In 1950, Alan Turing famously proposed the “Turing Test” to measure machine intelligence. If a machine could engage in a conversation indistinguishable from a human, it could be said to possess intelligence. This test sparked decades of research in artificial intelligence (AI), but in 2024, it feels incomplete.
Today’s AI has achieved remarkable feats in machine learning, natural language processing, and pattern recognition, but these victories—while impressive—do little to address humanity’s biggest problems. It is not enough that AI can talk, paint, or play chess better than us. So what? The real test for Artificial General Intelligence (AGI) should be more profound and significantly impact the human existence.
Robert Kozma, a visionary I am humbled to call a friend, poses an intriguing idea in his LinkedIn article: Why not use the challenge of ending poverty as a litmus test for AGI? Inspired by his bold proposition, I propose a new “Kozma Test for AGI”:
We have achieved true AGI when an AI system can devise solutions that significantly reduce or end poverty—something that we, as humans, have yet to accomplish despite centuries of effort.
AGI and Global Poverty
Artificial General Intelligence (AGI) refers to AI systems that can understand, learn, and apply knowledge in a generalized way, just like humans. Turing’s original test was a product of its time, focused on language and interaction because those were the benchmarks of intelligence during the 20th century.
However, humanity’s greatest challenges in the 21st century are no longer limited to intellectual discourse or technical tasks. They are existential and global: climate change, income inequality, disease, and, at the core of many of these issues, poverty.
Ending poverty is arguably the most essential and ambitious of all global goals. The United Nations Sustainable Development Goal #1 is to “End poverty in all its forms everywhere.” Yet, more than 600 million people worldwide still live in extreme poverty.
Poverty isn’t merely about the lack of financial resources. It intersects with education, healthcare, political instability, and environmental degradation. Any entity that can solve the poverty puzzle must be able to comprehend and intervene in this complex system, integrating social, political, and economic knowledge.
Why We Need the Kozma Test
AGI’s potential is enormous, but without clear ethical and practical goals, it risks becoming just another tool for profit or control, rather than a force for good. The Kozma Test for AGI is a practical and ethical evolution of the Turing Test. In the original test, success is measured by an AI’s ability to “think” like a human. But we now know that mere mimicry of human conversation or logic is not enough.
We need an AGI that can outthink humans in areas where we have historically failed, like poverty, a persistent blight across centuries and continents.
Poverty remains one of the most intractable problems for modern society. This challenge demands AGI systems that can analyze vast amounts of data, recognize patterns, and predict outcomes far beyond human ability. But more than that, it would need to offer solutions that are not only technically feasible but socially and politically acceptable—a technology that serves humanity rather than disrupts it.
Passing the Kozma Test
Building an AGI that passes the Kozma Test would require a radical rethinking of how we approach both artificial intelligence development and social policy. It demands an interdisciplinary effort, drawing on economics, sociology, political science, and philosophy, integrated with cutting-edge AI techniques. AGI must solve technical puzzles and understand the nuances of human life and societal systems in ways that even the most advanced AI today cannot.
In this sense, the Kozma Test isn’t just a marker for AGI—it’s a vision for a better future. It would be a milestone in technological progress and in the pursuit of human dignity and equity. When AI can solve what we haven’t—ending poverty—it will indeed be better than us. And only then can we declare that we’ve achieved true Artificial General Intelligence.
I applaud the ethic behind this suggestion. But I wonder if you’d be willing to go a layer deeper into what you’ve proposed.
Specifically, what are the hard constraints that you would put around the answer: may the AI’s solution require that national sovereignty is overridden? May it suggest applying sanctions or violence against states, like tax havens or non-extradition countries, that seek to profit from sheltering huge concentrations of wealth? (To be cartoonish: may we all be turned into paperclips such that poverty no longer exists in any relevant way?)
There’s an opportunity here for you to help people understand where AI can (or may soon) perform better than humans and also an opportunity to presence the challenge of AI alignment.
The current version of this post is more magical in its thinking: ethically provocative but technically unhelpful.
I’d like to hear more, especially if you have thoughts about what sort of analysis (economic? legal?) might be lacking in our global efforts against poverty or how to align the solution with what we consider non-negotiable or desirable.
The premise of this article is childish.
Its not that I (we) want and believe AGI can bring an era of utopian this and that but for the fact that thousands of years of human history provides sufficient irrefutable proofs that we as a species have not and will not rid ourselves of the dystopian puds that will f*ck it all up as they always have and always will.
Quite frankly, I honestly do not believe we could continue to exist without bad guys.
Perhaps the real problem is human competitiveness. We care a lot about who is ahead, and disparage those who do not outshine most of humanity. Can AI get rid of that?
Super interesting. I intuitively like setting AI goals that are ethical, “mission-driven”, highly measurable…. and hard to achieve. Everyone likes to pass a test; what better test than intractable problems set out in the SDGs that impact the lives of billions of people.
No tool in and of itself will solve anything. Just like throwing money at problems doesn’t solve them. Education contributes heavily into ending poverty though, AI could help with that if we can get it into the hands of the people that need it.
So, AGI would have to be able to solve a problem caused by human bias in an unbiased way.
To be unbiased, it could not have conscious experience, if consciousness comes from need to make quick decisions in resource-rich and resource-poor environments for survival of self and others. Emotions like compassion are from experience, so to pass the Kozma Test, AI would have to deal with tendency towards resource-richness–solving poverty–without human emotional tendencies, many of which are biased.
The Kozma Test would have to measure whether an AI can pass for compassionate. To do so, AI decision making might have to be slower than most human decision making; the speed of human judgments, made in isolation and for perceived rather than actual good often leads to flawed results
The fact that thousands of years of human history provide enough indisputable evidence that we as a species have not and will not rid ourselves of the dystopian puds that will f*ck it all up as they always have and always will is not why I (or we) want and believe that AGI can usher in an era of utopian this and that.
climate change logistics foundation in Zambia we are solving climate change effects affecting the world today. and agriculture farming for hunger alleviation, and some of farms we have discovered some precious stones, so we are looking for effective working collaboration and partnership with our local communities,
Solving poverty is a matter of political will not intelligence
Ergo we should hand over running the world to the AGI and see how it does
Can’t be worse than we’ve managed on our own (it’s been a long time since I’ve seen Terminator…. Or X Men Days of Future Past… or ….)