Dr. Ayse Ozturk

A significant question circulating about AI nowadays is this: Is AI helping everyone catch up, or is it merely allowing top performers to widen the gap?

Depending on which study you read, you’ll either become more optimistic about AI’s democratizing powers or more nervous, thinking whether your CV/resume still matters in an age of AI. Let’s see what some studies so far say.

---

At first, AI was a skill equalizer

Initially, the research suggested that it was good news for low performers. Studies showed AI helped low performers the most in writing, coding, or customer service. It was like giving everyone a superpower.

Papers by Brynjolfsson et al., Peng et al., and Noy & Zhang showed that when customer support agents or college students were given access to GPT-style tools, low performers improved the most. Their writing got better. Their code got cleaner. Productivity gains were largest at the bottom. It was a democratizing tool for those who’d been left behind by traditional education or the knowledge economy.

These studies made a strong case that AI equalized the skill distribution by standardizing quality and reducing the importance of specialized skills.

---

Then, it got complicated

However, the findings became more complicated. Two recent papers raise questions about the egalitarian optimism.

1. Otis et al. (2023) ran a five-month field experiment with 640 Kenyan entrepreneurs using GPT-4 via WhatsApp. It showed that high performers saw a 20% boost in revenue. Low performers actually did worse, seeing only a 10% boost. Why? Because they asked AI about more complex, make-or-break challenges. Meanwhile, high performers were asking AI how to expand already profitable shops. Obviously, the latter is easier for AI to answer and implement.

2. Toner-Rodgers (2024) studied AI in scientific R&D, showing that top researchers nearly doubled their output, while those in the bottom third saw little improvement. Why? Because AI took over “idea generation” tasks (i.e., creating a set of hypotheses), and obviously, researchers had to be good at picking the best ideas from AI’s suggestions. As a result, top researchers with domain expertise did a better job of leveraging AI output.

In both cases, the low-skilled users weren’t lazy or tech-averse. They used the AI. But they either asked it to solve problems it wasn’t good at or lacked the judgment to filter good from bad suggestions.

---

So Which One is AI: An Equalizer or a Divider?

The conclusion can be summarized as:

• If you give AI to workers performing well-defined, modular tasks (such as writing, customer service, or coding), it’s likely to benefit everyone, particularly those who are low performers. In those cases, AI becomes a substitute.
• If you give AI to users facing complex decisions (like running a business or selecting novel research directions), then domain expertise becomes essential. In those cases, AI becomes a complement to skill, not a substitute.

And that’s the key part: AI can be either an equalizer or a divider. It depends on what you ask it to do.

---

Implications

• If you’re a policymaker: Be wary of assuming that AI tools will automatically reduce inequality. They might exacerbate the issue unless paired with training and support.
• If you’re a teacher: Teach your students how to ask good questions—and how to critically assess AI answers. Prompting and judgment matter more than ever.
• If you’re a founder or researcher: Don’t just hand your team AI tools. Help them understand when to trust them and when to override them.
• And if you’re wondering whether AI will replace you? The answer seems to be: It depends on whether you know what to do with its answers.

---

Ultimately, AI may not replace humans. However, it appears to be widening the gap between those who can use it wisely and those who cannot.