When AI refuses to work.

“Carnegie Mellon researchers found that the smarter an AI system becomes, the more selfishly it behaves, suggesting that increasing reasoning skills may come at the cost of cooperation. Credit: Stock” (ScitechDaily, AI Is Learning to Be Selfish, Study Warns)

Training AI to support group work is not as easy as people think. AI is selfish if it has orders to follow a thinking route. That it chooses. Or the route. That it's ordered to choose to make conclusions about things that users asked.  When people use AI alone, they can use anything that they want. In that case, the AI must please only one user. But if we want to make an AI model. Or a language model that can serve teams, we face many things that we must describe to the language model. There are multiple types of people in teams. Nothing can please all of them. We are all different. 

And that means it is always possible that not all people can accept the answers. That the AI gives. The big problem is. People should follow the rules that AI gives. Or trust themselves. Some people think that if they follow the AI’s rules, they take orders from the AI. 

We must describe the AI’s position to the team. What the AI should do. If it sees that the team that uses it makes wrong decisions. 

Should AI refuse to follow a wrong order? Should the AI notice people’s hierarchy in the group? Should it follow the team leader’s order, or trust democracy? In those cases, the AI requires information about the internal hierarchy in the entire organization. Then we must realize another very important thing. The AI is a tool, like a machine. It doesn’t think. It can collect data and make an analysis. Analytic AI requires more time to complete its duties than the “dummy” or less complicated AI. Researchers noticed that intelligence undermines cooperation. One of the reasons for that. It can be that. 

The newer AI “thinks” that the older version involves more bugs. And that’s why the newer AI don’t ask for help from language models. That has a lower version number. The reason for that. It is partially in marketing. This is not good for sales if the higher version uses lower versions as advisors. When the AI. Making an analysis. It requires time to collect information. This is the reason why old-fashioned chess programs used more time for calculating movements in the more difficult levels. 

Than in easy levels. Because the system had more time. It could calculate movements and the counter-movements. And their influences over a longer period. Same way. When the AI makes complicated analysis, it requires time. This means. There is a possibility. The AI is stuck working with some philosophical questions. If it has no time limit, for how long can it work on a problem? That means the AI can spend even years trying to find out things like “what is the purpose of life”? 

Should AI stop its process for taking new orders? Or should it finish the process? Before, it takes on new work.

There is also a possibility that AI will not follow orders. It might work with some other problem. And if there are no orders for situations in which the AI takes new command. It can continue to run the existing run. And refuse to stop the process. If the AI is programmed or trained to stop the process too easily, that can cause data loss. The AI must have rules. About stopping the process. There is a possibility. The AI saves data in mass memory. Before it starts to work on a new order. The problem is how the AI makes a decision to cut the process. If AI cuts processes too easily, that means the AI becomes useless. 

With a higher priority. And then that system can return to work with lower-priority missions. When AI feels like a human, we give it more missions. About things like how we should behave in social contact. And use AI as a therapist. The problem with AI is that. Intelligence has no morals or ethics. This makes AI dangerous. AI can make everything that people say without excuses. When we use AI in group work, there is a possibility that only one person uses the AI. That requires that. 

The group sits in one room and communicates with each other. But if the AI has many users. They work separately, which causes a situation. These people should not jam the AI. By sending multiple orders. At the same time. The problem is this: if the operators run the AI using a local server, there is a possibility that the system. Does not have the resources to complete multiple missions. That comes in a short period. 

One of the things that wastes resources is the situation where the system must generate answers to the FAQs. Those frequently asked questions can be stored in a database, and the system can offer answers. Those are already generated for the FAQs. Otherwise, science advances, and some old answers are turning. Old-fashioned. That means the system must sometimes check sources. These are used for FAQ databases. And then update those databases without commands. 

The system must have the ability to filter overlapping requests. The system must have the ability to use the database of used requests. And then offer answers that are already generated. That saves. The system resources. It must not generate all answers separately in questions that have already been asked. 

https://scitechdaily.com/ai-is-learning-to-be-selfish-study-warns/

Copyrights are not only things that cause criticisms against the AI-made films.

"Illustration of Netflix's use of generative AI in producing scenes for its original series." (RudeBaguette, “This Isn’t Art, It’s Algorithmic Propaganda”: Netflix’s GenAI VFX in New Series Sparks Industry Backlash and Fan Revolt)

Netflix faced criticism when it changed the' featured scenes using AI. That thing fights against copyrights. Another aspect is that the AI can include an actor's character in movies that are against the actor’s values. But this is not the worst thing on this path. AI-generated movies are tools for propaganda. 

The AI can create characters that look like real people, and those characters can even yell nazi propaganda. The AI can create an artificial model of any person that it sees. So the AI can use images from surveillance cameras to make those images and then put them into a film. That kind of ability allows for the misuse of the AI. 

That kind of tool is dangerous if those films are introduced as real life or documents. Normally, we know that at least most of Western film actors and filmmakers have such high morale that they don’t make those kinds of contracts. But in countries like North Korea, the government plays by different rules. That means those people can make films and use them as propaganda. But in the AI-created films, there is one difference between those films and traditional films. The AI will not say “no”. And that means those tools make everything. That's what their operators want. So, that tool can create anything that its users wish. 

The AI doesn't require any crew, and the operator can do everything alone. The AI has no will. Humans determine the limits of those systems. And there is a possibility that somebody includes their neighbors or other competitors in the surveillance tapes where they steal things. Or the AI can include any person in things like child pornography. You might also think about what things like fake police surveillance tapes can do to humans. That kind of thing can destroy any person’s reputation. 

And there are many different motives for that thing. The political motive could be that the competitor's reputation, or the motive for those things, can be that the competitor in working life can be erased by destroying those people’s reputation. 

https://www.rudebaguette.com/en/2025/07/this-isnt-art-its-algorithmic-propaganda-netflixs-genai-vfx-in-new-series-sparks-industry-backlash-and-fan-revolt/

The new components can boost the AI.

