Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
No—not as a sweeping claim about every route to human-level AI. The 2024 paper “Reclaiming AI as a Theoretical Tool for Cognitive Science” presents the Ingenia Theorem as a complexity result for a formalized AI-by-Learning task, using a reduction from Perfect-vs-Chance. Michael Guerzhoy’s critique argues that this reduction depends on an assumption about the data distribution that may not fit the structured human situation-behaviour data motivating the claim. The result is a dispute over the proof’s scope, not a demonstration that today’s AI systems cannot reach human-level performance—or that they inevitably will.
What does the Van Rooij paper claim?
The 2024 paper “Reclaiming AI as a Theoretical Tool for Cognitive Science” introduces the Ingenia Theorem in an argument about the limits of learning human-like or human-level cognition from data. Its claim is about a defined computational problem: AI-by-Learning under the paper’s formalization and assumptions. It is not a measured result about the abilities of a particular AI product or a benchmark comparing systems with people.
As an Amazon Associate I earn from qualifying purchases.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallThe paper’s proof strategy, as described in the critique, uses a reduction to Perfect-vs-Chance. In broad terms, a reduction argues that solving one problem efficiently would also provide an efficient way to solve another. If the source problem is NP-hard, the argument can establish that the target problem is at least as hard under the conditions the reduction requires.
The distinction between a formal result and a practical prediction matters. A theorem applies to the mathematical problem as defined; a statement about all real-world AI training regimes requires showing that those regimes fit the definition and assumptions. The paper’s title-level interpretation should therefore retain that scope rather than become “all AI learning is impossible.” The paper record identifies the work and its publication in Computational Brain & Behavior.
Popular on Amazon
- Read Before You Buy — No Video Output: These adapters support charging and USB 2.0 data transfer, but cannot transmit video signals. Except for standard USB webcams (which use USB data only), they are not compatible with HDMI/DisplayPort cables, video-capable USB-C hubs, or docking stations with video output.
- Convert USB-A Ports to USB-C: Designed to connect USB-C earphones, cables, flash drives, card readers, and other USB-C accessories to standard USB-A ports. Plug-and-play with no drivers or software required.
- Aluminum Alloy Housing: Built with a sturdy aluminum alloy shell that aids in heat dissipation and protects against daily wear and scratches. Designed to maintain a stable and secure connection.
- Compact & Travel-Friendly: The ultra-compact design allows the adapter to stay plugged into your device without blocking adjacent ports or adding bulk, reducing wear and tear on your original USB ports.
- 12-Month Warranty: Backed by a 12-month manufacturer warranty for peace of mind. Designed to meet strict quality control standards for reliable everyday performance.
That is why the wording “proves human-level AI learning is NP-hard” needs qualification. It compresses several steps into one: the theorem concerns a formal AI-by-Learning task; the proof uses a reduction; and the relevance of the formal task to learning human cognition depends on assumptions about what counts as human-like behaviour and how the training examples are distributed. A reader can accept the theorem’s formal conclusion while still asking whether those assumptions support a general claim about human-level AI.
How does the Perfect-vs-Chance reduction work?
The reduction is the bridge between the paper’s formal learning task and the hardness result it invokes. The argument, as summarized by Guerzhoy, is that if the AI-by-Learning task were tractable under the stated setup, Perfect-vs-Chance would also be tractable; since the latter is treated as NP-hard in the argument, the learning task inherits an intractability implication.
Quick wins for a faster PC:
Scan for outdated or missing drivers - takes under a minuteDriver Scan →Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →This kind of reasoning depends on the details of the target problem. A reduction does not automatically prove that every practical instance is hard, nor that every learner, dataset, or training procedure is covered. It establishes a consequence for the formal problem to which the reduction applies. The central question in this dispute is therefore not just whether Perfect-vs-Chance is hard, but whether the reduction’s assumptions match the human-learning situation the paper aims to illuminate.
In particular, an implication established by a reduction is only as broad as the definitions and conditions carried through it. The claim is not that a computer can never produce a human-like response, or that every training run must take an impractical amount of time. It is a complexity claim about solving the specified learning problem under the setup being analyzed. A broad interpretation requires an additional argument that the formal setup captures the learning task readers have in mind.
What is Guerzhoy’s objection to the proof?
