Researchers found a way to make AI learn intuitive physics like a human
Scientists taught a deep learning system to learn intuitive physics the same way human babies do.
Why it matters
This mechanism could be key in bridging the gap between humans and AI, as well as inform future psychology studies about cognition.
By definition, “human intuition” seems to denote a barrier between us and artificial intelligence. It’s why we have gut feelings and reflexive — sometimes impulsive — reactions inexplicable by logic, and therefore not simply transferrable to computers. I mean, we can hardly parse our own reasoning for instinctive behavior, so how could we develop algorithms to encode it?
In collaboration with AI research laboratory DeepMind in the UK, this team developed an artificial intelligence system that learned “intuitive physics,” that is, commonsense understanding of how our universe’s mechanics work, just like a human baby.
It’s named Physics Learning Through Auto-encoding and Tracking Objects, or PLATO — undoubtedly a nod to the Greek philosopher famous for his allegory of the cave, a thought experiment that probes the nuanced nature of knowledge and meaning.
“Current artificial intelligence systems pale in their understanding of intuitive physics, in comparison to even very young children,” the study authors wrote in their paper. “Here we address this gap between humans and machines by drawing on the field of developmental psychology.”
What’s intuitive physics?
If you were to show a baby a red ball, then block it with a large book, the child might be a little shocked at first. She might wonder, “Uh, did that red ball just…disappear?” But if she sees this situation happen enough times, she’ll eventually realize, “Oh, it’s still there even though I can’t see it. Stuff doesn’t just disappear randomly. We have physics!”
This is called object permanence, and between birth and age two, it starts to meld into what we consider our intuition.
Fast-forward to adulthood, when an item is blocked from our view, we don’t ever deliberate the fact it’s still there. We just know. And the new study’s team wanted to help PLATO get to the point where it just knows physical stuff like that. Intuitive physics.
Here’s how everything went down.
Basically, the study team first perused decades of developmental psychology research about how babies learn intuitive physics. Slowly, after reading through that literature, a shared theme began to emerge — “the idea that physical understanding is supported by breaking the world into a discrete set of objects,” Luis Piloto from DeepMind said in a press conference Monday.
In other words, babies seem to learn intuitive physics by observing objects move around, fall down, interact, appear and disappear. Gotta see it to believe it, you might say. Zeroing in on that principle, the researchers developed a deep learning model, which is a system based on massive datasets that can sort of gain skills over time and therefore adjust its own code. This is PLATO.
Then, the team showed PLATO 28 hours of animated videos about simple physics that involved lots of objects.
For instance, PLATO watched a ball falling to the ground or rolling behind other objects — and even “impossible” scenarios that defied the laws of physics. Things like objects moving through each other. Scenarios you might find in a magician’s handbook.
Algorithmic power, NATO and Artificial Intelligence
NATO has formally approved its first Artificial Intelligence (AI) strategy as it seeks a leading position in the adoption of AI for defence, but it may face some critical hurdles ahead in implementing the strategy, according to Simona Soare.
NATO defence ministers have formally adopted the Alliance’s first artificial intelligence (AI) strategy. The document lays out six ‘baseline’ principles for ‘responsible’ military use of AI – lawfulness, responsibility and accountability, explainability and traceability, reliability, governability, and bias mitigation. It also provides an insight into key implementation challenges.
The strategy is meant to provide a ‘common policy basis’ to support the adoption of AI systems in order to achieve the Alliance’s three core tasks – collective defence, crisis management and cooperative security. The strategy is also designed to challenge established Alliance processes for procurement, technology development and wider engagement with the private sector and academia.
Only a summary of the strategy has been made public. However, it reveals four critical obstacles to implementation that NATO will face: reconciling the objectives of member nations; securing sufficient political and financial support; bridging any disconnect between the Alliance’s policy and operational units; and managing the transnational bureaucracy that will implement the strategy.
As well as being a consensus-building policy document, the strategy attempts to position NATO as the leader of AI adoption in defence. It reiterates the allies’ commitment to transatlantic cooperation on the development and use of AI in security and defence, an important element of which is ensuring inter-operability and standardisation.
There are still hard questions, however, about how NATO will coordinate different national approaches to managing the development and application of AI in defence, combined with restrictions on technology use, access, sharing and transfer. For countries like the United States, it is a priority that allies agree practical guidelines for the operational use of AI-enabled systems and the necessary data-sharing, a challenge that should not be underestimated. Some allies, meanwhile, are not satisfied with the granularity of the six principles of responsible use, while others consider that overemphasising the normative approach risks ceding technological advantage to peer competitors.
Similar tensions are playing out in the European Union. The EU’s proposal for an AI act is more restrictive for high-risk, high-impact applications of AI, though its impact on defence will be indirect, as it do does not apply to the military domain. In the defence realm, the European Defence Agency’s Artificial Intelligence Action Plan for Defence shares more similarities with the NATO strategy. While the plan is not public, it reportedly includes a list of use cases for military applications of AI which member states may consider for collaborative development and principles of responsible development and use.