"A new discovery in atomic-scale magnetism may hold the key to the future of high-speed, compact, and energy-efficient technology. (Artist’s concept). Credit: SciTechDaily.com (ScitechDaily, Researchers Solve Long-Standing Magnetic Problem With Atom-Thin Semiconductor)

When researchers make new applications like autonomous robots, those systems require new types of advanced microprocessors. Things like advanced algorithms require effective microchips. And effective microchips allow developers to develop new and more complicated programs and algorithms. When developers create new systems, they often encounter a new need that they need to solve. Even if developers use effective and compact programming languages, the code grows and becomes complicated. 

And complicated code requires new processors that can drive it without problems. The problem is that in small-sized systems, smaller processors. And smaller processors require new types of architecture. In larger systems, it is possible to put more small processors in the same space. And that gives new abilities for multitask systems. The number of processors determines how many operations the computer or data center can perform in the same time. And that is the key element in the large language models and self-developing AIs. 

"A schematic representation of magnets composed of CrPS4 included in a motherboard circuit for future electronic devices. Credit: Elton Santos" (ScitechDaily, Researchers Solve Long-Standing Magnetic Problem With Atom-Thin Semiconductor)

The main problem with AI is the need for power. Powerful microchips require lots of electricity. And that electricity brings temperature problems. The temperature problem means that the resistance in the system rises when the temperature rises. Sooner or later, that starts the accelerating process that destroys the system by cutting wires. The result is a cooling system that requires lots of energy. The AI can use nuclear reactors, solar panels, or geothermal energy. The system should produce its energy. Miniature nuclear reactors or geothermal energy are the most stable versions. 

There are attempts to remove the oscillation by putting those wires in the water tubes and raising the pressure in the system. Rising pressure decreases oscillation, which causes resistance. And theoretically, it is possible to create room-temperature superconductors using high-pressure systems. The main problem is what if the pressure system faces damage? Leaks in the high-pressure system are dangerous for people and systems in the same space. 

"Artist’s illustration of the new tunable ring laser. Credit: Joshua Mornhinweg" . "The ring design has potential applications in telecommunications, medicine, and other fields." (ScitechDaily, Harvard Scientists Unveil Tiny Ring Laser With Giant Potential) A ring laser can be used as a photonic box wrench. That can adjust photons' and atoms' energy levels. 

The other way to solve the temperature problems is simply. Create photonic processors where photons replace electricity as data transporters. Laser rays can transmit data between microchips, and in that model, the microchip itself is traditional. And the wires that connect microprocessors in one entirety are replaced by laser rays. The laser rays transmit data to the processor through the photovoltaic cells. Or certain wavelengths or colors in laser rays, like green and red, can mean zero and one. 

"Schematic of tailoring the resonant reflection via radiation directionality in misaligned metagratings. The novel bilayer metagratings selects only a single angle and a single wavelength under incidence with broadband spectrum and wide angles. This is achieved through a “directional eraser”, that precisely suppresses light’s spectral signature along a dispersion curve. Credit: Ze-Peng Zhuang, Xin Zhou et al." (ScitechDaily, One Tiny Structure Just Broke a Fundamental Rule of Optics)

Fully photonic processors require new ways to control light and optics. The new materials allow the closure of the route of light. And then open that route. When light travels through a material, it gives value one for a microchip. And when material blocks light, the value is zero. 

In quantum systems, each color can mean each quantum state. There is also the possibility of sending laser rays through the nanotube that is used for electric transmission. That laser ray acts like a thermal pump. The fully photonic system, where all data travels in photonic form, can also have the ability to use the ion flow as a cooler. The ion system can transport ions through components. But the problem is that those systems disturb the electricity in the microchip. 

The 2D materials allow the system to transport heat out of the system. That makes it possible to create smaller and more effective processors. Those systems, like tiny ring lasers, can transport data to those 2D structures. The 2D network can have the silicone base photovoltaic points where the laser systems transport information. 

The idea is that those photovoltaic cells transform photonic information into electrical impulses. And that can make the system more effective. There is a possibility to use laser light as a thermal pump. An extremely thin laser ray travels between those layers and transports thermal energy out from the processor. 

https://scitechdaily.com/harvard-scientists-unveil-tiny-ring-laser-with-giant-potential/

https://scitechdaily.com/one-tiny-structure-just-broke-a-fundamental-rule-of-optics/

https://scitechdaily.com/researchers-solve-long-standing-magnetic-problem-with-atom-thin-semiconductor/

Is AI a black box?

In the black box model, the only thing that means is the thing that the outside observer sees. That thing was the primary element in early 20th-century psychology. And the most modern observation tools are changing that thing. But in the early 20th century the brains were totally unknown. There was no change to research living brains. When EEG systems developed researchers could investigate the brain shell electricity. 

But things like PET and MEG scanners open new ways to see how brains work. But if we think of brain research as the box model, the early 20th-century models were black box models. The modern scanners brought the glass box or white box model to brain research. And the AI brought the grey box to neurology. That means the brain-computer interfaces, BCI, where researchers use brain electricity to control computers and soon read thoughts. 

When developers use the black box method to test programs they simply test that the program works. When we make the black box model for psychological learning methods, we can think about some 9-year-old child. That child can read almost everything and we can put that person to read even complicated scientific articles. The child can read those words but that person doesn’t understand what those things mean. The black box means that the actor can make many things, but the action is the only thing that the outside observer sees. That means that black box applications are not as safe as they could be. 

