge it seems to possess and the apparent ease with which it generates human-sounding/looking content based on a simple prompt. But, as Steven Wolfram points out, despite the many things ChatGPT can do, there are many activities for which it is inadequate or incorrect. Much of this has to do with the way that the GPT-3 model that powers ChatGPT has been designed and trained. It wasn’t created to solve specific problems or provide certain answers accurately, though, as the article points out, it does an excellent job of sounding convincing. Another ChatGPT realist is Steve Wozniak (apparently, something is going on with people named “Steve”), who warns that ChatGPT can “make horrible mistakes.” These limitations don’t just stop with text-based models. BuzzFeed News has an excellent article that explains the strange symptom of nearly perfect photorealistic generated images of nonexistent people – most of which have messed up fingers. So, while there are some amazing things that ChatGPT can do and some hilarious content generated, don’t believe all of the hype. It’s not ready to take your job, and if it did, it would likely get fired on the first day (ok, it might be able to last until day five 😊).
It’s Not as New as You Might Think
One of the most significant signs that ChatGPT isn’t ready to take your job anytime soon is a brief reflection about what ChatGPT is, or the least what makes it so special. A way to frame this is to remember that Google’s Language Model for Dialogue Applications (LaMDA) – the language mode that powers the newly announced ChatGPT competitor, Bard, in the same way, that GPT-3 powers ChatGPT – was announced during Google I/O 2021. LaMDA started as a project from Google Research in 2017 and has been developing since then. This is the same model that was described as (nearly) sentient by Google Engineer Blake Lemoine in 2021. Powered by LaMDA, Bard was announced recently; it was demoed in early February 2023, and it didn’t go so well. LaMDA has been in development for six years and was touted as being sentient nearly two years ago, causing Google’s parent company, Alphabet, to lose $100 billion in market value in a single day by not being trustworthy enough for a rehearsed demo. As we have already discussed, ChatGPT / GPT-3 isn’t ready for full trust yet either, and, relatedly, GPT-3 has been around since 2018.
The point here is that there is a key divergence between the hype and what will truly impact the job market in the coming years. ChatGPT and other generative AI services are tools at best in the hands of the adaptive knowledge worker, though some would even say that they aren’t that. In fact, as a tool, some are seeing significant productivity gains for certain use cases, but this bump in productivity simply means that humans are being more productive by using technology, not that humans have been somehow replaced. It’s important to appropriately frame ChatGPT or other exciting 2022 generative AI tools like GitHub Copilot as enablement tools that now provide more people access to advanced natural language neural network AI models than ever before. The “fancy tech” innovation – the AI models themselves – have been around for well over six years, but this recent disruption and hype is that now more people than ever before can get access to these models.
We’ve Been Here Before… Sort Of
Because the quality of generative AI is still questionable and these models have been under development for more than just the recent past, the rate of change – both the technologies themselves and the jobs they impact – appears to be much slower than all the recent hype likes to imply. This is why I think we’re at least ten years away from anything that would significantly leverage AI to change the job market or job security. At that rate of change, while there is likely to be some job disruption in the long term, widescale job displacement or job insecurity is highly unlikely.
Looking at the history of innovative changes with similarly disruptive potential, such as email or effective web search, there was a significant adoption and maturity timeframe from the point that these technologies were introduced until they were truly widescale (e.g., 20 years and ten years, respectively). With such a slow proliferation curve, while couriers and research assistants were certainly impacted, most of the world adopted the new technologies and adapted their work.
Some may argue that not only were email and effective web search created during a far slower-moving internet age (1970-2000), but that they were not so indistinguishable from human creativity as ChatGPT, Bard, GitHub Copilot, Mid Journey, and similar generative AI. Therefore, not only will disruption happen more quickly, but when it does, it will happen to a far greater extent: displacing more people’s jobs in favor of AI. Well, I suppose this is possible, and I agree that the world is different now than when email and Google rose to search dominance. But as has been discussed, generative AI is yet to be consistently or dependably high quality. This means that the adoption time clock hasn’t even started yet, and historically, this stage is followed by another phase of development and refinement once adoption begins to accelerate. It’s possible to compare generative AI to earlier technologies (email, search, or the web) too far. Still, the point here is that if you’re worried that the AI overlords will take your job, you should breathe easy. While unsettled right now in certain sectors, your job and the state of the job market are and will remain unaffected by generative AI for the foreseeable future.
If there’s one thing I’d like to leave you with, it’s this: you should not worry about your job or the future of humanity because of ChatGPT or generative AI. While it’s not entirely clear what the adoption rate, maturity/accuracy/release schedule, or the impact of the next great model will be, there is an incredible amount of misinformation, scare tactics, and general unrest around generative AI now. Here are three takeaways or next steps that I would suggest you take to respond to generative AI appropriately:
- Be curious. Read about it from multiple sources. Get educated. Read the news, tech journals, blogs, and other sources, but if what you read causes you to be afraid, then you should likely leave that source behind. For now, there’s nothing to worry about.
- Play with it. ChatGPT is free to play with. MidJourney and DALL-E are free to play with. I would advise that you set up an account and take a few of these tools for a test drive. It will soon become apparent what these services are and are not, but also, as has already been discussed, these are tools that may – even with their relatively low level of maturity – help you be more efficient at your job today.
- Plan for it. While we’re at least 2-3 years away from entirely usable generative AI – when today’s hype becomes tomorrow’s reality – there will be another 5-7 years after that when adoption will progressively impact the workforce and job market. I don’t have a crystal ball, and these are estimates based on my time spent in IT roles for the last 25 years. Still, you should begin to dream and be curious (not fearful or ruminative) about a world in which your uniquely human contributions will be refined to those things which are not supported by generative AI. 10 years is a long time in IT years and my suspicion is that by then not only will AI look different, but also so will the jobs used to support it.
In closing, I’ll leave you with a quote from Forrester analysts Charles Betz and Chris Gardner, “When systems become too automated, their behavior in key respects becomes harder and harder to predict, and setting them straight when they go wrong requires deeper and deeper expertise.”
While the jury is out regarding the extent or way this quote will apply to generative AI, it is mainly beyond debate because history has shown us time and time again that as we make technological advancements in the long term, more humans are put to work, not less.