AlphaCode vs. Openai: How Artificial Intelligence Meets Human-Generated Programming Skills (and Why) Humans Will Need It
AlphaCode, an artificial intelligence system which can often compete with humans at solving simple computer science problems, has been appreciated by a group of artificial intelligence researchers. The AlphaCode tool, which was released in February by DeepMind, has now been published in Science1, showing that it beat half of the humans at a code competition.
In the past week or so, social-media users have been enthralled by the ability of a newchat called chat gp to provide occasional meaningful sounding (and sometimes unbelievably ridiculous) mini-essays. According to researchers, the state-of- the-art artificial intelligences don’t have the ability to replace human programmers.
The new version of the natural-language system by Openai was released on November 30. Both ChatGPT and AlphaCode are ‘large language models’ — systems based on neural networks that learn to perform a task by digesting massive amounts of existing human-generated text. In fact, the two systems use “virtually the same architecture”, says Zico Kolter, a computer scientist at Carnegie Mellon University in Pittsburgh, Pennsylvania. There are differences in training and execution, but the main difference is that they are trained upon different data sets and therefore for different tasks.
Researchers have pointed out that much of the work that goes into a large software-engineering project — say, designing a web browser — involves understanding the needs of humans who are going to use it. These are difficult to describe with the simple, machine-readable specifications that an AI can use to produce code.
Kolter says that it’s unclear whether it will ever be possible for machines to generate large-scale software systems from scratch. But “my best guess is that tools like these that can generate portions of a program will likely become ‘second-nature’ kind of tools for programmers”, he says.
By applying the same underlying technology used to create GitHub Copilot, it will be possible to build Copilots for virtually any complex, repetitive aspect of knowledge work, allowing knowledge workers to spend their time on higher-order cognitive tasks, and effectively transforming how a great many of us interact with technology to get things done.
Without a sense of wonder, it is hard to appreciate the sheer magnitude of innovation taking place in the area of artificial intelligence. Over the last few weeks, researchers across industry and academia have published work that advances the state-of-the-art in nearly every area of Artificial Intelligence, from benchmarking to feats beyond what we could have imagined a few years ago.
Historically, computer programming has been all about translation: Humans must learn the language of machines to communicate with them. Codex lets us use natural language and the machine will translate our intentions into code. It is a translation of the human imagination to any piece of software.
That is a remarkable step forwards in productivity for developers alone, a community of knowledge workers who are wrestling with extraordinary complexity and unprecedented demand for their talents. But it’s just the first step of many that will be taken in 2023 as we see this pattern repeated across other sorts of knowledge work.
Our increasingly complicated and information-dense world requires more knowledge work every year, imposing ever-greater demands on those workers in every field and industry. Copilots for Everything could give a genuine revolution to work where productivity has been fairly constant since the introduction of the internet.