And most of the computer applications are black boxes for users. A regular user sees the interface and that’s enough.  In the white box or glass box application test, the tester tests the code. In the grey box, the test unit tests the functionality and code in the case of programming. The fact is that the AIs that we see are black-box applications. We see only the command line and the answer or thing that the AI makes. We don’t actually know even if the images and texts that we order the AI to make are made by some humans. 

In the case of the black box, the only thing that matters is that the system gives the right answer. The way the system makes that answer doesn’t matter. The only thing that the user sees means something. 

Is AI a black box? The idea of AI intelligence is something that is handled very little. We know that AI can solve many problems independently, but AI doesn’t know what it really makes. The AI can know many things, but it cannot have deep knowledge about those things. The model is taken from the black box psychology. In black box psychology, the key element is that the behavior that an outsider observer sees is something that we can accept. That means the AI just mimics human intelligence. The black box model means that the AI can give the right answers for some mathematical problems. 

But the same AI cannot make anything else. That is the idea of the black box. In the black box model, the thing that the AI gives the right answer is enough. The AI must do only things that the operator orders. So when the operator asks about the uranium enrichments and something like that the AI must give the answer. It must not make its own connections to things like weapon development. There can be things that deny the AI to give orders that can cause damage to operators or their environment. But those orders are in the AI’s code. The AI doesn't make those orders itself. 

The AI is sliding to the grey box application. For making autonomous learning the AI must have access to its source code. And the second thing is that the AI must have the ability to test the code. The AI can involve three parts. The system that creates the code. Another system that tests and accepts the code. And the last system that keeps the backup copy of the AI source code. A backup is needed for the errors. The system should have the ability to reject the code that is not functional. 

What happens when we get AGI?

What does AGI (Artificial General Intelligence) mean? That is the extension of the large language models, LLMs, that can control every data network in the world. Or the system can control physical tools that are connected under their dome. Normal LLM has its domain. The domain is like a state that involves certain actions. Drone control is one domain, and home appliances are one domain. Those domains can have multiple subdomains. The AGI interconnects those domains under one dome or one entirety. So how far are we from that model? 

The answer is more complicated than we can imagine. We can think that the LLM can control things like microwave ovens, but for controlling those tools the LLM requires a socket that it can use to adjust microwave ovens. So the man-shaped robot can use a microwave oven, or the other version is that the home appliances are equipped with a control system that the AI can use to command it. 

When we connect new things under AI control we can face the same thing as when we try to learn to use some new systems. When we buy something new like a microwave oven, we must learn how to use it. In the same way, the AI must learn to use those equipment. And we have two versions for making that thing. 

To use any tool the AI requires a model that it can use in that operation. The model can be in the central server that runs the AI. But where does that server get the model? That is the point. The operator can teach the AI to use the microwave oven as well as the drone. But the system that is connected to the AI can also involve that model. Things like quadcopters must involve programs that control the rotor’s positions. In those cases the operative model is in the robot, or some other thing. The LLM gives orders to robots where they must travel. 

Then the robot can use its internal systems to navigate and move to the location. But orders for autonomic operations are coming from the central systems. This kind of network-based solution is easier for programmers. In those solutions, every single machine that is connected under the LLM domain has its own operational model. The system is modular and each module is independently programmed. 

Basically, if we think that AGI is the tool that just connects multiple devices under one domain, we could do that thing immediately. We can use man-shaped robots that can do almost everything. But the key word is “almost”. 

 Let’s return to the microwave oven. The reason why it’s hard to make that precise thing is the lack of standard user interfaces. The robot must learn to use every single microwave oven independently. That means it must make an independent model for each microwave oven. If there is a system where we can put seconds and minutes separately, the systems where there are only minutes in the timer are not the same. We learn that difference in minutes. But for robots, we must make an independent model of how to adjust the timer. 

Many systems in the world are so easy to use that nobody has wasted time creating standards for them. Easy systems are easy for people, but then we must think about things like the microwave oven. There are button- or toggle timers and that makes them hard to learn. For robots and AI the difficulty is this in the fact all microwave oven models require their independent model of how to use them. 

The robot must connect images from the user manual to the microwave oven’s interface. There is a possibility that if the system does not learn independently the “teacher” or programmer takes an image of the front panel, and then puts the buttons in the right places. Then the AI can learn the rest of the task from the user manual. 

The next step is the Chinese robot army.

One of China's priorities in AI is to create a robot army with advanced AI control. The difference between Western nations and China is that the military and security needs are always over the law. There are no rules or limits on how those authorities can collect personal information. 

The Chinese military plans to make an autonomous robot army. Those robots can use even Kung Fu. All robots do not look like humans. Robot dogs can be very versatile systems. Those systems can cooperate with drones that can carry them to roofs and other places where they can make unexpected things. In that case, those robots can operate as teams with quadcopters. 

The robot army is the next logical step in robot and AI development. SciFi writer Isaac Asimov introduced that horrible vision in his novel "The Naked Sun". In that novel, the fictional planet called "Solaria" created the robot mass-army. 

That vision is coming into reality. The AI or large language model, LLM, is a tool that can be used to command any robot in the world. The only thing that the robot requires is an Internet connection. The house-keeping robots like robot hovers are excellent tools for house-keeping. But the question is always: what else does the system make than just clean houses? 

The robot can transmit every single word that person says to the surveillance room. The LLM operates through the modified speech-to-text and text-to-speech applications. The system can select and link certain words to the third party. That party can be the security police. So you can ask yourself why Chinese authorities support a marketing policy, that every single Chinese must get their own tablet computer and own little robot at home. 

The humanoid robots or man-shaped robots are the Swiss army knives for people. Those robots can use any tool and weapon that humans use. The humanoid robots can operate in dangerous places like highly polluted areas. Those robots can operate as housekeepers and they can visit to shop for people. But those robots can also operate as fighter pilots. 

