Thursday, May 11, 2023

Winter is Coming

I fear we are heading to a computer science winter. Now why would I say that when CS enrollments are at record levels and AI is driving incredible excitement in the field? Back in 2016 I wrote

We won’t see a decline in demand for computer science students until we automate ourselves out of a job.

With advances in machine learning, especially generative AI, you can now use AI tools with little to no knowledge of computer science and mathematics. You can do quite a bit more with just a basic knowledge of Python programming and APIs. And tools like Github co-pilot make programming that much easier.

In 2005 during the last period computer science saw small enrollments, I suggested computing became a commodity and we lost the excitement in the field, leading to a decrease of interest and students. It didn't help that potential students had a (mostly mistaken) perception that all the computing jobs were being outsourced to India.

We may soon see a time when computing loses its excitement again if everyone can just interact in English (or any other language). Students might now have a perception that computing jobs will be outsourced to AI. The recent tech layoffs don't help. Even the ones interested in computing might focus more on the various low-cost certificate programs instead of a full CS degree.

What can we do? We need to reframe our degrees or create new ones to recognize the move to data-oriented computing. We need to embrace teaching responsible AI but without fighting the future. 

CS is in a great place right now but we have to continually adjust to ensure our future relevance or we may no longer have it.


  1. Who are the people who design the AIs that the jobs will be outsourced to? They are computer scientists. Computer scientists will not lose their relevance before they have automated everyone else's jobs too.

    1. You could say the same thing about the Internet in 2005 but until new technologies came around (cloud, mobile) the excitement in CS was limited.

  2. Are we (particularly in complexity theory) still asking the same questions? And are those questions going away? Would you or Bill or anyone read a ChatGPT or other AI-generated proof that P = NP (cause we're in a different world here, we assume that all it takes to prove this is to provide an algorithm but the AI tool may come up with a non-constructive proof)? And even with that, its like you say in your book, the thing is more about what happens after that. Say "they" prove it. Then nobody wants to read it. Then somebody finally reads it / translates it / publishes it / writes an algorithm from it that shows that something. Then we'd be in a new era of CS.

    Right now. We're in an era, where its "I've got this new calculator* but I've got to show you how to use it or you'll get the wrong answers." These things are being trained on bad data and answering questions incorrectly, so badly that its a running joke right now. We know that they will get better because things always get better. The question is who is doing the guiding towards "better" and who is the judge of "better".

    Right now we have an opportunity to have an educated society take part in the answering of these questions, or a fearful society.

  3. I will definitely read a P vs. NP proof whether or not it is AI-generated. However, the most likely scenerio is that AI won't be capable of solving hard math problems like P vs. NP, but it will be capable of putting many people out of their jobs.

  4. During my master in human ressource management and dynamics of organisations, a sociology teacher told us that at the end the 60's, early 70's, people were scared computers would put many out of a job. "Workforce would not be needed anymore!"
    It turns that no study has ever shown that the job market shrunk globally (at least in Europe) because of computers. In hindsight, quite the opposite happened.
    I think we are seeing a repetition of the same fear with AI and chatGPT. Those will act as strong cost deflation forces, enabling tons of new businesses.
    Besides, I have worked with international global payroll EPRs (I mean under the hood + hands in the dirt). These are incredibly complex considering legislative requirements. I am confident I won't see any chatGPT generate an operational and fully fonctional global payroll solution (other than copying an existing one) in my lifetime.
    In addition, a 3 stars engineer cannot do a 5 star engineer job. The latter will still be needed forever. Scarcity will trigger high compensations and thus motivate many.
    Winter is coming? I am confident it won't last forever...

  5. The problem still relies in human resources. In seeing people who look like you, from your neighborhood or where you grew up and made it and are doing good and have a story to tell. It's a problem is can you understand my weird problem that is harder to put into words than it is to solve.

  6. "I fear we are heading to a computer science winter."

    What you seem to have meant was a low-skill computer programmer winter. If department sizes retreat to the student-to-faculty ratio levels they used to be at and leadership can prevent cuts (they claim they cannot fund more faculty lines because interest might slump, fine, when interest slumps do not cut).

    The "information sciences" departments that has students that are hyped-up as "we're just as good at it as computer science majors" are the ones to be fearing this.

    As the true computer science world shifts, departments that produce computer scientists and true software engineers can shift with it as needed.

    Departments grown for students who can not cut it in real CS cannot shift when what is still needed are real computer scientists and software engineers.

    Students who have trouble debugging their own code or can hardly imagine writing full systems are not going to be able to glue together and debug the code the "GPT"s of the world spit out.