How will the Software Engineering be in 20 years?

[aeroxis]

20 years ago, we were dealing with PHP mainly and the LAMP stack, or JSPs and deploying to JBoss/Tomcat. Today we deploy apps written in any language to containers.

Where do you think software engineering will be in 20 years?

[mikewarot]

If all goes well ---

In 20 years, we'll finally have capability based security in our operating systems. We won't need virtualization, containers, "safe" programming, or the chiding of users.

We'll finally get back to the levels of productivity that we had back in the days of HyperCard, Visual Basic, Delphi and SmallTalk.

We'll have renamed the mutex and lock to "bottleneck". We'll have a far better understanding of the physics of software when you have multiple processors.

Source code won't be straight text. You'll be able to include other media in the source.

If all goes wrong ---

Software will only be written in legally authorized environments by legally licensed people using clearly specified "safe" techniques.

General purpose computing won't exist.

All data will be periodically scanned to protect the children, protect against forbidden ideology, etc.

[redanddead]

I think it will be more accessible, democratized, for better or for worse.

I imagine coding in some kind of visual or VR way, something more intuitive and human.

[batzy]

Maybe no-code, low-code things will improve and, wil reduce the job market and the salary scales. Who knows..

[jschveibinz]

Here’s a radical thought: software engineering will for the most part cease to exist in 20 years.

Why? How? Let’s think about this like a Sci-Fi story…

To begin, it is not difficult to imagine that AI systems will trained on prior art in the same way they are trained on images, essays, etc. If you saw this in a movie, it would be totally believable.

So in this way, simple text prompts will produce entire, tested software programs.

This is just like asking DALL-e to show a “picture of a goat standing on top of a cow in a field of sunflowers on a sunny day.”

Furthermore, it is quite possible that this will be done as part of the AI “operating system” of the computer.

In other words, the computer system will naturally request or seek specific training sets to “learn” how to solve a specific problem as part of its natural functionality in meeting the request of the user.

For example, let’s say we are on a secret mission and we need a totally new way of communicating with headquarters that can’t be easily broken in 24 hours.

Once upon a time…

We ask our computer system to solve an important problem. We need a new, secure communications capability.

Prompt: “Build a novel secure communications capability”

The computer responds ”ok,” then requests specific parameters, or the specifications for the system:

“Are you able to answer these questions,” it says.

“Frequency range, modulation preference, error coding, permissible bandwidth, frame size, handshake, probability of intercept, message length, etc.”

You aren’t a comms engineer, so you say “no” and then ask for the computer system to ask for higher level requirements.

The computer says “ok,” and then asks “how secure do you need your comms system to be, how long is the message you want to send, how many times do you need to use this, can it be seen or must it be hidden and how much time do I have to send a message?”

…or something like that.

The computer may also request the necessary quality and documentation required. In other words, it asks “does this software need to be readable by a human?”

With these answers that you provide, then the system requests access to prior art from open source secure communications systems that already exist.

You are prepared for this part, so you provide the necessary details.

From these data, the AI computer system takes over and produces several potential solutions, tests these against the specs, and then iterates to produce a final set of solutions.

In the end, with a little bit of human input, the system produces an optimal solution to the secure communications problem.

And it only takes ten minutes to do all of this.

The end.

It’s just a story, but given the state of AI, I would predict that we will be seeing the first of these systems in commercial use in the near future-certainly before the end of the decade.

And I would not be shocked if these already exist behind closed doors.

It is difficult to imagine how this will affect the economy. It may not be pretty.

But disruptions happen dramatically. In other words, relatively quickly with mass adoption.

There were horses for transportation, and then there weren’t.

There were factories and factory jobs, and then there weren’t.

It took weeks of research at a library to find information, and then it took ten minutes.

I believe that change will happen dramatically in software sometime soon. If you use your imagination, you might see that change has to happen because of economic pressure. There aren’t enough programmers, the salaries are getting very high, and technology demands are growing exponentially.

[s-t-876]

>I believe that change will happen dramatically in software sometime soon. If you use your imagination, you might see that change has to happen because of economic pressure. There aren’t enough programmers, the salaries are getting very high, and technology demands are growing exponentially.

okay so in twenty years that would be 2042, that is around mid century that has many negative climate change predictions, and it assumes our energy supply by that time will continue to increase with the demand for more tech, people somehow will manage to carry on and adjust and adapt to climate change with clean energy and it will not be severe enough to limit tech ambitions.

based on what you believe should software developers[ today ]continue to think of themselves as employees-first? or owners first? an AI can't own anything, corporations own the AI. In this context are you thinking of software engineers as co-owners or labor?

[cratermoon]

This all sounds nice except for one thing. It's not clear that "AI" can exist without programmers. There's a bit of a rule of thumb about a system not being able to create a thing more complex than itself, which means that even the best machine learning systems need human laborers to care for it.

Of course, much of that human labor will be essentially slave labor, much like the social media and recommendation systems simply use inputs taken from millions of user interactions, but some of it will have to be expert.

For example "the system produces an optimal solution to the secure communications problem". OK, but how is it determined that the solution is optimal? The system cannot within itself prove it is optimal (Gödel's incompleteness theorem), so something external to the system must be involved.