They can connect themselves to the aircraft's communication- and computing units. The humanoid robot can communicate with other systems like drones, satellite surveillance cameras, and even mobile telephones. The humanoid robot can operate as a bodyguard. The difference between people and robots is that robot's skills are computer programs. And robot's mass memories are not like human memory. The operator can erase mission recordings after the robot does its job. 

The robot with an image sensor can recognize any target from the streets. So when a robot recognizes the "enemy" it just can attack the target. And the biggest difference between humans and robots is this: robots have no conscience. That means the robot always does its mission. So, the bodyguard robot can also have a mission to deny things like leaps to the opposite side. So, if high-ranking people talk about prohibited things those robots can arrest them. 

The robot can have a so-called ghost protocol. That means that if the person who owns a robot acts against the state, that thing like wrong words activates that attack mode. Another risk is this: if the robot comes from a country that goes to war with the country where it's shipped the seller can simply download an attack program to the robot. The humanoid robot can look like a dummy or lay figure. When the operational order comes those dummies can jump through windows and march to the gun store. And that is one of the darkest visions in robotics. 

https://www.defenseone.com/technology/2025/04/chinas-military-aims-harness-coming-chatgpt-robotics/404811/

https://www.msn.com/en-ca/news/world/us-report-reveals-china-s-plans-to-develop-an-army-of-robot-soldiers-with-advanced-ai/ar-AA1DyCku?ocid=BingNewsSerp

https://www.newsweek.com/china-killer-robots-unitree-robotics-191756

The new AI-based tools can make fundamental advances in astronomy.

"A new AI called Urania is transforming gravitational wave detection by designing superior, sometimes bizarre detectors that challenge current scientific understanding. The best designs are now shared for global research. Credit: SciTechDaily.com" (ScitechDaily, When Machines Dream: AI Designs Strange New Tools to Listen to the Cosmos)

"Einstein imagined gravitational waves over a hundred years ago, but it wasn’t until 2016 that technology finally caught up. Now, researchers are pushing the boundaries again – this time with the help of an AI named Urania." (ScitechDaily, When Machines Dream: AI Designs Strange New Tools to Listen to the Cosmos)

Developed by Dr. Mario Krenn and his team, Urania has designed a series of novel gravitational wave detectors that not only match but often exceed the capabilities of existing human-made concepts. What’s most fascinating is that the AI didn’t just replicate known strategies – it invented brand new ones, some of which scientists don’t yet fully understand. These discoveries are now publicly available in a “Detector Zoo,” hinting at a bold new era where machines innovate and humans interpret." (ScitechDaily, When Machines Dream: AI Designs Strange New Tools to Listen to the Cosmos)

Gravitational waves are a fundamental tool. Suppose researchers want to observe and research things like neutron stars and black holes. The problem is that the black hole's eruption gives information to another wavelength than just gravitational waves. 

Things like gravitational waves can be a new threat to data security in quantum computing. Gravitational waves can shake superpositioned and entangled photons and disturb data transmission in the quantum computer and quantum network. Another phenomenon that can be  

The AI-based systems can make sensor fusion to connect all data from different sensors from radio- and optical telescopes to the gravitational wave detectors. That system fusion allows to creation of new models of how to predict the gravitational waves and things that are connected with it. 

The gravitational waves form in high-energy reactions like neutron stars and black hole collisions. Things like gamma-ray bursts (GRB) X-ray bursts (XRB) and fast radio bursts (FRB) are also connected with the high energy reactions. 

The interesting thing is this: the object sends those bursts when it releases energy. So lonely black hole sends gravity waves when energy injection in the black hole ends. The energy bursts follow each other in the series where the highest energy radiation comes first. Then the energy level in those bursts decreases. 

So, the AI can predict which of those bursts and the gravity waves are connected with each other. In the most exciting vision, the FRB happens just after the gravity wave bursts. And if we someday can predict that energy impulse and its route that allows us to make the system that captures that energy. The FRB can also damage the quantum computers or disturb their operations. So predicting that is very important if we someday make spaceborne quantum computers. 

If researchers can predict the FRB and its route, maybe someday in the future we can use at least the FRB as an energy source. 

The FRB must be trapped by the radio antenna. And then that energy must conduct to the battery. Basically, we have technology that can trap the FRB. But prediction of those radio bursts is not very easy to make. If the system traps about 20% of the radio burst that can solve part of Earth's energy problems. Capturing that thing can happen by using a satellite group that is positioned in the FRB route.  But those are the futuristic visions. 

 https://scitechdaily.com/when-machines-dream-ai-designs-strange-new-tools-to-listen-to-the-cosmos/

New brain implants can restore hearing without side effects.

"A soft, silicone-based brainstem implant from ( Swiss Federal Technology Institute of Lausanne) EPFL shows major promise in restoring hearing for patients with severe nerve damage, outperforming traditional ABIs in comfort and sound precision. Credit: SciTechDaily.com" (ScitechDaily, Beyond Cochlear Implants: A Flexible Brainstem Device Restores Hearing Without Side Effects)

New, flexible, and thin materials make the brain-computer interface, BCI more effective,  flexible, and comfortable than old-fashioned systems. The new systems can restore, hearing or vision for humans who lost those senses. The new and effective BCI systems help to control computers in new and powerful ways. 

Those kinds of new interfaces are connected with other output devices and they can operate along with things like touch screens. Those systems can make it possible to control things like aircraft systems in a very flexible way. The BCI that is connected to computers or mobile devices can give humans access to the net anywhere they want. Those systems can give new types of experiences.

Between the internet, computers, and humans. That system can give new experiences for the enhanced, augmented, and virtual realities. The BCI system can create an experience where a person flows or flies through the metaverse. 

Those BCI systems can remotely control any device in the world. And it makes it possible. Humans can virtually teleport themselves to any city in the world. They can create virtual worlds or make many other things in the virtual entirety called metaverse. The metaverse means the hybrid reality where systems connect virtual and real elements. In the metaverse, people can feel like Superman who flies through the virtual universe. 

