AI Doom or AI Boon: will Generative AI lead to White Collar collapse?
While “narrow AI” applications fundamentally changed how we live and work by helping people and businesses do at scale what even armies of workers cannot, many in and outside of the AI community have long believed that tasks requiring creativity remained the sole province of humans. We’ve therefore become comfortable in the assumption that white collar professionals were (mostly) safe from the AI jobs apocalypse— “surely, AI will eventually replace some jobs, but not my job.”
As such, for all the wonder and delight that generative AI models like Midjourney and ChatGPT have lately evoked, they’ve also induced no small amount of anxiety. Does their ability to replicate cogent and internally consistent text, to remix images, and to reproduce or suggest code completions mean these AI dogmas were wrong? Will graphic designers, software developers, and communications professionals suddenly become obsolete when competing against tireless AI that can produce outputs faster and at a fraction of the cost?
The answer, as always, lies somewhere in between the extremes.
The limitations of Generative AI (for now)
It’s become difficult to make pronouncements about what AI can and can’t do for two reasons:
- 1. AI’s exponential progress tends to upend whatever we come to believe are its limitations. The public introduction to large language models and image transformers is a perfect recent example.
- 2. What consumers are seeing today in models like ChatGPT is still less capable than what Google, Meta, and even OpenAI themselves, have in their back pocket. Given that GPT-3 is a 175-billion parameter language model, and that Google and others have trained—but not yet released—1-trillion+ parameter models, what’s just around the corner will likely be much more impressive.
With those two caveats in mind, the current limitations of generative models are important to consider. For large language models (LLMs) like GPT-3.5—the basis of ChatGPT—text outputs can either lack substance or be outright wrong, despite how convincing it may read to us. It’s not hard for example to coax output from ChatGPT that is confidently wrong as anyone with an understanding of mathematics, software development, or any other particularly technical subject can attest.
Another problem plaguing generative AI is that they’re temporally locked to the data they were trained on—essentially, a snapshot in time that stretches backward from its most recent data, but not forward to the present. What this implies for models like ChatGPT or Microsoft’s new Bing Chat is that it sometimes gives outputs that are out of date. For image generators, this may mean they’re unable to replicate visual styles or fashions they haven’t yet been exposed to. (This doesn’t even take into account the fact that image generators still have difficulty replicating things that human artists never would, like fingers for example, or that they make contextual errors that are at once hilarious but also clearly wrong.)
The new white-collar worker, powered by AI
These problems probably aren’t insurmountable by future AI models however, as it’s likely that AI will overcome them sooner than later. Nonetheless, these issues persist for the moment, making strong arguments for the continued necessity of highly trained professionals and their judgment.
What value does generative AI bring to businesses, then? Like narrow AI before it, its benefits are of scale, both in productivity and output. Coupled with generative AI, writers, coders, and designers now have the ability to mock-up first drafts of copy, code, and images before refining them to finished products faster than ever. It can also help with idea generation and spur creative solutions with its ability to return millions of suggestions in seconds. Generative AI is therefore a tool to supercharge workers; not replace them.
What it won’t do is understand the context of a business’s culture or voice, losing the finer points of driving a message home. It won’t know the security, performance, or technical debt requirements of a business’s code base. It also certainly can’t speak to the aesthetic needs of a brand that make its design language unique. For all these scenarios, highly trained professionals remain vital as even in this age of generative AI, people are still the best “source of truth,” judgment, and expertise. They help ensure that AI outputs are true and beneficial to their businesses and the audience they serve, and as a result remain irreplaceable.
In much the same way, your business can rely on the judgment and expertise of Defined.ai. Over the past eight years, we’ve led the AI industry with high-quality, ethically sourced data for AI development. We believe that “if data is the lifeblood of AI, people are the lifeblood of data.” It’s therefore our mission to harness the best that people have to offer, ensuring your AI trains and benefits from data with their expertise and lived experience. Don’t take our word for it, though; see for yourself from the multitude of data samples available on the Defined.ai Marketplace.
How will generative AI supercharge your workforce? Whatever ideas you have, Defined.ai can help—reach out and let us know how we can best accelerate your business, powered by AI.
Leave a comment