You've heard the hysteria a million times: "AI is going to take our jobs!" And for certain types of tasks — specifically, those that are simple, manual, repeatable — that may come true sooner than you think. With the uptick in AI tools like ChatGPT, people are already performing certain tasks much quicker with the help of AI.
- Need to write a short blurb for a proposal? Have AI write the first draft.
- Need a chatbot for your website? Have AI triage your customers.
- Need help fixing some code? Have AI be your debugging duck.
The truth of the matter is, AI is just better at doing certain tasks more efficiently than the human brain. However, it is far from perfect.
Why AI Needs our Assistance
1. Used improperly, you can easily get the wrong answer.
The simple fact of the matter is that AI can be wrong. Dead wrong.
Need ideas? Great! Need facts? Stay away!
— Mike Pearl, writer for Mashable, on ChatGPT
Take ChatGPT, for example. It is essentially an advanced chatbot that is able to produce language convincing enough to be confused with a human (i.e., it passed the Turing test). The large language models from which GPT-4 was derived contained 570GB of data from books, Wikipedia, and other internet sources, totaling a whopping 300 billion words fed into the system.
It knows quite a lot and it's good at producing convincing responses.
But that's the thing. The purpose of ChatGPT is not to be an all-knowing encyclopedia — it is to produce human-like responses. And so while its responses might sound really good, sometimes it's just as accurate as a hyperbolic salesman… it needs a good fact check. Stephen L. Carter, Yale law professor and contributor to Bloomberg, put it quite well:
ChatGPT can lie, but it's only imitating humans.
Many people struggle with admitting they're wrong, but more specifically, AI doesn't always have the ability to distinguish the truth. And that's because…
2. It takes a lot of work to perfect AI systems.
Let's take another example with self-driving cars.
While you've almost certainly heard of Tesla's "full self-driving" features, not only is it far from perfect but it's not even the best on the market, achieving only SAE L2 automation. The first vehicle to reach SAE L3 was the Japan-exclusive Honda Legend in 2021.
While there are driverless taxi companies like Waymo and Motional working on officially cracking SAE L4 within the decade, experts say that there's "no realistic chance" that full-on, L5 self-driving will be available before 2030. Given that the history of self-driving vehicles goes back as far as the 1920s with the first truly autonomous car appearing in the 1980s, you may be wondering…
Why is it taking so long to solve this problem?
Simply put, a task like driving requires you not only to be able to follow the rules of the road (simple instructions) but also to react to unexpected situations (complicated predictions), the latter being the main sticking point for AI. Humans have the capacity to learn from new experiences and make decisions instantly through millions of years of evolution whereas computers have limited memory storage and require more explicit programming, even in black box AI situations.
There are just so many different situations to prepare for. It's hard to capture all of them.
So despite the cutting-edge research available today in neuroscience, we still can't definitively describe what exactly allows human brains excel at these types of problem. They just do. (For now!)
What is Shared Autonomy?
A viable solution to the trickiest automation problems
Now that we've discussed some of the shortcomings of AI and how hard it is to perfect, let's talk about one promising approach to "solving" this problem in the short-term: shared autonomy.
Shared autonomy is a hybrid approach where both humans and machines collaborate to complete an automation task.
In this human-robot collaboration, the automated machine will run as long as its able to on its own but is able to explicitly request help from a human operator when needed. In other words, shared autonomy applies the 80/20 rule to these tasks: allow AI to solve 80% of the problem really well and then have humans to take over for the most difficult 20% of corner cases.
On the consumer end, the Tesla "autopilot", for example, can disengage at any time and requires the driver to intervene, such as if snow is covering the lines on the road. (But of course, the driver needs to be prepared for that.)
On the enterprise side of things, Third Wave Automation, an intelligent automation company based in California, applies this approach to supply chain automation through the use of automated forklift trucks.
The automated forklift can autonomously complete warehouse tasks such as picking up a pallet in a rack, but when it gets stuck, a human operator can log onto a remote operation system to complete the task manually and give back control to the vehicle. The intervention provided by the human operator can then be captured as new machine learning data to help the machine perform better on the next similar task.
This continuous learning process means that the system gets better the longer it is used, meaning a single remote operator can assist more and more vehicles to create "fanout" and gradually expand an autonomous fleet of vehicles over time, which can reduce operating costs and help combat labor shortages.
Impact on Human Workers
While this article is intended to focus on the technology itself, I'd like to briefly readdress the question from the start of this article: "Doesn't this mean AI will take our jobs?"
In some sense, yes, and the impact is significant according to a 2020 study by MIT and Boston University, primarily impacting manual and blue-collar jobs. This means that human labor will become increasingly skilled over time as demand for manual labor decreases.
On a positive note, this means that the back-breaking manual labor jobs that younger generations are refusing to work anyway will disappear and those older generations of experts (including those with work-related disabilities) can then utilize their years of expertise to perform these skilled shared autonomy tasks.
So while it is possible that AI may eventually be able to replace most (or even all) jobs, that is something much further ahead in the future. The best advice for us humans in the job market right now?
Work with AI, don't try to compete against it.