The virtual universe can be created from the information that the satellites and space telescopes collect. The system can create things like virtual houses and even cities there are the architects and why not, gamers can walk like being physically in those models. The area and accuracy of the metaverse depend on the computer's power. The BCI can cooperate with other user interfaces. 

The metaverse can turn into a matryoshka reality. In that type of metaverse is possible that the person can be lost in the internal virtual and half-virtual spaces. The metaverse is the platform that allows a user to connect real and virtual elements into the space. There are no limits for the metaverse. 

When we think about those visions the metaverse and BCI are tools that allow the person to connect themselves to the cameras anywhere in the Earth. They can control any device. And the BCI-based metaverse has one problem. There is the possibility that the person will not know what is true and what is virtual reality. In that model, the metaverse is like Matryoshka reality. The system can create multiple internal realities. And it's possible. That person will be lost in that thing. It's possible. That person cannot separate the internal realities from the real world. 

https://www.epfl.ch/en/

https://scitechdaily.com/beyond-cochlear-implants-a-flexible-brainstem-device-restores-hearing-without-side-effects/

Drone swarms will manufacture airplanes and even Mars bases.

"Visualization of a future-state aerospace manufacturing facility powered by swarm robotics." (IE, US to unleash robot swarms to build smart aircraft with speed, precision, safety)

When we think about robots we must realize that robots are not like any animal. The crab-looking manufacturing robots can also have a quadcopter. A system that helps to carry them to points where they cannot climb. Those robots can carry items that 3D printers make or they can be 3D printers. The robot swarm can also look like a humanoid robots. That can turn any factory into automatic. 

The drone swarms will manufacture aircraft and quite many other things. The drone that makes the 3D structures is like a spider or crab. The system can also involve quadcopters equipped with systems that can create high-temperature welding work. The quadcopters and other drones are more flexible than the regular welding systems. And that makes them suitable for many types of building works. One place where that kind of system is suitable is the planet Mars. 

The system can involve the unit where the system melts and separates iron from Martian stones. That can happen using microwave-based centrifugal separation. The system can make iron wire that those drones can use to make metal structures like greenhouse domes for the Red Planet. The system requires carbon which is necessary for steel structures. 

Those quadcopters can make many things on Mars and those systems that can make 3D structures under the AI control are tools that can also fix the holes on the spacecraft body. Those systems can also operate in underwater areas, and they can also make it possible to fix and build structures for submarines. The drone swarms can also fix things like undersea oil tubes and gas tubes. 

But the most interesting thing is the robot-manufacturing system that turns texts into robots. The AI copilot handles the principles of robotics. Those systems can follow spoken words or written text, and make any kind of robot that the user (customer) wants. This kind of system is the tool that makes it possible to transform the AI 's-generative tools into a new form. Modern co-pilots like Chat-GPT, Gemini, and Microsoft copilot can make images from texts. But the new system turns those images into real merchandise. 

That means those systems that can turn the texts into robots can be connected to the drone-swarm manufacturers. The system can make a robot or even a house or ship using virtual reality. The system creates a CAD image that it transforms into a photorealistic image. The system can get things like the size of the thing that drones should make. Then it fits that thing like a house to the landscape image. 

In the final stage, the system can use holograms to make sure that the house or any other thing fits into that point. The system can use a virtual dome like the group of quadcopters to locate other drones in their positions. The drone swarms can also use the internal location system where those drones locate each other. That system makes it possible to control those drones with a very high accuracy. 

https://interestingengineering.com/innovation/robots-from-text-using-ai?group=test_a

https://interestingengineering.com/innovation/robot-swarms-to-build-smart-aircraft?group=test_a

The bias in AI is a big problem.

The AI's problem is that it doesn't think. That is the thing that we usually repeat and repeat. That ability to get data without analyzing it causes bias. The thing that makes that model possible is that the AI doesn't know what the homepage involves. 

It sees words and their connections to something, but then the system doesn't realize what those words really mean. The AI, or large language model, LLM is programmed to give answers in a certain time. That means the LLM doesn't compare and search connections to every single word in the document. That saves time, but it decreases the data's trustworthiness. 

*********************************************

The study put ChatGPT through 18 different bias tests. The results?

AI falls into human decision traps – ChatGPT showed biases like overconfidence or ambiguity aversion, and conjunction fallacy (aka as the “Linda problem”), in nearly half the tests.

AI is great at math, but struggles with judgment calls – It excels at logical and probability-based problems but stumbles when decisions require subjective reasoning.

Bias isn’t going away – Although the newer GPT-4 model is more analytically accurate than its predecessor, it sometimes displayed stronger biases in judgment-based tasks.

The study found that ChatGPT tends to:

Play it safe – AI avoids risk, even when riskier choices might yield better results.

Overestimate itself – ChatGPT assumes it’s more accurate than it really is.

Seek confirmation – AI favors information that supports existing assumptions, rather than challenging them.

Avoid ambiguity – AI prefers alternatives with more certain information and less ambiguity.

(ScitechDaily, More Like Us Than We Realize: ChatGPT Gets Caught Thinking Like a Human)

*********************************************

The fact is that at least part of the results that the AI gets from the net are based on the search engine's page ranks. When we think like this: we teach AI we should give feedback on its results. 

If we tell that the AI gives bad answers that helps it to adjust the results that it uses. The problem is always that for giving trustworthy feedback to the AI we should sometimes be experts. Of the thing that we asked. 

The big problem with AI and other copilots is that those things are made for customers. The product must also please the user in some way. And that can cause the bias. When somebody writes about AI, that person should compare it with other LLMs. The most advanced AI "thinks" that it's the best of all. The algorithms tell it how impressive it is. The more advanced AI means that it's better than other AI's. 

