How to augment people with AI to build a better working world

Democratization, diversity and inclusiveness

Until now, creating high-quality information goods like software, music or artwork required humans to undertake hundreds if not thousands of hours of practice. But thanks to LLMs’ natural language interface and GenAI, these skills are now widely available to anyone with elementary level education. Until now, creating code required humans to understand computer languages: C++, Javascript, Python, and so on. Today, computers understand our languages, English, Spanish, French and Chinese. Even children can craft text prompts that generate workable software and professional-looking graphics.

With greater democratization and reduced barriers to entry, the diversity of people working in disrupted industries should increase. For instance, software engineering is one of the least diverse of all. Approximately 90% of all software engineers globally are male and more than 50% are Caucasian.⁷ ⁸ A lack of diverse talent in the technology sector raises particularly acute social and ethical problems with the development of AI techniques, where algorithmic bias risks perpetuating existing social inequities and privacy risks disproportionally harm underrepresented or underprivileged minority groups. Yet with GitHub Copilot, we can expect engineering talent to increasingly come from unexpected places. Today about 80% of software engineers globally have a college degree, but that share should decline with reduced barriers to entry.⁹

However, probably the biggest positive impact of GenAI on inclusiveness is via the way these models pull workers toward the skill and productivity frontier. Empirical evidence is building that the latest AI advances drive the largest productivity improvements in workers with less experience, training, competency and skill attainment. At the same time, the most productive, experienced, competent and skilled workers show little benefit from working alongside an AI, as they are already at or near the productivity frontier.

Lower-skilled workers benefit the most from the knowledge embedded in the training data, which gets reproduced on demand during task execution, and GitHub’s Copilot model provides the clearest example of this effect. The algorithm is trained on production-quality code stored in its database, where the best superstar engineers are overrepresented in the training data. Inexperienced software engineers are unlikely to get their code into the production codebase, and so the training data is likely of very high quality. This enables novice coders to draw upon all the knowledge of their superstar colleagues and competitors, made accessible by the LLM. GitHub Copilot thus reduces the competitive advantage of superstars by forcibly redistributing their knowledge and expertise to augment the performance of those less skilled.

This observation has enormous implications for inequality, as much of it has been driven by a widening skill gap. The share of economic profits accruing to “superstar” firms has risen 300% since the late 1990s. According to recent estimates from leading economists, such “superstar” effects are responsible for much of the increase in inequality in recent years.¹⁰ And this is what makes GenAI so unique in modern economic history. David Autor, MIT’s world-leading economist researching how technology impacts inequality, argues GenAI will help rebuild the middle class thanks to the way these models help close the skills gap.¹¹ This is in sharp contrast to most recent digital innovations, which tend to exhibit a low share of labor value-added and thus have contributed to widening inequality.

There are good reasons to believe GenAI will have a positive human impact by augmenting rather than substituting for people, supporting social inclusion and economic growth simultaneously.

The views reflected in this article are the views of the author and do not necessarily reflect the views of the global EY organization or its member firms.