Over the last few weeks, AI has come up in every conversation with friends and family. ChatGPT has truly made AI mainstream. Broadly, people fall into two buckets: excited and scared.
People are excited because it is clearly the next tech revolution and they want to build something. It’s a healthy mix of people who’ve worked in tech and domain experts who want to solve a specific pain point in their industry.
The worry is understandable. AI is capable of running 24/7 and doing tasks for you. It feels like it will take away jobs. Coincidentally, OpenAI (company behind ChatGPT) released a paper claiming that:
80% of the U.S. workforce could have at least 10% of their work tasks affected by the introduction of GPTs, while around 19% of workers may see at least 50% of their tasks impacted.
I’m going to argue that AI will replace tasks, not jobs.
AI is not 100% accurate today.
If output is not 100% accurate, you will want to check it before putting it in front of your customers or colleagues. In other words, for the vast majority of tasks, we’ll need a human in the loop vs. AI automatically doing things.
This will reduce dramatically as AI learns. Learning will happen across the board but also for a specific customer that the product is serving.
Last week, Intercom announced Fin: it’s new customer support bot driven by GPT-4. They trialed the bot internally (i.e. customer service agents use it first) before exposing it externally. It uses sources from a company’s knowledge base to answer questions and offers seamless handover to a human agent if it cannot answer the question. It’s a great example of how the accuracy problem will get solved over time: learn from your mistakes, and use a human when it’s really required.
Complete task automation
For a job to be replaced entirely, every task has to be automated.
I don’t think this is going to happen as quickly as it might seem. Doubling down on the point above, there are certain tasks where the consequence of inaccuracy might be too grave to even trial AI.
Consider a customer care agent for a service like Uber. If it’s a sensitive issue (e.g. a passenger reporting that they feel unsafe), you will not want AI handling that. It does not matter how well AI might perform, you simply don’t want to take the risk.
The single biggest reason I do not believe AI will replace jobs is because of coordination.
Most of our work today is cross-functional. For example, a copywriter writes copy and many other things. They might need to talk to marketing to align on the new tone of voice. After that, they might want to talk to business intelligence to understand how content has performed to date and how they should adapt their output.
Even if we had 100% accuracy, we’re not in a position where AI can coordinate work for us. The OpenAI paper alludes to this in it’s weaknesses section:
- Tasks or activities where while an LLM could theoretically help or accomplish the task, adopting it to do so would require multiple people to change their habits or expectations (e.g. meetings, negotiations) – Tasks or activities where there is currently some regulation that requires human oversight or norm that suggests human judgment or empathy (e.g. making decisions, counseling), and – Tasks or activities where there already exists a technology that can reasonably automate the task (e.g. making reservations).
Think about how many jobs have at least one of the above.
Augment > replace
For most jobs, we will see AI augment rather than replace. Earlier this week, Microsoft announced the Microsoft 365 Copilot. Microsoft is integrating AI into business apps like Powerpoint, Excel and so on.
Microsoft’s announcement had some really interesting thoughts on how they are thinking through human computer interaction in the age of AI. Human agency or decision making is given priority over everything else. They acknowledge that AI is often wrong but is “usefully” wrong. What they mean is, AI being wrong might be signal that it needs more guidance or that the underlying data it was provided was insufficient. Their vision is for tools to learn how you work as opposed to us having to learn how tools work (the world we are used to).
AI takes away certain tasks and gives you the ability to do more. I’ve been using Github Copilot and ChatGPT Plus to write code for a while now. It is incredible but I still need to guide ChatGPT on what I want (usually with >2 tries) and then make a decision on whether the output is appropriate. Using it to augment my daily tasks has been an incredible productivity boost. This is how I believe AI will influence our lives in the foreseeable future.
AI penetration by job type
Certain jobs might end up being more automated than others. These jobs typically have a single type of task or a bunch of related tasks that can be easily automated. These tasks usually involve repeatable actions like retrieving knowledge (paralegals) or taking a limited set of answers (first line customer support).
The OpenAI paper investigates the relationship between a bunch of factors and the extent to which AI can take over:
- Wages: contrary to previous work, they find that higher wage occupations have more exposure to task automation with AI.
- Skill importance: jobs requiring science and critical thinking have less tasks that AI can do. Jobs requiring programming and writing have more tasks that are likely to be handled by AI.
- Formal education: jobs requiring people with formal education are more likely to have their tasks automated with AI.
Here are two tables showing jobs with the highest exposure to task automation and jobs with the lowest exposure to task automation:
The dynamics make sense when you take a step back. Jobs with higher wages and requiring more formal education tend to be digital. These jobs have tasks that can be taken on by AI. On the other hand, a carpenter can rest assured that it’s take a while before AI can fix your sofa.
If you are concerned about AI automating away your job, please don’t be. It will automate tasks and not the job entirely. This is because it’s not 100% accurate, cannot do 100% of tasks and is unable to coordinate the way humans can.
My recommendation is to take that worry and channel it into learning a little more about AI. One of the most ground breaking innovations with AI is that it allows anyone to communicate with a computer. The ability to prompt (guide ChatGPT to help you with a task) is going to be a very important skill in the future.
Previously, you had to know how to code. With AI, you communicate with natural language. If you can instruct a computer to help you do your job, you are going to be an extremely valuable candidate in the job market.