Can AI learn to code?

营销策划 ctotodevelopers (源链)

AI is coming along with artistic impression. There is Paul and e-David that can draw , an AI that can write a Beatles inspired son g, followed by an AI that wrote a Christmas song . There’s been a lot of work on natural language processing, but natural language understanding remains elusive . It’s hard to keep up with all of AI’s creative works and how fast AI will go from pattern-based expressions to truly creative ones.

Coding, a significant AI challenge

Getting an AI to code is going to get a lot of attention. Microsoft Research and Cambridge University are experimenting with DeepCoder an AI that codes by reusing existing lines of code from other programs. It can solve challenges that require around five lines of code . It’s a significant achievement, though DeepCoder’s researchers admit that even basic programming challenges will take a lot more research and “Generating a really big piece of code in one shot is hard, and potentially unrealistic.”

Still, for those of us who grew up coding or began our careers as software developers, this achievement makes us wonder how hard it will be to succeed at the next challenge and how fast will we see breakthroughs in this line of research. Is it a matter of breaking down coding challenges to the right set of expressions for AI to process, or is there a fundamental creativity in coding that will be difficult for AI to replicate? Having AI assemble code rather than program fresh lines of code seems like a good starting point knowing that it is often easier and faster for humans to reuse code rather than build new.

But the hardest challenge may be in deciding what coding problems to apply AI and the optimal way to present coding tools or skills to an AI engine. Trying to get AI to code even simple programming assignments is a challenge because coding is an expression. It’s closer to natural language challenges, whereas successful AI has been in areas of pattern recognition (computer vision, basic forms of art and music) and decision making (playing games, self-driving cars).

We’ll have to wait and see!


Which Programming Languages Should You Learn in 20... One of the most common questions I get asked is this: “Which programming languages should I learn in 2018?” It’s a valid question, but it’s d...
The Quantum Pascal’s Triangle (2002) pascal Fromweek188 of This Week's Finds, October 11, 2002 The Quantum Pascal's Triangle John Baez Let's see what happens when you put Pasca...
Ready Made Plots make Work Easier A while back Simon Jackson and Kara Woo shared some great ideas and graphs on grouped bar charts and density plots ( link ). Win-Vector LLC ‘s Nina Z...
Won’t get fooled systematically (2010) Won't get fooled systematically I once had an idea for a game about deception. It would be like a tutorial for an ordinary game, but the instruction...
The Fallacy of DRY Originally posted on my blog as The Fallacy of DRY . DRY, standing for Don’t Repeat Yourself , is a well-known design principle in the ...
ctotodevelopers责编内容来自:ctotodevelopers (源链) | 更多关于

本站遵循[CC BY-NC-SA 4.0]。如您有版权、意见投诉等问题,请通过eMail联系我们处理。
酷辣虫 » Can AI learn to code?

专业 x 专注 x 聚合 x 分享 CC BY-NC-SA 4.0

使用声明 | 英豪名录