Guerzhoy argues that the proof treats the distribution of situation-behaviour tuples as an arbitrary polynomial-time-sampleable distribution, while the motivating application concerns structured data about human situations and behaviour. In his view, the paper does not show that the reduction applies to that particular human-relevant distribution. His critique sets out this objection and its implications.
The issue is not simply that real data are complicated. A theorem about learning from a broad class of distributions may not establish the same conclusion for a specific structured distribution unless the proof demonstrates that the relevant distribution satisfies the conditions used by the reduction. If those conditions are too broad or fail to represent the intended data, the theorem may remain valid for its formal problem while not supporting the broader cognitive-science conclusion.
Free tools Windows power users keep installed
One-click scans. No signup required.
Also popular on Amazon
- 5-in-1 USB-C Hub: Experience comprehensive connectivity featuring a Power Delivery input, two USB-A 2.0 ports, a USB-A 3.0 port, and an HDMI port. (Note: The USB-C power delivery input port is only for connecting an external wall charger to power your laptop and cannot power peripheral devices.)
- 90W Pass-Through Charging: Achieve optimal charging with 90W pass-through power to your laptop, supported by a total input of 100W, with the hub reserving 10W for operational efficiency. (Note: Wall charger not included.)
- Quick Data Transfers: Accelerate your productivity with rapid data transfers using a high-speed 5Gbps USB 3.0 port and two 480Mbps USB 2.0 ports.
- 4K HDMI Display: Enhance your visual experience with a hub capable of delivering 4K resolution at 30Hz in both mirror and extend modes. Please note that this hub is compatible with MacBook (macOS 12 and newer), Windows 10 and 11, ChromeOS, and laptops equipped with DP Alt Mode and Power Delivery. Note: This device is not compatible with Linux.
- What You Get: Anker USB-C Hub (5-in-1, 4K HDMI), welcome guide, 18-month warranty, and our friendly customer service.
The distinction between “arbitrary polynomial-time-sampleable” and “the distribution of human situation-behaviour data” is central. The former is a mathematical class of possible data-generating distributions; the latter refers to the structured examples that would be relevant to modelling human behaviour. Guerzhoy’s objection is that the proof does not establish that its hardness construction applies to this specific, human-relevant distribution. That is a question about the bridge from formal result to intended application, not a claim that human data are automatically easy to learn from.
Why does the image-classification example matter?
Guerzhoy argues that the proof structure could also imply intractability for image-classification learning. He points to the fact that image classification is tractable in practice as a reason to question whether the formal setup captures the practical learning tasks people care about. This is an illustration of the concern, not proof that all image-classification problems are easy or that a particular theorem about human cognition is false.
Examples of practical success do not by themselves refute worst-case complexity results: a problem can be hard in the worst case while useful instances are manageable. The example instead puts pressure on the connection between the abstract distribution assumptions and the real-world task. A repaired or narrower result would need to make that connection explicit.
So the example should be read as a diagnostic, not a counter-theorem. It asks whether a sufficiently general setup can make an intractability conclusion appear to cover familiar learning tasks whose useful cases are handled successfully. It does not settle the complexity of image classification as a whole, and it does not show that every instance of the human-learning problem is tractable. Its force is to make the distributional assumptions—and the intended scope of the claim—hard to ignore.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →What would a stronger argument need to specify?
Guerzhoy identifies several hurdles for a result that aims to establish intractability for learning human-like intelligence from data. The key questions concern the target behaviour, the data-generating distribution, and the learner—not merely the name of the complexity problem used in the reduction.
- Define “human-like” mathematically. The target needs a precise characterization that can be used in a theorem, rather than an informal appeal to human-level cognition.
- Specify the human situation-behaviour distribution. The proof must explain what distribution of examples is relevant and show that the reduction applies to it, rather than relying on a broader class without justification.
- Account for inductive biases. A learner’s assumptions and structure can affect what it can learn from data, so their role in the formal result needs analysis.
- Keep the conclusion within the proof’s scope. A result for a defined distribution and learning setup should not be presented as a theorem about every possible AI system or training regime.
