From Manual Coding to AI-Powered Software 2.0

The era of manually coding your program is long behind us. Now it is Software 2.0.

The evolution of technology has brought us to the next level. Recently, it was even possible to make a program without having to know its fundamental code first. A project leader at NVIDIA GameGAN explained that their team successfully recreated PAC-MAN by having AI learn the rules of an environment only by observing the screenplay of moving agents.?

Such progress means that someday game developers, or developers in general, don't need to code every object interaction manually. This opens the potential for a Software 2.0 phenomenon, where machine learning can replace traditional software coding.

A Closer Look at Software 2.0

Software 2.0 differs from Software 1.0, or ‘traditional code’, in terms of the programming language used and how it is generated. According to Andrej Karpathy, Software 2.0 is written in a human-unfriendly language, a more abstract one, and is created using computational resources, such as neural networks.

Instead of writing the code manually, Software 2.0 uses a trained neural network to create the desired outcome. It spreads through backpropagation. Backpropagation is a gradual adjustment of the neural network nodes until it reaches the expected result. It’s just like Karpathy stated, Software 2.0 is code written by optimization based on an evaluation criterion.

That being said, Software 2.0 needs a huge source of data to give a satisfactory result. Those data need to be labeled. In this case, TechTalks guesses that, at some point, software engineers’ roles will turn into “data curators” or “data enablers.”

Examples of Software 2.0

Apart from NVIDIA GameGAN, there are several other programs and/or services that are transitioning to a Software 2.0 stack.?

Karpathy pointed out that features like visual recognition used to require very simple machine learning. However, now they use large datasets and search in the space of Convolutional Neural Network architectures.

While Karpathy doesn't mention any programs that use Software 2.0, Ahmad Mustapha provides us with several examples. The first one he mentioned was Prisma. It's a photo-editing mobile app, and it utilizes neural networks and AI to add effects to images. Prisma uses neural style transfer technology, Mustapha added, to redraw an image based on another image style.

Another example is Lobe, a platform for building Software 2.0 products. The website claims that it "helps you train machine learning models with a free, easy-to-use tool." You can train your app to recognize different colors, different facial expressions, and so on.

The Benefit of Software 2.0

With such advanced technology, there are several benefits of Software 2.0. The ability to train your app to recognize particular patterns is useful in identifying and filtering harmful materials, Rofi Labs stated, such as explicit images, texts, videos, and audio contents. It can also filter off-brand and poor-quality content and makes it easier to remove.

Aside from removing harmful content, it can also help in recognizing faces for security purposes. FaceID is one of the face recognition apps. Other than unlocking phones, Kaspersky mentioned that it is also used in law enforcement, for example, to filter civilians from people on a watch list.?

The Future of Software 2.0

Most of the articles used in this research mentioned the possibility of technology, in this case, Software 2.0, overtaking the roles of human software engineers. While it might happen someday, it's still a long journey to reach that point. There are a lot of things to adjust and work on, and that definitely cannot be done in just a single night.

It is important to remember that technology is built to support human tasks. This includes Software 2.0 as well. Ultimately, it's still a machine that requires monitoring in order to minimize failure to zero percent. It won't replace humans; it's like adding a new partner for a better and easier future.