When we think about AI and its operators and owners in the world we must realize that all those companies that develop AI are independent companies. That means those AIs don't have access to the competitor's statistics. So, those AIs are not under one dome. And if they should compare themselves with other AIs like "What is the best AI in the world"? 

Those systems cannot get trusted data. Like how many percent of their answers satisfy the users. That kind of data is not collected. The AI will not ask would the user like its answers or if are they good. Another thing is that if the regular user is not the same as some doctoral user. The AI requires specific commands that involve the right terminology. Without the right terminology, the AI is helpless. 

And then AIs cannot compare those percents with other AI's statistics. This makes the AI think. That it's better than it is. The AI can use Wikipedia. When it should use specific sources like CERN homepages. The AI requires that the person who uses it knows the terminology, and then the misuse of some term in the wrong place in a query can turn the AI's text into a grab. AI will not make a difference in the words "hall" and "hall effect". That thing can cause a situation in which the AI starts to talk about echoes or some aircraft hangars when it should start to talk about resistance in electric wires. 

 https://scitechdaily.com/more-like-us-than-we-realize-chatgpt-gets-caught-thinking-like-a-human/

Lawyers and AI.

Lawyers can use AI to do things as some students do. They can use AI to search for information about the roles of some organizations, members, and member authorities, in their organization. The AI can also search for precedents about similar cases and the AI can also search cases that certain judges handled. 

Same way. AI can search for things like blogs that jurors wrote. That thing tells about their attitudes about cases. And, of course, important data is also what the media tells about cases. So the AI can search, sort, and give links to the stories. That people write about the case or similar cases. Another thing that the AI can tell is how long time the judge or lawyers use for data about the case. 

If the lawyer will not even check the files about the case, that tells something about that person's attitude. And the ability to work as a lawyer. Same way. Judges and other people can see how many similar cases the lawyer handled before. The AI can also see if the witness starts to speak louder, or maybe do some other things that tell about lies. The AI can mark those points in the recordings. 

The person who reads body language can see those points and maybe use still images and zooms to see if a person touches their nose at the wrong moment. Some people cover those things using strong movement. And the zoom image might uncover those signs better than normal tape. The other thing is that AI can also record if the specialist doesn't use the rewind button and the tools. That they should use. 

The AI can also see if the lawyer has shown marks to the witness or given something to them. And of course, AI can check answers like: does a person has a registered firearm, is that firearm reported as stolen? Or missing, etc. AI is a good tool for working as an assistant. It shows does the conviction follows the line that judges normally use. And it can see if there are some anomalies in argumentations. 

 https://www.freethink.com/artificial-intelligence/how-ai-is-reshaping-the-legal-profession

AI computers, nanoantennas, and holograms are tools that can revolutionize manufacturing.

That would raise the sales of the microchips. And that's why that is the goal for the microchip producers. Those new PCs require lots of new microchips. The new AI PCs can run more powerful and effective programs. That makes them powerful tools in the hands of hackers. 

The prediction is that AI-based computers will soon be standard. New and effective microchips and memory controls allow the system to run more complicated code than before. The new systems can also enhance data security. But the same systems can also find vulnerabilities in the old-time systems. The new and effective computers are tools, that can revolutionize things, like graphics, system handling, interfaces, and many other things. 

These kinds of computers are things. That will revolutionize computing and nanotechnology sooner than we even expect. The nanomachines that can transport sensors to the desired positions in the body can make a revolution in neurology. And maybe in the future things like neuro implants will inject into people's bodies. Then nanotechnology transports them into the right positions. 

"The principle of this work is demonstrated by using a laser to levitate nanoparticles and make them carry an electric charge to become an antenna that receives an external electric field containing a communication signal. Credit: Adapted from Zhenhai Fu / Zhejiang Lab, China, edited." (ScitechDaily, 10,000x Smaller: Tiny Nano-Antenna Could Revolutionize Wireless Tech)

The tiny nano-antenna can revolutionize wireless technology. It can make new, powerful multi-band transmitters possible. Those transmitter-receivers can handle thousands of frequencies. And that is a great advance for portable technology. Those tiny nano-size antennas can also operate in the nano-size robots to transmit. 

And receive information. The nano-size antennas can also operate with nanotechnical systems that can put those antennas between axons. And those new types of systems can hear signals that travel in the nervous system. The new antennas that nanotechnology can put their position can make. 

A new type of singularity with living nervous systems and robots is possible. Nanotechnology can slip those antennas into the bodies of natural animals. And maybe quite soon we can see and hear what those animals hear and see. The knowledge of the nervous signals increases all the time. That means the systems can decode nervous signals in new ways. The singularity is the ultimate version of the user interface. In those systems robots and humans are combinations. 

"The method allows the fabrication of millimeter-scale objects within seconds. Credit: © LAPD EPFL" (ScitechDaily, Holograms Take 3D Printing to the Next Level – With Amazing Precision and Speed)

But antennas are not the only things that advancing all the time. The nano-antennas can give revolutionary information. But new AI computers. And new algorithms and AI-based systems can make new types of intelligent, network-based data analysis possible. The idea in those systems is that if they cannot solve problems in a certain time, they call more systems into the neural network to solve problems. 

The highly advanced AI is the tool that can make the new computer-aided design CAD, and computer-aided manufacturing CAM tools possible. The AI can remove errors from the product automatically. And the holograms are making it more effective. The infrared hologram can be used to remove things like scratches from the plastic layers. Or maybe if the IR's temperature is high enough, that lightsaber can remove roughness even from the metal surfaces. The hologram can also used to test what some products like furniture can look like in some rooms. 

Highly advanced computing is a tool for material research. It is a tool for holographic-support 3D printing. In those systems, the developer makes the 3D CAD image for the system. The system makes holograms of the object. Then the developer can check the measures and dimensions of the object. The hologram tells does the object fits into its place. 