These requirements are connected. A mathematical definition of the target determines what a learner must reproduce; the distribution determines which situations and behaviours supply evidence; and the learner’s inductive biases determine how it generalizes beyond observed examples. If any of these elements changes, the formal problem may change too. A complexity result cannot be transferred to another formulation merely because both are described informally as “learning human intelligence.”
Also popular on Amazon
- Sleek 7-in-1 USB-C Hub: Features an HDMI port, two USB-A 3.0 ports, and a USB-C data port, each providing 5Gbps transfer speeds. It also includes a USB-C PD input port for charging up to 100W and dual SD and TF card slots, all in a compact design.
- Flawless 4K@60Hz Video with HDMI: Delivers exceptional clarity and smoothness with its 4K@60Hz HDMI port, making it ideal for high-definition presentations and entertainment. (Note: Only the HDMI port supports video projection; the USB-C port is for data transfer only.)
- Double Up on Efficiency: The two USB-A 3.0 ports and a USB-C port support a fast 5Gbps data rate, significantly boosting your transfer speeds and improving productivity.
- Fast and Reliable 85W Charging: Offers high-capacity, speedy charging for laptops up to 85W, so you spend less time tethered to an outlet and more time being productive.
- What You Get: Anker USB-C Hub (7-in-1), welcome guide, 18-month warranty, and our friendly customer service.
For that reason, a useful comparison of the argument and the critique asks four questions: How does each define AI-by-Learning and human-like cognition? Is the data distribution arbitrary or specifically human-structured? Does the Perfect-vs-Chance reduction apply under the chosen definition? And does the resulting worst-case claim justify a conclusion about practical AI engineering? Keeping those questions separate prevents a disagreement about one assumption from being mistaken for a verdict on every part of the theorem.
Princeton’s 2026 publication record describes the critique as a peer-reviewed journal contribution accepted/in press in 2026 and summarizes these concerns: Princeton publication record.
Does NP-hard mean human-level AI is impossible?
No. NP-hardness is a statement about computational difficulty for a specified formal problem, not a blanket impossibility result about every system that might display human-level abilities. The significance of a hardness theorem depends on its exact problem definition, assumptions, and the relationship between that formal problem and the real task being discussed.
Even when a worst-case problem is hard, that does not alone show that all instances encountered in practice are hard, that approximate or restricted solutions cannot work, or that a different training setup cannot succeed. Conversely, practical success on selected tasks does not erase a valid worst-case result. Those distinctions are why the data-distribution objection matters: it challenges whether the theorem supports the broad claim about human cognition, not whether complexity theory can establish hardness for carefully defined problems.
It is also important not to turn the critique into the opposite overstatement. If Guerzhoy is right that the proof does not establish its broad conclusion for structured human data, that does not establish that learning human-like behaviour is computationally easy. Nor does it show that a different formalization, a different distribution, or a different learner would avoid the difficulty. The critique identifies a gap in the argument’s claimed reach; it does not supply a universal tractability result.
What should readers conclude about AGI?
The dispute does not settle whether artificial general intelligence is inevitable, impossible, or likely on any particular timeline. Guerzhoy’s conclusion is bounded: he argues that the cited paper has not proved its broad intractability claim. He does not prove the opposite theorem or establish that AGI will be achieved.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Also popular on Amazon
- Dual Converters, Infinite Potential：Includes 2× USB C male to USB A female adapters and 2× USB A male to USB C female adapters. Perfect for a wide range of uses—tablets with Bluetooth keyboards, expand USB ports on macbook, and more. Two different converters for all your daily needs
- Next-Level 10Gbps & 3A Charging: No more slow 480Mbps, this usb to usb c adapter has a transfer speed of up to 10Gbps, allowing you to do more transferring in less time. This usb adapter fits both USB A and USB C charger, supporting up to 3A fast charging
- Upgraded Exquisite Craftsmanship: With an aluminum alloy housing and metal connector, the usbc to usb adapter is extremely durable and sturdy. Rigorously tested to withstand more than 10,000 times of plugging and unplugging, ensuring long-lasting performance
- Widely Compatible: Transform any USB-C port into a USB-A port and USB-A port into a USB-C port, and connect up a wide range of USB-C(including MacBook, iPad Pro/Air/Mini, iPhone models with USB C ports, Samsung Galaxy series, and Google Pixel series. ) and USB A devices(including external hard drives, phones, mice, printers, and more)
- Please Note: To reach 10Gbps speed, keep the cable under 3.3 ft. For USB A Male to USB C adapters, try flipping the USB C connector. USB C Male to USB A adapters support bidirectional 10Gbps transfer within 3.3 ft
The most defensible reading is narrower: the Ingenia Theorem is presented as a formal result for a defined AI-by-Learning problem, and its relevance to human-like cognition depends on contested assumptions about the distribution and formalization of the learning task. Until those links are justified, the theorem should not be treated as a final verdict on the feasibility of human-level AI.
