Outsourcing Thinking: Will AI Make Human Problem Solving Obsolete?
As AI systems like GPT-4 automate more of the problem-solving process, what role will human ingenuity and creativity play? Let's explore the evolving dynamics between humans and AI.
My name is Jacky, and I am an AI engineer at a Infinitus, an AI healthcare startup. I was formerly a software developer at Looker, product manager at Singlestore, and solutions architect at Oracle Cloud. Currently, I am one of the “first crop” of folks applying artificial intelligence systems like LLMs to solving problems at organizations. I would like to think I have a fairly practical and realistic look of the world of solving real world problems with AI and LLMs.
Read time: 5 minutes
Since the dawn of humanity, us humans have relied on our ingenuity, adaptability, and reasoning to solve problems.
We've followed a tried-and-true framework: identify the issue, research, plan, execute, monitor progress, refine the solution, launch, and evaluate results. But I am postulating that we may be seeing a great change to this paradigm thanks to modern AI systems like GPT-4, Claude, Bard, and more. These large language models can act as simple logic engines to automate pieces of the problem-solving process.
In this blog post, we'll talk about how humans have historically solved problems, where AI shows promise (and struggles), and what the future may hold as these systems advance. Could AI someday manage the full problem-solving workflow on its own?
The Classical Human Problem-Solving Framework
Problem-solving is a skill deeply embedded in human history. It has paved the way for every innovation, from fire, the wheel, the printing press, to the Internet and beyond (P.S. I wrote about this in my previous article). The steps we follow has traditionally followed the following framework**:
Identify & research: The beginning of any problem solving is the identification of the problem + getting as much context as possible. This is also where the majority of information gathering happens.
Plan & prepare: Next, we brainstorm the ideas and solutions, pull together resources, and get stakeholders on board.
Execute & monitor: The plan is then put into motion and we must keep track of progress.
Review & refine: After this, we test things out and tweak the solution as needed.
Launch & evaluate: Last but not least, we launch the finished product and see measure its real impact. Continuous iteration and improvement also happens at this step.
For thousands of years, this has been a fully human endeavor, and we’ve adapted this framework to tackle all sorts of challenges over the centuries. Sure, modern day humans have had the assistance of machines like computers, code, robots, Wikipedia, search engines, etc, but operation of these machines always started and ended with humans.
** P.S. I am not saying this is the only way we solve problems, but it is a general framework we use to solve most problems.
A Small Glimpse Into The Future
New systems like GPT-4 represent a fundamental shift in artificial intelligence.
No longer does AI need to be mere scalpels solving extremely specific problems — narrow use cases like beating experts at Go/Starcraft, image recognition, or sentiment analysis. Current LLMs, given a well-defined problem and enough background, can act as basic reasoning and logic engines. Research shows that LLMs can do more than just summarizing long texts and writing annoying emails to your coworkers — they have slowly become “anything tools”.
No longer does AI need to be mere scalpels solving extremely specific problems.
GPT-4 (in my case, the GPT-4 API) especially excels during the Execute & Monitor phase. If the problem has clear requirements, it can generate high-quality solutions that get the job done quick and accurately. I've seen this firsthand at my job - it seriously cuts down on the time humans spend on execution. But I still need to meticulously fact-check and review its work; the outputs aren't flawless yet. But it’s 80-90% there.
The Future: GPT-5 and Beyond
Imagine LLMs like GPT-5, Gemini, Claude 3, Llama 3, and more, progressing to a point where they can manage the entire problem-solving loop. They would identify a problem, research it, plan and prepare for its solution, and even engage stakeholders—human or machine. It wouldn't stop there. These models would review their own performance, refine the solution, and re-iterate, all without human intervention.
Imagine LLMs like GPT-5, Gemini, Claude 3, Llama 3, and more, progressing to a point where they can manage the entire problem-solving loop.
In this future, research and information gathering will no longer be a human pursuit. We are already seeing GPT-powered tools like AutoGPT that can do rudimentary research, information gathering, planning, and execution. They are absolutely not production-ready nor anywhere near as “automated” as their fancy marketing suggest (and to be honest, I think things like AutoGPT are some of the most overhyped and overrated things in the LLM world right now), but this is a small glimpse of the future.
Another part of the human problem solving loop — post-launch processes like A/B testing might become a quaint relic. The machine would iteratively improve its solutions based on real-time evaluations, far exceeding human capabilities in the speed and scale of refinements. Again, we are already seeing automated GPT-powered A/B testing tools like Coframe.
And just for your information, all these ideas + tools launched just within the past 12 months as of the writing of this blog post. What the next 12 months will look like is anyone’s guess, because like with the launch week of GPT-4, it truly was a week that felt like a year in terms of AI developments.
What the next 12 months will look like is anyone’s guess, because like with the launch week of GPT-4, it truly was a week that felt like a year in terms of AI developments.
The Big Question: What Happens to Us?
When machines cover the whole loop of problem-solving, from identification to evaluation, what's left for us? It may sound like a dystopian narrative, but it's essential to remember two things about human nature:
Innovation: Our ingenuity and knack for problem-solving won't disappear; they'll find new avenues.
Insatiable curiosity: Human curiosity and drive to understand is limitless. When one frontier is conquered, new ones continually capture our imagination.
It’s not about machines replacing us, but about humans and machines discovering a new equilibrium. We will move on to new challenges, perhaps those that we couldn't even fathom without the help of AI. And so, the dance between human creativity and machine efficiency will continue, on and on, into an ever-evolving future.
An optimistic thought is that we will discover problems so advanced that their solutions can only be solved combining the best of modern day AI + human ingenuity. Let’s focus our time on thinking about what those problems may look like?