Artificial intelligence


AI is Learning To Code, But Ho...


AI is Learning To Code, But How Soon Could They Surpass Programmers?

AI is Learning To Code, But How Soon Could They Surpass Programmers?
The Silicon Review
02 November, 2021

Artificial intelligence is defined as intelligence demonstrated by machine, rather than from natural sources, and is one of the most beloved topics to programmers now more than ever. Teaching computers to fulfil the role of humans is becoming easier and more practical, with AI learning to do all sorts of things, from diagnosing certain conditions to beating people at Tetris. But how soon will these artificial minds be able to replace the biological ones that created them?

What Can AI Do Now?

Artificial intelligence is developing at an amazing rate. Whilst at first all it could really do was sort certain items, the technology has now developed hugely. From basic comprehension of the English language, AI rose to be able to write poetry of its own, begin to play games and create art and pictures. For the poker fans out there, some AI programs have even learned to play poker, able to understand bluffs and calculate the odds of their winning hand in seconds. It’s good enough to give even the players at GGPoker a run for their money.

But it seems now that AI is advancing further into the field of human ability and even human jobs. With the newest startup SourceAI setting up its own AI companion to make lives easier, it seems that computer programs will begin taking over human jobs potentially in the near future. But what does this new AI do so well?

SourceAI’s Coding Assistant

SourceAI is able to write code - A machine that can write the instructions for another machine. Whilst it seems like far-flung future technology, this AI is very much in the here and now. The AI is currently not available to the public, though a demo is available online and corporations can look into buying the licence ahead of release. This amazing program is able to take plain English, Japanese or any other human language, figure out what is being asked of it, and can write a program to fulfil that request in over forty different coding languages from Java to SQL.

Not only can this AI interpret a request and code for a user, it is also able to scan through an engineer's code to find errors and correct them. This means that this AI, once in full production, could be used to proof-read code and make corrections whilst the human engineers work on implementation of new features and code. This is a program that can make the coding process much easier and simpler, speeding up turnaround times across all phases of the project.

How Does It Actually Work?

Unfortunately, peering behind the curtain at this wizard does reveal the process to be a lot less magical. The AI is not capable of fully inferring and understanding the user’s request the way a human would, but instead scans the request for patterns that it has learned from GPT-3’s huge neural network. GPT-3 is the most advanced AI language program that has been released so far, and is used to gather scraps of information from all over the internet to teach derivative programs how to interpret language.

Once it has spotted some recognisable patterns in the request, it compares them to its own “memory” of other similar patterns and what code fulfilled those requests. It then breaks apart the lines of code it needs and reformats them into a functional program that fits what the user asked for. Pretty neat right?

AI is a complicated field, made more complicated and more insane by the day as more and more people create fantastic systems with it. Whilst AI programs likely won't take over any human-powered jobs for the time being, it shouldn’t be shrugged off. With the advancements coming at an exponential rate, who knows where we could find machine thinking going next?