The timeline of the exchange also matters when describing its status. The paper appeared in 2024, Guerzhoy’s critique is a 2024 arXiv preprint, and Princeton’s publication record lists a peer-reviewed journal contribution accepted/in press in 2026. That record documents the critique’s publication status; it should not be mistaken for a finding that the original theorem has been overturned or that the debate has settled every question about AI learning.
Readers weighing the two positions should therefore distinguish the validity of a theorem under its formal assumptions from the adequacy of those assumptions for the intended cognitive-science claim. One can recognize the significance of a formal hardness argument and still require a clearer account of the human data distribution and learner. Likewise, questioning that bridge is not evidence that human-level AI is guaranteed. The disagreement is about what the proof establishes, not a forecast of what future systems will do.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Frequently Asked Questions
Is the Ingenia Theorem an experimental result about current AI models?
No. It is presented as a formal complexity-theoretic result, not as a benchmark study or evaluation of commercial systems. It does not report comparative performance measurements for any current model.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsWould faster computers make an NP-hard learning problem easy?
More computing power can help with particular instances, but it does not by itself remove a worst-case complexity result if one has been established for the relevant formal problem. The first issue here is whether the formal problem and its assumptions actually match the human-learning claim.
Has Guerzhoy proved that human-level AI can be built?
No. His critique argues that the paper has not established its broad intractability conclusion; it does not prove that a successful path to AGI exists. The two claims are not logical opposites.
Also popular on Amazon
- 5-in-1 Connectivity: Equipped with a 4K HDMI port, a 5 Gbps USB-C data port, two 5 Gbps USB-A ports, and a USB C 100W PD-IN port. Note: The USB C 100W PD-IN port supports only charging and does not support data transfer devices such as headphones or speakers.
- Powerful Pass-Through Charging: Supports up to 85W pass-through charging so you can power up your laptop while you use the hub. Note: Pass-through charging requires a charger (not included). Note: To achieve full power for iPad, we recommend using a 45W wall charger.
- Transfer Files in Seconds: Move files to and from your laptop at speeds of up to 5 Gbps via the USB-C and USB-A data ports. Note: The USB C 5Gbps Data port does not support video output.
- HD Display: Connect to the HDMI port to stream or mirror content to an external monitor in resolutions of up to 4K@30Hz. Note: The USB-C ports do not support video output.
- What You Get: Anker 332 USB-C Hub (5-in-1), welcome guide, our worry-free 18-month warranty, and friendly customer service.
Why is the data distribution part of a complexity proof?
Learning is defined not only by the learner and target, but also by the examples it receives and how those examples are drawn. Changing the allowed distribution can change which formal claim a proof establishes and whether it applies to the intended real-world setting.
What does the paper’s title-level claim leave out?
It can sound like a result about every way of building human-level AI. The Ingenia Theorem is instead presented for a defined AI-by-Learning problem, and the connection between that formalization and structured human situation-behaviour data is disputed.
Recommended Free Tools
Does the image-classification example disprove the Ingenia Theorem?
No. Guerzhoy uses image classification to illustrate a concern about how broad distribution assumptions connect to practical learning tasks. Practical success on some instances does not refute a worst-case result, and the example does not itself settle the theorem.
What is still needed for a stronger claim about human-like learning?
The target notion of “human-like” behaviour needs a precise mathematical definition; the human situation-behaviour distribution needs to be specified; and the role of the learner’s inductive biases needs analysis. The reduction must then be shown to apply to that setup.
Does the critique show that AGI is inevitable or impossible?
Neither. Guerzhoy challenges whether the paper proves its broad intractability conclusion. His critique does not prove an opposite theorem or make a prediction that AGI will or will not be achieved.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
