With speech-to-code tools, teaching intent and context is a definite constraint
Even as the hype around code-suggesting tools like Copilot hasn’t died down, it is time for making artificial intelligence to generate code with textual prompts. Companies are increasingly depending on applications requiring human programmers to develop and maintain the software, and according to a 2019 IDC report, the existing pool of programmers wouldn’t be sufficient for future industry needs. Though this implies that speech-to-code will come to the rescue, there is every chance that it will replace humans. Or is it just a misplaced assumption? Under the hood, the machine learning algorithms are trained to generate code on textual commands by analyzing millions of code snippets helping develop low-code applications. But how a programmer makes a machine learning algorithm understand his intent is the challenge. Put otherwise, how a code developer asks, recommends, and advises the computer.
Code suggestion tools basically have optimized IDEs (Integrated Development Environments), trained for many years around a model of programming as dictation. Therefore, programming as a process of conversation that requires teaching intent and context always generates an altogether different set of constraints every time it faces a new context demanding new workflows and hence new tools. It is reasonable to say conversational code is not fool-proof yet for it to replace humans entirely from the programming scene even after overcoming usual constraints with machine learning programming such as limited availability of training data, limited computational resources, and the complexity of the interface between algorithms and people. While the first two constraints can be overcome over some time with machine learning algorithms working around routine tasks such as code completion, code search, and bug detection, the last one will definitely augment the need for human presence. While the low-code or no-code will meet the development requirements in long term, it will never be the panacea. Systems will only become easier to build for common processes and use cases but companies cannot make conversational programming as a mainstay of the tech stack development.
Source: analyticsinsight.net