Another question that remains to be answered is the extent of NATO’s ambition to adopt AI. The strategy is meant to be implemented in a phased approach, partly to build political support for AI military projects. Initial ambitions seem modest, reportedly focusing on mission planning and support; smart maintenance and logistics for NATO capabilities; data fusion and analysis; cyber defence; and optimisation of back-office processes. As political acceptance grows and following periodic reviews of the strategy’s implementation, the goal is to also include more complex operational applications.
Finally, the AI strategy runs parallel to NATO’s Military Strategy, a military-led process launched in 2019, and its Warfighting Capstone Concept, which examines alliance requirements in future operating environments. However, the AI strategy is a stand-alone document. To avoid creating narrow implementation tracks, meaningful early engagement between NATO’s policy and military communities would be beneficial to cut across any disconnect between threat-based assessments of the impact of AI on military capabilities and politically driven processes for the development and use of AI.
The executive summary of NATO’s AI strategy does not reflect any alignment of the roles and resources of the different NATO and national innovation bodies. It is unclear from the summary how the NATO Innovation Unit, Allied Command Transformation, the Science and Technology Organization and the NATO Communications and Information Agency will coordinate to implement the strategy.
The Alliance aims to exploit AI developments in the commercial sector by adopting an open innovation model and deliberately moving away from its present procurement model. However, this will require an effort to map out the relationship between old structures, such as the NATO Industrial Advisory Group, and new engagement channels with the private sector, such as the Defence Innovation Accelerator for the North Atlantic and others created by the AI strategy.
While NATO has adopted the AI strategy, there is no dedicated line of funding for it. Finance will depend on a combination of common budget funding and off-budget mechanisms such as the NATO Innovation Fund. Besides the uncertainty over the availability of funding, some Alliance agencies are concerned that their budgets could be cut and redistributed towards the implementation of the AI strategy. The allies have set a USD1 billion target for the NATO Innovation Fund. However, whether this amount is sustainably generated and distributed over the long term, and by what means, is more important for encouraging innovation than the announced figure.
The promise of AI for military applications has been clear for some time; less obvious is the route to deliver on it. For all the implementation challenges it faces, the Alliance’s AI strategy represents a step in the right direction.
Weaponization of Neuroscience
Hervé Le Guyader
This essay is part of the author’s contribution to the NATO Operations 2040 study by the NATO Innovation Hub.
While it has been said that everything could be weaponized, neurosciences and, more broadly speaking, Nanotechnology, Biotechnology, Information Technology and Cognitive Sciences (NBIC) are clearly providing state and non-state actors some true game changers.
The story narrated in this essay begins in 2018 with weak, and not so weak signals, and ends in 2040 with NATO triggering Article 5 because of NBIC attacks on some of its allied Nations. During these 22 years, pivotal decisions are taken at NATO Summits, fundamental choices are made for the design of the successor to the Alliance’s main surveillance and control system, and NATO manages to embark a large number of nations, far beyond its core allied nations, into a pragmatic educational program on global security.
All of this because of the “Weaponization of neurosciences” challenging topic that was to be addressed.
This essay uses fiction and mixes actual facts and events, fairly logical foresights and some fictitious extrapolations drawn from a couple of long term key geostrategic initiatives launched by today’s big players. Of course, the roles played in this story by those big players could be interchanged, albeit with some work.
Using a few dramatization tricks, at the cost of being a bit provocative to try and keep the reader’s interest doesn’t mean not being serious at voicing out one’s deep beliefs.
In this particular case:
Yes, “Human mind” should be NATO’s next domain of operation,
Yes, AWAC’s successor must address NBIC,
Yes, global security is what’s at stake today, and it will take more than professionals of the defense, security and military sectors to address it efficiently.
However difficult it will be.
Brussels, July 17, 2026, NATO Summit: “Human mind”, the 6th domain of operation
Excerpt from the Brussels Declaration, issued by the Heads of State and Government participating in the meeting of the North Atlantic Council in Brussels 16-17 July 2026.
Article 11 … To stay secure, we must look to the future together. We are addressing the breadth and scale of new technologies to maintain our technological edge, while preserving our values and norms. We will continue to increase the resilience of our societies, as well as of our critical infrastructure and our energy security. To effectively do so, NATO and Allies, within their respective authority, must constantly take stock of the pace and breadth of scientific research being conducted, in particular outside the Alliance. Nanotechnology, Biotechnology, Information Technology and Cognitive Sciences (NBIC), whose development rate is staggering, have an immense potential to deeply transform our societies, but the dual nature of this potential poses a new set of challenges to our security.
