$200K for the Hardest AI Test: How Google DeepMind Is Redefining the Path to AGI

$200K for the Hardest AI Test: How Google DeepMind Is Redefining the Path to AGI

In 2026, Google DeepMind introduced a structured framework to measure progress toward AGI, backed by a $200,000 Kaggle challenge focused on testing complex cognitive abilities — especially social intelligence.

What happened: a shift from talk of AGI to measurable metrics

Google DeepMind has offered a practical approach to one of the industry's most elusive topics: assessing progress toward AGI (artificial general intelligence).
Instead of debating "what AGI is," researchers have focused on measurement. A hackathon with a $200,000 prize pool was launched in partnership with Kaggle , challenging participants to develop tests to evaluate complex cognitive abilities in AI.
The key idea is that if intelligence cannot be clearly defined, it can be systematically measured.

$200K for the Hardest AI Test: How Google DeepMind Is Redefining the Path to AGI

New cognitive model: 10 abilities instead of one indicator

The study draws on psychology, neuroscience, and cognitive science. Based on this foundation, 10 areas were identified that form the basis of general intelligence.
These include perception, learning, memory, reasoning, and problem solving. Particular attention is paid to social cognition—the ability to understand the context of communication and respond appropriately to it.
This component is considered one of the most complex for modern models.

Most modern systems work well with logic, text, and data. However, social situations require a different level of information processing.
Context, intentions, emotions, ambiguity—all of this makes social intelligence tasks fundamentally more complex.
DeepMind is focusing on precisely this level of complexity. Verification goes beyond answer accuracy—it also evaluates the appropriateness of behavior in various interaction scenarios.
This brings AI testing closer to real-world use conditions.

How the assessment protocol works

The proposed evaluation system includes three stages.
First, the models are tested on a broad set of cognitive tasks. Then, similar tasks are performed by humans from a representative sample. In the final stage, the results are compared, allowing us to understand how close the AI ​​is to human performance in each skill.
Importantly, the comparison is not made with an abstract "average person," but with the distribution of results. This makes the assessment more accurate and scientifically sound.

Hackathon: An attempt to turn theory into a tool

The "Measuring progress toward AGI: Cognitive abilities" competition focuses on five of the most challenging areas: learning, attention, metacognition, executive functions, and social cognition.
Participants create benchmarks that are then tested on real models through the Kaggle Community Benchmarks platform.
The prize structure reflects a focus on quality: the best solutions receive funding, but most importantly, the opportunity to influence future industry standards.
With the rapid development of AI, the market is facing a lack of universal metrics. Model performance is often assessed based on narrow tasks, which does not reflect the true level of intelligence.
DeepMind's approach could become the basis for standardization. If the industry adopts unified evaluation criteria, this will change
model development,
investment decisions, and
competition between companies.

In fact, we are talking about creating an “intelligence metric” for machines.

For financial markets, such initiatives have an indirect but significant impact.
Technological breakthroughs in AI are strengthening the position of companies investing in research, impacting US stock markets and global capital flows.
In 2026, the artificial intelligence sector will remain a key growth driver. The emergence of clear metrics could accelerate the reallocation of investment toward the leaders.
For Forex, this means a strengthening of the influence of the technological factor on exchange rates through investment flows and macroeconomic expectations.

Why AGI remains uncertain

Despite its progress, DeepMind has not provided a definitive definition of AGI. The term continues to be used as a general term for general-purpose systems.
Estimates for the timeline for the emergence of AGI remain controversial. Some experts believe it's still a long way off, while others point to rapid progress.
In this context, the new taxonomy is an attempt to capture at least intermediate guidelines.

From a practical standpoint, the key moment will be when the system shows human-level performance or higher in all ten cognitive areas.
This does not automatically mean that AGI will appear, but it will create a measurable point from which to evaluate progress.
This approach reduces uncertainty and makes AI development more transparent.
Google's DeepMind initiative reflects a significant shift in the industry: from abstract discussions of AGI to concrete metrics and benchmarks. The attempt to measure intelligence through a set of cognitive abilities could become the foundation for a new stage of AI development. In a context of growing competition and investment, such standards will determine which technologies truly approach general intelligence and which remain narrowly specialized solutions.
Written by Ethan Blake
Independent researcher, fintech consultant, and market analyst
March 23, 2026

Join us. Our Telegram: @forexturnkey
All to the point, no ads. A channel that doesn't tire you out, but pumps you up.
FX24

Author’s Posts