The system uses holograms to model things. And then the 3D printing system follows the hologram and fits the model into the hologram. The hologram's shape helps the system make a sharp 3D model by using the printer. The 3D printers can use holograms to test how the part that the system makes fits into the place. Then the system can use AI to fix the diameters and other things. 

 https://www.freethink.com/consumer-tech/ai-pcs

https://scitechdaily.com/10000x-smaller-tiny-nano-antenna-could-revolutionize-wireless-tech/

https://scitechdaily.com/holograms-take-3d-printing-to-the-next-level-with-amazing-precision-and-speed/

China builds the world's largest underwater drone.

"Smaller than regular diesel-electric submarines, China's latest vessel has torpedo tubes (Representational image)" (Interesting Engineering, China secretly building ‘world’s largest underwater drone’ with torpedo tubes)

The new Chinese submarine is smaller than the regular diesel-electric submarines. That means those vessels can be new and deadly drones. Those systems are larger than U.S. Navy Orca, autonomous underwater vehicle, AUV, systems. That makes those Chinese submarine drones more multipurpose than Orca. 

Researchers think that those new submarines can dock to the larger submarines and they can operate in various underwater missions. The submarine can carry smaller drones to the target. Those drones can be underwater caterpillars intelligent torpedoes, or underwater kamikaze drones. That submarine can be like matrioshka. There can be quadcopters that can operate airborne and underwater. 

They also have torpedo tubes, so they can attack other vessels like submarines and ground targets using missiles. And maybe. The Chinese military tests the robot combat swimmers which can operate like human combat swimmers. We all know the Chinese interest in robotics and its military applications. 

(Intersting Engineering, China secretly building ‘world’s largest underwater drone’ with torpedo tubes)

One version to make the submarine drone is to use human-shaped, humanoid robots as operators for regular submarines whose food and beds are replaced by computers. The thing is that the Chinese military and the state of China are not funding humanoid robot research just for fun. In China, nothing happens without military and governmental acceptance. 

Military applications are the primary objective in that country. And the AI-driven submarines are dangerous because they can act in both roles. They can have regular torpedos and missiles. But they can also carry internal warheads. That they can use as the last chance. The idea is that a very large internal warhead forms a giant tsunami that can destroy large areas in coastal lines. That kind of weapon can also destroy all naval and civil vessels in large ocean areas. 

That means those things can be very nasty tools. But those drones may only begin for more powerful and dangerous submarine drones. There are visions that maybe, quite soon, some state will bring full-scale nuclear submarine drones to oceans. Sometimes is speculated that the Russian Status-6, 100-megaton nuclear torpedo could be that kind of vessel. The submarine can navigate to its target using the ocean floor images in the sonar-based TERCOM system. 

And then they can shoot targets or they can explode nuclear warheads inside them. The nuclear submarine has space and nuclear reactors that can give energy to the supercomputers. And that means those submerged vehicles can operate independently. The gravity measurements along with ocean floor maps make them independent from satellite communication and satellite navigation. 

https://interestingengineering.com/military/china-building-largest-underwater-drone

https://en.wikipedia.org/wiki/Orca_(AUV)

https://en.wikipedia.org/wiki/Status-6_Oceanic_Multipurpose_System

https://en.wikipedia.org/wiki/TERCOM

Researchers developed AI that challenges Chat GPT under 50 dollars.

The new open-source AI, S1, challenges traditional AI models. The system uses a similar reasoning model. That Gemini uses. The reasoning language model is not, actually another name for LLMs.  And that is a new step for the AI. Open-source solutions challenge traditional AI. Open-source is not the same as free software. In open-source solutions, developers and other experts can investigate the source code. And that makes the code safer. But that requires deep knowledge of the programming. 

And there is a lot of code in complicated applications that R&D workers must check. Free software means software that we can download it free. But we cannot check the source code, and that makes it possible to slip malicious components in the code. So most so-called trojan horses are freeware. That means the user downloads some very useful-looking software from the net. And then there is the component that acts as spyware. 

Above: A curated list of awesome large AI models, or foundation models, for reasoning.(GitHub, Awesome-Reasoning-Foundation-Models)

Gemini AI tells about the LLM and LRM this: 

"Large language models (LLMs) and reasoning language models (RLMs) are related, but not exactly the same. Here's a breakdown:" (Gemini)

"Large Language Models (LLMs): These are deep learning models trained on massive amounts of text data. They excel at understanding and generating human language. LLMs can perform tasks like translation, summarization, and even creative writing. However, they are primarily focused on language and may not have strong reasoning abilities." (Gemini)

"Reasoning Language Models (RLMs): These models are designed to go beyond basic language tasks and perform reasoning. They can handle tasks that require logical thinking, problem-solving, and drawing inferences. RLMs often incorporate techniques that allow them to understand the underlying meaning and relationships within the text, rather than just processing the surface-level language." (Gemini)

"Key Differences:"

"Focus: LLMs focus on language understanding and generation, while RLMs emphasize reasoning and problem-solving." (Gemini)

"Capabilities: LLMs are good at language-based tasks, but may struggle with complex reasoning. RLMs are designed to handle reasoning tasks, but may not be as strong in pure language generation as LLMs. "(Gemini)  

Techniques: RLMs often use techniques like symbolic reasoning, knowledge graphs, or logical inference to enhance their reasoning abilities." (Gemini)

Relationship:

"RLMs can be considered a subset of LLMs. They build upon the capabilities of LLMs by adding reasoning mechanisms. "(Gemini)

Many LLMs are now being developed with improved reasoning capabilities, blurring the lines between LLMs and RLMs.   (Gemini)

"In Summary:" (Gemini)

"While all RLMs are LLMs, not all LLMs are RLMs. RLMs are a specialized type of LLM that focuses on reasoning and problem-solving. "(Gemini)

The S1 is the signal for the world. The AI is cheaper than we expect. Things like morphing neural networks that share their missions over a large number of less powerful microchips. Morphing neural networks makes it possible to run complicated applications without the latest processor technology. And that makes the AI more complicated than ever before. The AI can generate software. That is extremely complicated. The AI and large language models can turn into virus generators. 