For decades, NATO and Allies, and our competitors too had been used to operate in a three-dimensional environment, where air, land and sea represented familiar, distinct but interoperable operational context.
The 2014 Wales Summit identified that Cyber-attacks presented a clear challenge to the security of the Alliance and could be as harmful to modern societies as a conventional attack. By way of consequence, NATO and Allies agreed that cyber defense was part of NATO’s core task of collective defense.
The 2016 Warsaw Summit then recognized cyberspace as a domain of operations in which “NATO must defend itself as effectively as it does in the air, on land, and at sea”.
Three years later, the 2019 London Summit declared, in the article #6 of its final declaration, Space as an operational domain for NATO, recognizing its importance in keeping us safe and tackling security challenges, while upholding international law. Of note, the same article also stated “We are increasing our tools to respond to cyber-attacks, and strengthening our ability to prepare for, deter, and defend against hybrid tactics that seek to undermine our security and societies. We are stepping up NATO’s role in human security. We recognize that China’s growing influence and international policies present both opportunities and challenges that we need to address together as an Alliance.”
Progresses in NBIC make it today possible for our competitors to develop new ways to reach their offensive objective. While propaganda and influencing strategies have always existed, the depth and sophistication of NBIC-fueled hybrid attacks today represent an unprecedented threatening level inasmuch they target the most vital infrastructure we rely on: the human mind.
Norfolk, we have a problem As it turned out, preparation to the 2026 Summit had not been as exhaustive and sturdy as it should have been. Years of under-budgeting and under-staffing had taken their toll and Article 11, the “Mind hacking” article, was one key casualty. Enthusiastic accolades were shared and raucous applause heard across the world but, soon enough, once the dust had settled, impartial observers were prompt to identify two main fault lines: • IC, and notNBIC While Article 11 had correctly presented NBIC, as a whole, as being the issue to address, only one and a half (or thereabout) out of its four components had been in reality looked into with the necessary rigor: Information (technologies) and their own, specific capacity to tamper with human Cognition. But Nano, Bio technologies, and their own impact on Cognition (hence the 1.5 vs. 2.5 approximation) had, in reality, been put on the backburner. • Doctrine? Rules of engagement? Training? DOTMLPFI17? Under public and diplomatic pressure, NATO had managed to reach consensus among nations on this fairly disruptive concept of Human mind as a domain of operations and to pull off a unanimous decision but, unlike the five first domains of operation, “NBIC warfare against human mind” was pretty much terra incognita, most certainly in terms of lessons learned. People had been fighting for hundreds of years on land, at sea, for a little bit more than a century in the air, for a few decades in cyberspace and space. Historians, scientists, defense specialists, military and civilian experts and practitioners had built considerable knowledge regarding wars waged over land, sea and air. More recent conflicts had added Cyber and even Space warfare data and analysis to the mix, and dozens of exercises, executed at the coalition (NATO) level had allowed for all concerned parties to optimize their readiness level. But human mind as domain of operation??? What’s the equivalent to the “smoking gun”, how can it be detected, identified, attributed to … something, somebody??? Where is my OODA loop??? My C4ISR?, What are the ad hoc CCIR18’s??? And then, the real killer issue: What would cause triggering Article 5? Adversaries and competitors were merciless in overtly mocking NATO’s apparent unpreparedness, stressing the “existential risks this “marketing rather than strategic” decision was creating for the human race”. More covertly, troll farms19, fake news factories20 and 50 Cent Army21 worked double, triple, quadruple shifts to make sure gullible (remember Weapons of Mass Cretinization?) folks would go down streets and avenues around the world with new, anti-NATO slogans. Ironically, the same time pressure that had prompted NATO to issue its declaration in 2026, in an admittedly rushed out fashion, ended up also applying to its competitors who, in turn, made a series of bad moves that ended up in “incidents”. Epitomizing the NBIC threat and serious enough in their disastrous.
We therefore recognize the human mind and bodey as a domain of operations in which NATO defend itself as effectively as it does in the air, on land, at sea, in cyberspace and in space.
