Technology doesn't simply take away jobs - DSBoost #34

A lot of people fear that AI will take away jobs. But is it true? Let’s dig deeper and figure out with a study.

🎙️ Podcast of the week

Which Professions are threatened by LLMs - Dataskeptic

What does history say?

All technological advancements brought some type of change to the job market. There are winners and losers but generally, there are more winners overall.

Many of the jobs that exist today did not exist 15-20 years ago. New technology can of course take away some jobs, but also create them.

Generally, technology doesn’t simply take away jobs, it makes us more efficient, but our supervision and actions are still required.

But LLMs are something different.

We all know how powerful Chat GPT and other language models are. They can write essays, copy, and even code. The opportunities seem endless. It can be a problem but an opportunity at the same time. The question is how we use it, so the implementation will be the decisive point.

Implementation of AI and Language Models is now on the task level. We can do different tasks with them, but a job usually includes 30 or even 50+ tasks. They cannot really run entire processes that include several relatively complex tasks yet. 

In this paper, the research team used a dataset that documents the tasks required for a job. Then they measured how exposed different jobs are to LLMs and AI.

They have 3 categories [1]:

• Tasks are labelled ‘No exposure’ if direct access to the LLM cannot reduce the time required to complete the task by half or more. These tasks require a high degree of human interaction.

• ‘LLM+ exposure’ means that having access to the LLM alone may not reduce the time it takes to complete the task by at least half, but with additional software developed on top of the LLM could achieve the reduction.

• ‘Direct exposure’ means that direct access to the LLM can reduce the time it takes to complete the task by at least half.

The Results

Findings suggest that around 80% of the U.S. workforce could have at least 10% of their work tasks affected by LLMs. Approximately 19% of workers might see at least 50% of their tasks impacted.

Higher-income jobs potentially face greater exposure. For example, lawyers and doctors are more vulnerable than factory line workers, so it is possible that some of the tasks for lawyers will be automated and their salaries will decrease. On the other hand, if a lawyer can use AI well, they may get more productive and get ahead in their career, so inequality can increase even further in the long run.

Here are some jobs that are on the top of the list based on exposure:

• Interpreters and Translators

• Survey Researchers

• Poets, Lyricists, and Creative Writers

• Animal Scientists

• Public Relations Specialists

• Writers and Authors

• Mathematicians

• Tax Preparers

Additional Sources

[1] - https://arxiv.org/abs/2303.10130

📚Featured content

Am I too old to learn Machine Learning?

From Reddit:

No, it is not too late at all. I only started learning this in earnest a year ago and at 31 I have just started working in the ML field a few months ago. I do have a coding background but that is not as nearly as valuable as you think it should be.

The Skills You Need

Programming, math and statistics are the major areas you will need to learn. Without programming, you won't have a lot of luck progressing in ML so unless you already have decent coding skills, start here first:

• Programming: If you don't know a language yet, learn Python. If you already know a language, there are likely libraries you can use in the language of your choice but they will not have the support and community to solve basic problems you will encounter when learning so I would still advise investing the time in getting familiar with Python. If you already know one language, the learning curve to learn a new one is exponentially smoother than when you first started learning programming so don't let learning Python worry you.

• Math: What you will need to know depends on your goals and what problems you see yourself solving using ML.

  • Linear algebra: Linear equations, linear regression, vectors and matrices, Gaussian elimination, sub-spaces and affine spaces. This will help you in a few ways when starting out. You will have a better idea of what "features" of data are valuable (and importantly, which are not) when deciding what data feeds into your ML model. […]

  • Multivariate calculus: Partial derivatives (chain rule, second derivatives). Do not stress if you don't have a solid math background in the beginning. When starting out, I would advise just for a superficial understanding of what this is and circling back once you have seen it in practice. This will help you with what is happening in regards to "optimisation", and how an algorithm finds the best given your specified constraints. […]

• Statistics: Get a handle on basic statistics initially and circle back when you understand more about the algorithms you're using so you can get a better feeling of which is better suited to your problem domain.

[David]

In my opinion, it is never too late. There is always you can use from your previous experience that will be valuable in a Data Science / Machine Learning role. One could be domain knowledge, but not only that, any soft skills such as presentation, writing, managing, etc. can be really useful and a unique selling point when changing your field.

The fundamental skills you need to know are programming (especially Python) and basic statistics and maths. Complex maths is not generally required, only when you become very specialised or need to develop algorithms, which is rarely the case.

So in my opinion, first learn the basics of Python and SQL, and continue with more specific libraries such as pandas, numpy and matplotlib. Revise basic concepts of maths and statistics. Finally, practise a lot, first with toy datasets and later with real-world datasets.