Deepmind invents AI tool to write novel computer code. In a coding campaign and placing in the middle, DeepMind’s AI has created an AI that can write code to solve any problem given to it, as demonstrated by taking part in a coding challenge and finishing — well, somewhere in the middle. It won’t be replacing software developers anytime soon, but it’s promising and may assist with basic chores.
DeepMind, a subsidiary of Google, is working to generate smartness in as many forms as possible, and coding is certainly one of the tasks that many of our great minds are focused on.
Obviously, it’s not the first to try something like this: OpenAI has a similar Codex natural-language coding initiative, and it is used by GitHub Copilot as well as a Microsoft test that allows GPT-3 to finish your sentences.
In their study, DeepMind’s researchers defend their approach by stating that they’re not just interested in mastering AI but also in creating entirely new domains of application:
“Recent large-scale language models have demonstrated an impressive ability to generate code, and are now able to complete simple programming tasks. However, these models still perform poorly when evaluated on more complex, unseen problems that require problem-solving skills beyond simply translating instructions into code.”
![AlphaCode AI from DeepMind produces code at a competitive level 1 AlphaCode AI from DeepMind produces code at a competitive level](https://techbriefly.com/wp-content/uploads/2022/02/dims-e1575998404106.jpg)
However, even if OpenAI has something to say about it (and we can probably anticipate a riposte in its next paper on these lines), competitive programming problems generally entail a mix of interpretation and ingenuity that current code AIs don’t show.
AlphaCode AI from DeepMind trained a new model
DeepMind trained a new model on GitHub libraries and a collection of coding problems and solutions to tackle the domain. Simply put, but not an easy task. They then deployed it on the 10 most recent (and needless to say, unseen by the AI) competitions from Codeforces, which is responsible for this sort of competition.
Its accuracy was mediocre, placing it in the middle of the pack, just above the 50th percentile. That may be an average performance for a human (not that it’s simple), but for a machine learning method’s first try, it’s quite incredible.
“I can safely say the results of AlphaCode exceeded my expectations,” said Mike Mirzayanov. “I was skeptical because even in simple competitive problems it is often required not only to implement the algorithm, but also (and this is the most difficult part) to invent it. AlphaCode managed to perform at the level of a promising new competitor.”
The following is an example of the sort of problem that AlphaCode fixed, and its solution:
![AlphaCode AI from DeepMind produces code at a competitive level 2 AlphaCode AI from DeepMind produces code at a competitive level](https://techbriefly.com/wp-content/uploads/2022/02/alphacode-problem.jpg)
![AlphaCode AI from DeepMind produces code at a competitive level 3 AlphaCode AI from DeepMind produces code at a competitive level](https://techbriefly.com/wp-content/uploads/2022/02/alphacode-solution.jpg)
This is, of course, still a work in progress. It’s not yet enterprise SaaS grade stuff, as you can see. Don’t worry; it’ll come later. Right now, all we need to show is that the model can handle and comprehend a complicated written question at once and provide a sensible, functional response most of the time
“Our exploration into code generation leaves vast room for improvement and hints at even more exciting ideas that could help programmers improve their productivity and open up the field to people who do not currently write code,” writes the DeepMind team.
At this demo site, you may learn more about how AlphaCode was created and the solutions to various problems. DeepMind’s share price by 2/2/22 is 2,959.09 USD.