Meatbags vs machines: DeepMind plans hackathon to draw line between human and AI brains
(2026/03/18)
- Reference: 1773852676
- News link: https://www.theregister.co.uk/2026/03/18/google_deepmind_agi_hackathon/
- Source link:
If a bot actually achieved artificial general intelligence (AGI), how would we even know? Google DeepMind boffins have come up with what they say is an empirical, scientifically grounded framework to measure progress toward AGI, and they're looking for a few good devs to actually flesh it out.
Once upon a time, "artificial intelligence" was used to describe a machine that was essentially indistinguishable from humans on a wide variety of thinking-like tasks. But as machine learning applications, starting with OpenAI's ChatGPT, captured the public imagination and the flywheel of tech industry growth-hype spun into hyperdrive, the term "AI" was defined down to mean computer programs that use mass-scale matrix multiplication to perform complex tasks, reasonably well, with relatively little oversight. With the movement of these goalposts, the loosely defined term "AGI" has taken the place of what AI used to mean.
DeepMind wants to tighten up that definition. The team at the Google-owned AI research and development shop [1]reported this week that they had developed a "cognitive taxonomy" for measuring the tech industry's progress toward universally useful AGI, along with a three-stage test for benchmarking AI system performance against human capabilities.
[2]
For those hoping for some groundbreaking psychological insights or ideas here, sorry, but what the researchers are proposing is simple. Run AI models and humans through the same cognitive benchmarks, say DeepMinders, and you'll get a good estimate of when a single AI is able to meet or exceed human capabilities on all ten areas of its taxonomy, which are divided into two main areas, as the team describes in a [3]paper .
[4]
[5]
First up are eight basic building blocks of human cognition, which have been previously defined by other researchers: Perception, generation, attention, learning, memory, metacognition, and executive functions.
Those eight building blocks combine in various ways, say the DeepMind researchers, to form two composite faculties that are equally important: Problem solving, and social cognition, which is defined in the paper as the ability to process and interpret social information and respond appropriately in social situations.
[6]
DeepMind's concept of what AI capabilities mapped onto the 10 areas of its taxonomy, and against human performance - Click to enlarge
A taxonomy is all well and good, of course, but without systems in place to test how AI models perform against humans it doesn't really do much. So the Google team has proposed a hackathon to draft the community to help.
"We are launching a new Kaggle hackathon," the team explained, "to design evaluations for five cognitive abilities where the evaluation gap is the largest: learning, metacognition, attention, executive functions and social cognition."
[7]AI gone rogue: Models may try to stop people from shutting them down, Google warns
[8]Google DeepMind minds the patch with AI flaw-fixing scheme
[9]AI superintelligence is a Silicon Valley fantasy, Ai2 researcher says
[10]Worry not. China's on the line saying AGI still a long way off
A $200,000 prize pool has been established for [11]the contest , and several entries are already posted publicly and in progress. Two teams in each of the five areas will be awarded $10,000 each, and four overall winners will each be awarded $25,000.
AGI is widely believed to be a [12]long way off at this point, and some experts have declared it a fantastical [13]waste of time altogether. There's not even a clear agreed-upon definition, beyond being an AI that's able to perform well across subject areas. The DeepMind team didn't bother to clarify much about what they thought AGI meant beyond saying that it's "often used as a shorthand to describe various kinds of highly capable AI systems" that we need to actually start measuring progress toward while we dither on definitions.
[14]
The team hopes that, by doing something - anything - to help measure progress toward AGI, it can "move the conversation around AGI from one of subjective claims and speculation toward a grounded, measurable scientific endeavor."
Winners of the hackathon will be announced in June. ®
Get our [15]Tech Resources
[1] https://blog.google/innovation-and-ai/models-and-research/google-deepmind/measuring-agi-cognitive-framework/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/measuring-progress-toward-agi/measuring-progress-toward-agi-a-cognitive-framework.pdf
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://regmedia.co.uk/2026/03/18/deepmind-cognitive-taxonomy-map.jpg
[7] https://www.theregister.com/2025/09/22/google_ai_misalignment_risk/
[8] https://www.theregister.com/2025/10/07/google_deepmind_patches_holes/
[9] https://www.theregister.com/2025/12/11/ai_superintelligence_fantasy/
[10] https://www.theregister.com/2025/03/05/boffins_from_china_calculate_agi/
[11] https://www.kaggle.com/competitions/kaggle-measuring-agi
[12] https://www.theregister.com/2025/03/05/boffins_from_china_calculate_agi/
[13] https://www.theregister.com/2025/12/11/ai_superintelligence_fantasy/
[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[15] https://whitepapers.theregister.com/
Once upon a time, "artificial intelligence" was used to describe a machine that was essentially indistinguishable from humans on a wide variety of thinking-like tasks. But as machine learning applications, starting with OpenAI's ChatGPT, captured the public imagination and the flywheel of tech industry growth-hype spun into hyperdrive, the term "AI" was defined down to mean computer programs that use mass-scale matrix multiplication to perform complex tasks, reasonably well, with relatively little oversight. With the movement of these goalposts, the loosely defined term "AGI" has taken the place of what AI used to mean.