Annex: Recommendations This is an annex to the main “Weaponization of neurosciences” essay, aimed at providing some recommendations related to the three main points summarized in its conclusion. Going too far without interacting first with NATO, based on its reaction to the paper, would probably be unrealistic and useless, so here are some fairly concrete recommendations for the two first main points raised by the essay. The third point raises some strategic and geopolitical issues that clearly need to be better appreciated in order to provide plausible recommendations. Point # 1: Human mind as NATO’s 6th domain of operation. Reaching that level may be a long shot but, whether or not that objective is achievable, the reality of the human mind hacking threat is undeniable and NATO must react in a concrete manner, and do it quickly. The code name proposed for NATO’s response is: “Human mind hacking: Light, camera, action!”, a three-year project. • Light: Because it is a developing and complex subject, Human mind hacking needs light being shed on it to be made clearer and more decipherable. This will start with an exhaustive state of the art study addressing the nature, plausibility, development of that threat, together with an impact assessment of attacks already perpetrated. That particular task may be coordinated by the Innovation Hub. Evidence gathering, structuration of the study do not raise any particular issue and can be distributed among several military and non-military int’l partners, but particular attention must be given to the quality of the deliverables so that they lend themselves well to the two next steps of NATO response. This is a 10-month effort, going from April 2020 till February 2021. Updates every six month. • Camera: Because the relevance and potential impact of the Human mind hacking issue address the full gamut of stakeholders, from leaders to first responders involved in complex, hybrid crises, from their awareness and understanding of the situation to decision-making process, cameras (figuratively speaking, of course) are needed to capture and broadcast in the most efficient manner the takeaways from the study summarized above, and to do it with messages customized to targeted audiences. While this effort must start immediately (April 2020) and be sustained for the whole duration of the project with regular updates to the material that will be generated, a first production of communication material will have to be out by September 2020. • Action: Led by ACT, and starting in April 2020, this third pillar to the project has two primary objectives o As an in itinere work package, from Month 1 and for the three years’ duration of the initial effort: setting in motion the production of the entire DOTMLPFI and coordinating its progresses, o As an immediate priority: Make sure that each and every exercise, wargame scenario, training material … includes Human mind hacking material generated by the (“light” and “camera”) two other components of the project. Weaponization of neurosciences, HLG, ENSC, February 2020 33 Point # 2: Allied Future Surveillance & Control (AFSC) This is obviously a major project for NATO in terms of strategic importance and in terms of budget. Even if the current AWACS has benefited from many updates along its existence, AFSC’s design faces unique challenges because of the complexity of today’s conflicts (see main report re: hybrid, complex warfare …), let alone the exponential growth of (NBIC) technologies. In other words, AFSC, in whatever shape/s or form/s this “system of systems” will take, will epitomize the depth and sophistication of NATO’s understanding of tomorrow’s conflicts. I am convinced that addressing all possible threats is a vital necessity. My recommendation is to extend that mind set to the whole design process. Point # 3: Security is not merely a military issue. Global security is a society issue. To develop that point and come out with concrete recommendations capable of providing some added value and not merely “state the obvious” would necessitate a better understanding of how NATO and chiefs of government, but also NATO and large international institutions work together and craft common agendas. One point, though, goes without saying: the communication material put together by the “Human mind hacking: Light, camera, action!” project needs to be designed with these partners in mind. Considering the geographical and political breadth of this issue, this is probably the most challenging point of the “Weaponization of neurosciences” recommendations to address
There is no established legal protection for the human subject when researchers use Brain Machine Interface (cybernetic technology) to reverse engineer the human brain.
The progressing neuroscience using brain-machine-interface will enable those in power to push the human mind wide open for inspection.
Facebook is building brain-computerinterfaces
“Do you want to work for the company who pioneered putting augmented reality dogears on teens, or the one that pioneered typing with telepathy?” You don’t haveto say anything. For Facebook, thinking might be enough.
Facebook hired Dugan last year to lead its secretive new Building 8 research lab. She had previously run Google’s Advanced Technology And Products division, and was formerly a head of DARPA.
Facebook built a special Area 404 wing of its Menlo Park headquarters with tons of mechanical engineering equipment to help Dugan’s team quickly prototype new hardware. In December, it signed rapid collaboration deals with Stanford, Harvard, MIT and more to get academia’s assistance.
Elon Musk and the goal of human enhancement
Brain-computer interfaces could change the way people think, soldiers fight and Alzheimer’s is treated. But are we in control of the ethical ramifications, extending the human mind …
Soon afterwards, the serial entrepreneur created Neuralink, with the intention of connecting computers directly to human brains. He wants to do this using “neural lace” technology – implanting tiny electrodes into the brain for direct computing capabilities.
There is call for alarm. What kind of privacy safeguard is needed, computers can read your thoughts!
In recent decades areas of research involving nanotechnology, information technology, biotechnology and neuroscience have emerged, resulting in, products and services.
We are facing an era of synthetic telepathy, with brain-computer-interface and communication technology based on thoughts, not speech.
An appropriate albeit alarming question is: “Do you accept being enmeshed in a computer network and turned into a multimedia module”? authorities will be able to collect information directly from your brain, without your consent.
This kind of research in bioelectronics has been progressing for half a century.
Brain Machine Interface (Cybernetic technology) can be used to read our minds and to manipulate our sensory perception!
Man is now made whether we want it or not into a commercial biomechanical platform.How is that possible? This is due to the lack of general knowledge in the field of nanotechnology, new networking technology, artificial intelligence and machine learning.
Mathematical models and enormous computing capacity in the cloud as well as
artificial intelligence make this possible.