Why the future of AI depends on human insight

Artificial Intelligence (AI) has advanced at extraordinary speed, reshaping industries, redefining productivity, and embedding itself in modern business infrastructure. For startups, in particular, AI is now both a competitive necessity and a key differentiator.

But beneath this momentum hides a structural challenge that the industry is only beginning to confront: AI is running out of meaningful data.

For over a decade, progress in AI has followed a simple formula – more data, more compute, better outcomes. That approach has delivered remarkable results: modern systems can generate content, analyse markets, and automate complex workflows at scale. Yet this progress has relied heavily on access to vast quantities of human-generated data.

That supply is drying up.

The data saturation problem

The pool of high-quality, human-created data is finite – and increasingly saturated. The same datasets are being reused across competing models, while AI-generated content is beginning to flood the internet.

Researchers and industry leaders have warned that this trajectory is unsustainable. Some estimates suggest that publicly available high-quality training data could be exhausted within the next few years, potentially as early as 2026. Others point to an imbalance between demand and supply, with the need for large, diverse datasets now outpacing the availability of suitable data.

Meanwhile, the scale of data consumption continues to accelerate. Training compute for frontier language models has been growing at 5× per year since 2020. This exponential demand only intensifies the pressure on already limited resources.

The consequence is not just scarcity but diminishing returns. Companies have increasingly found that simply adding more data does not guarantee better performance, particularly when models are repeatedly trained on similar or lower-quality inputs. In extreme cases, this can lead to what is known as ‘model collapse’, where systems degrade as they learn from their own outputs rather than human data.

For startups, this is a critical strategic inflection point. If every company is training on the same information, then AI becomes a commodity, not an edge.

The next frontier: human intelligence as data

If the first phase of AI was defined by access to information, the next phase will be defined by access to insight.

Human intelligence is not just about knowledge. It is shaped by context, experience, and the ability to navigate ambiguity. These are precisely the qualities that current AI systems lack, and they are also the least represented in traditional datasets.

Today’s models are highly effective at recognising patterns, but they still struggle with judgement, reasoning, and context; the core components of real intelligence.

This points to a fundamental shift in how we think about data. Instead of relying solely on static content such as text, images, and historical records, we need to capture how people think. That includes reasoning processes, trade-offs, emotional responses, and situational judgement.

These are not easily scraped from the internet, but they are significantly more valuable.

This is the thinking behind our flagship project, Humanix. It introduces a model where humans are not passive data sources, but active, compensated contributors. By capturing human reasoning, it becomes possible to generate entirely new forms of training data, which reflects not just what people do, but why they do it.

For startups, this creates a new opportunity. Those who can access or generate this richer layer of data can build systems that are more adaptive, context-aware, and competitive.

The controversy: how far should we go?

This shift is not without its challenges.

Teaching AI to think and behave more like humans raises important questions. If human reasoning becomes a primary input into AI development, what happens to the value of human creativity? How do we ensure that contributions are recognised and compensated?

See Also

How VFOs are helping small businesses unlock growth

There are also concerns about control. As AI systems become more capable – particularly if they incorporate elements of human judgement and improve continuously – the line between tool and decision-maker becomes less clear.

These concerns are valid. But avoiding them is not a solution.

AI is already influencing decisions, shaping markets, and embedding itself into critical systems. The question is not whether we continue to develop more advanced AI, but how we do so responsibly.

In my view, the answer lies in maintaining a clear distinction between AI capability and human control.

Advancing AI does not require relinquishing human oversight – it demands more human involvement. By positioning humans as active participants in AI development – not just as users, but as essential contributors – we create a system where control stays firmly with humans.

This fixes a deeper imbalance within today’s AI ecosystem. Right now, human input is treated as a passive resource, extracted without visibility or compensation. A more participatory model not only improves AI performance but also ensures that value is distributed more fairly.

A defining moment for startups

We are entering a new phase of AI development where the rules are changing.

The advantage will no longer go to those who simply scale existing models, but to those who rethink how AI learns. Data alone is no longer enough. The next breakthrough will come from integrating human intelligence itself.