DeepMind wants to tighten up that definition. The team at the Google-owned AI research and development shop [1]reported this week that they had developed a "cognitive taxonomy" for measuring the tech industry's progress toward universally useful AGI, along with a three-stage test for benchmarking AI system performance against human capabilities.
[2]
For those hoping for some groundbreaking psychological insights or ideas here, sorry, but what the researchers are proposing is simple. Run AI models and humans through the same cognitive benchmarks, say DeepMinders, and you'll get a good estimate of when a single AI is able to meet or exceed human capabilities on all ten areas of its taxonomy, which are divided into two main areas, as the team describes in a [3]paper .
[4]
[5]
First up are eight basic building blocks of human cognition, which have been previously defined by other researchers: Perception, generation, attention, learning, memory, metacognition, and executive functions.
Those eight building blocks combine in various ways, say the DeepMind researchers, to form two composite faculties that are equally important: Problem solving, and social cognition, which is defined in the paper as the ability to process and interpret social information and respond appropriately in social situations.
[6]
DeepMind's concept of what AI capabilities mapped onto the 10 areas of its taxonomy, and against human performance - Click to enlarge
A taxonomy is all well and good, of course, but without systems in place to test how AI models perform against humans it doesn't really do much. So the Google team has proposed a hackathon to draft the community to help.
"We are launching a new Kaggle hackathon," the team explained, "to design evaluations for five cognitive abilities where the evaluation gap is the largest: learning, metacognition, attention, executive functions and social cognition."
[7]AI gone rogue: Models may try to stop people from shutting them down, Google warns
[8]Google DeepMind minds the patch with AI flaw-fixing scheme
[9]AI superintelligence is a Silicon Valley fantasy, Ai2 researcher says
[10]Worry not. China's on the line saying AGI still a long way off
A $200,000 prize pool has been established for [11]the contest , and several entries are already posted publicly and in progress. Two teams in each of the five areas will be awarded $10,000 each, and four overall winners will each be awarded $25,000.
AGI is widely believed to be a [12]long way off at this point, and some experts have declared it a fantastical [13]waste of time altogether. There's not even a clear agreed-upon definition, beyond being an AI that's able to perform well across subject areas. The DeepMind team didn't bother to clarify much about what they thought AGI meant beyond saying that it's "often used as a shorthand to describe various kinds of highly capable AI systems" that we need to actually start measuring progress toward while we dither on definitions.
[14]
The team hopes that, by doing something - anything - to help measure progress toward AGI, it can "move the conversation around AGI from one of subjective claims and speculation toward a grounded, measurable scientific endeavor."
Winners of the hackathon will be announced in June. ®
Get our [15]Tech Resources
[1] https://blog.google/innovation-and-ai/models-and-research/google-deepmind/measuring-agi-cognitive-framework/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://storage.googleapis.com/deepmind-media/DeepMind.com/Blog/measuring-progress-toward-agi/measuring-progress-toward-agi-a-cognitive-framework.pdf
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://regmedia.co.uk/2026/03/18/deepmind-cognitive-taxonomy-map.jpg
[7] https://www.theregister.com/2025/09/22/google_ai_misalignment_risk/
[8] https://www.theregister.com/2025/10/07/google_deepmind_patches_holes/
[9] https://www.theregister.com/2025/12/11/ai_superintelligence_fantasy/
[10] https://www.theregister.com/2025/03/05/boffins_from_china_calculate_agi/
[11] https://www.kaggle.com/competitions/kaggle-measuring-agi
[12] https://www.theregister.com/2025/03/05/boffins_from_china_calculate_agi/
[13] https://www.theregister.com/2025/12/11/ai_superintelligence_fantasy/
[14] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44abraMTYNZFHy2KAPcRf0uwAAAAs&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[15] https://whitepapers.theregister.com/