Altran's 'Code Defect AI' and the Rise of AI-Assisted Coding Tools

"Altran has released a new tool that uses artificial intelligence to help software engineers spot bugs during the coding process instead of at the end," reports TechRepublic. "Available on GitHub, Code Defect AI uses machine learning to analyze existing code, spot potential problems in new code, and suggest tests to diagnose and fix the errors." Walid Negm, group chief innovation officer at Altran, said that this new tool will help developers release quality code quickly. "The software release cycle needs algorithms that can help make strategic judgments, especially as code gets more complex," he said in a press release....

"Microsoft and Altran have been working together to improve the software development cycle, and Code Defect AI, powered by Microsoft Azure, is an innovative tool that can help software developers through the use of machine learning," said David Carmona, general manager of AI marketing at Microsoft, in a press release...

In a new report about artificial intelligence and software development, Deloitte predicts that more and more companies will use AI-assisted coding tools. From January 2018 to September 2019, software vendors launched dozens of AI-powered software development tools, and startups working in this space raised $704 million over a similar timeframe.... "The benefits of AI-assisted coding are numerous," according to Deloitte analysts David Schatsky and Sourabh Bumb, the authors of AI is Helping to Make Better Software. " However, the principal benefit for companies is efficiency. Many of the new AI-powered tools work in a similar way to spell- and grammar-checkers, enabling coders to reduce the number of keystrokes they need to type by around 50%. They can also spot bugs while code is being written, while they can also automate as many as half of the tests needed to confirm the quality of software." This capability is even more important as companies continue to rely on open-source code.
The Register got more details about Altran's Code Defect AI: The company told us that the AI does not look much at the source code itself, but rather at the commit metadata, "the number of files in the check-in, code complexity, density of the check-in, bug history of the file, history of the developer, experience of the developer in the particular module/file etc." Training of the model is done only on the project being examined...

Nice!

[nospam007]

An wetware manager AND an AI manager looking over my shoulder?
What's not to like?

Re: Will it help detect reams of needless boilerpl

[spongman]

The getters & setters have their advantages over plain fields as they allow you to chiange implementation without requiring a rebuild of the caller (which maybe built by a completely different company). Languages exist that make declaring & using getters&setter without requiring any boilerplate.

Re:

[infolation]

I put the TFA through my BS-TFA AI analyser and, after detecting the use of the acronym 'AI' 12 times, but 'machine learning' only 2 times, it's deduced that this coding-tool won't work.

Not for long

[SuperKendall]

An wetware manager AND an AI manager looking over my shoulder?

When the AI figures out the manager may let an unacceptable number of defects through, it will quickly find a path to the managers demise.

Now when the AI starts to figure maybe it can also do the coding better, well....

Re:

[weilawei]

When the AI figures out the manager may let an unacceptable number of defects through, it will quickly find a path to the managers demise.

Is that you, Skynet?

So that's why

[yassa2020]

This is the vaporware Dorsey was telling Yang about. Rich people LOL

Looks at the metadata

[gweihir]

Not really at the source. That already says it all. This is _not_ a bug-spotting tool. This is a compliance tool, utterly worthless for improving the code. It is apparently made by people that believe that if coders just follow all processes, then the resulting code must be good and bug-free.

The mind boggles at this _extreme_ level of stupidity. On the other hand, MS was involved, so my surprise level drops significantly.

Many Problem Domains will Benefit, but not All

[JBrow]

I can appreciate the fact that in some limited problem domains bug-assisted tools will be useful. AI gets hyped way too much in-general, as it always has historically. Until I can articulate to a machine "I want {this}" and it generates bug-free C code... well...

Not the bottleneck

[phantomfive]

However, the principal benefit for companies is efficiency. Many of the new AI-powered tools work in a similar way to spell- and grammar-checkers, enabling coders to reduce the number of keystrokes they need to type by around 50%.

Typing speed isn't the bottleneck for programmers, thinking speed is. If typing speed is your bottleneck, then you have too much duplication in your code, learn how to find the duplication in your code and put it into a function. Then the next time instead of typing 10-15 lines, you can type one.

Reducing redundancy is far more powerful than this AI.

Re:

[weilawei]

Thinking speed is no bottleneck at all if you aren't thinking!

Re:

[phantomfive]

Omg, you're right! What was I thinking?

Re:

[weilawei]

What was I thinking?

That was your first mistake. Now if you'll just follow this link to install Microsoft IntelliThink on your Apple iBrain so that you can offload that...

Are you f-ing kidding me?

[the_skywise]

The company told us that the AI does not look much at the source code itself, but rather at the commit metadata,

Click. Whirr. ~the number of comments in your commit is under 25% of the lines of committed code, this is in violation of policy 2918 - you now have 15 seconds to comply.~

Re:

[LordWabbit2]

Exactly, how is it supposed to determine good code from bad code from commit metadata? All it's determining is "good" commit habits against "bad" commit habits. I could write two hundred lines of "hello world" and as long as it's well documented during commits it's going to be "fine".
Another bullshit bingo fest, the only company stupid enough to buy this software is the same company stupid enough to put a non programmer in charge of other programmers. Sadly that is a lot of companies.

Re:

[aviators99]

But somehow it looks at "code complexity". I can't reconcile those two things. I imagine in addition to

git --numstat

it does

git log -p

which happens to include all of the source code. At the same time you could call this "commit metadata"

Now We Know

[Retired ICS]

Now we know why Microsoft has such a high defect rate in their code. They combine "low wattage" (ie, incompetent) programmers with Algorithmic Inference in order to achieve such stunningly high defect rates.

We always believed this to be true, know it has been confirmed directly from the horses anus.

Just have Siri write code

[cjonslashdot]

Siri is state of the art. Would you trust the code that it would write?

Today's machine learning systems are pattern recognizers and pattern generators. They don't actually _understand_ what they are doing. They are not sentient. They don't build a world model the way the mind does. The code written by today's systems will be so trivial that it will be little help - perhaps more of a nuisance, like the "suggestions" that keep popping up in some tools.

the mythical man words-per-minute

[Don Bright]

" However, the principal benefit for companies is efficiency. Many of the new AI-powered tools work in a similar way to spell- and grammar-checkers, enabling coders to reduce the number of keystrokes they need to type by around 50%." that's like saying you can improve the efficiency of Film production by letting writers have word completion in their word processor. the most time spent on building software should be about thinking, planning, coordinating, communicating, and testing, "coding" is a small pi