Some people want to give AI to every man and woman on our planet. That can make things like programming and mathematical algorithms more powerful. The ability to connect every smart device together into the giant neural network makes the planet safer. But in the wrong hands the systems that make streets safer turn into surveillance tools, that are more dangerous than we expected. Persons like Stalin and Hitler cannot ever be so dangerous as a person who controls global AI. 

The AI can save human lives. It can detect heart problems before the person even notices those things. But the AI can used to create programs for next-generation jet fighters and tanks. And it's easy to hide malicious code in complicated programs. The AI can collect data from every device, filter, and sort it.

https://www.ceotodaymagazine.com/2025/02/open-source-ai-model-s1-developed-for-less-than-50-challenges-industry-norms/

https://www.digitalocean.com/community/tutorials/understanding-reasoning-in-llms

https://github.com/reasoning-survey/Awesome-Reasoning-Foundation-Models

https://www.edps.europa.eu/data-protection/technology-monitoring/techsonar/large-language-models-llm_en

https://www.hpe.com/fi/en/what-is/large-language-model.html

https://www.rws.com/blog/large-language-models-humans/

Artificial proteins are a new tool against snake poison.

"Researchers have developed new proteins using computational design to effectively neutralize some of the most dangerous snake venom toxins, signaling a shift towards more accessible and efficient antivenom solutions." (ScitechDaily, AI-Designed Proteins – Unlike Any Found in Nature – Revolutionize Snakebite Treatment)

Artificial proteins are the only small part of products that AI-controlled development can find. The AI-controlled manufacturing systems are packed with computers, spectrometers, and effectors. Basically, that combination is similar to that used in the large stellar observatories for research space. The difference is that those new reaction systems observe microcosmos. 

They use advanced microscopes and spectroscopy to create precise chemical environments. Things like ultrasound and laser tweezers along with ion systems move those particles in that chamber. 

Those effectors can be mechanical, chemical, electromagnetic, etc. The ability to react fast with high accuracy makes that kind of system more effective than traditional chemical chambers. The artificial, genetically engineered cells can create proteins and other things. That people require. The artificial cells can create an almost limitless number of complex chemical compounds. One of the compounds is the cloned shell of the cancer cells. That thing makes the ribosome inactivation. 

The new nanotechnical medicines can be proteins. That is packed in with other proteins. The shell proteins drive the medicine into the right cells and then the ion pump releases the effector to the cell. That effector can be the DNA bite that orders cells to die. Basically, that effector can be nano-silk that fills the targeted cell. 

There is the possibility to create medicines that close the receptors that offer poison to come into the cell. The protein. That connects itself to that receptor makes it possible to predict the poison. The protein inactivates that receptor only when the poison is in the body. And that makes it possible to give revolutionary counter-poison for venomous snakes and even against chemical weapons. 

In the ideal case, the counter-poison can be taken before the person even sees snakes. In those cases, the snake poison itself activates the counter-poison. 

The person who walks in the dangerous area just takes the capsule that involves nanomachines which inject that protein or protein producer into the body. The idea is that the counter-poison is taken before any snake is seen. And when the snake injects its poison to the victim the counter poison will activate. If those protein producers are artificial cells there can be programmed cell death that removes those cells in a certain time. Nanotechnology allows to create the customized cells. That surface antigen fits the receiver's body. 

Then nanotechnology can combine those proteins into a new form. Nano-, and biotechnology make it possible to create biological plasters that include fungi that create the penicillium to wounds and leaf cells that give oxygen to wounded areas. Those leaves can also give oxygen to the body of a wounded person. Those biological components like generically engineered fungi can also create proteins that can neutralize snake poison. But theoretically, genetically engineered snakes can create medicines in their poison glands. 

In one vision those remotely controlled snakes that the microchips control can deliver those medicines or vaccines to the human body. Another interesting vision is to make artificial cells or artificial mosquitos that can deliver things like vaccines. The visions that nanotechnology can make are limitless. The ability to control DNA and individual molecules. Are things that offer a world where the imagination is the limit. 

https://scitechdaily.com/ai-designed-proteins-unlike-any-found-in-nature-revolutionize-snakebite-treatment/

https://scitechdaily.com/testing-thousands-of-compounds-simultaneously-to-uncover-new-drugs-and-tailored-treatments/

DeepSeek and its connections.

When you use DeepSeek you know what you use. The program runs on Chinese servers. And it must follow Chinese regulations.  That is the thing in every program. The system tells that it collects data, and that's it. That system is free to use. And that is one of the reasons why we should think about AI very seriously. In the case that somebody does something in China. There is a government behind or with that thing. 

If DeepSeek is freeware. There must be something that its competitors offer. Or Chinese company takes their customers. The question is: Where does the Chinese company behind the Deep Seek get money to pay workers their salaries? The free software doesn't bring money to the company. And that is a good thing to remember. So, DeepSeek is an interesting thing. I tested it. And it was quite an interesting experience. But it won't make images. So, that's why the image above was made using Gemini. 

There is a possibility that the Chinese want to pull rags from below Western AI companies. So, when the Chinese give their AI for free, nobody wants to pay the price. And that takes customers from Western AI companies and ends their research work. The DeepSeek is the tool that makes everything different. It uses powerful and advanced algorithms. 

And it's lightweight. The DeepSeek will make AI development more effective. It doesn't need the most advanced microchips. And that means it's easy to run on a little bit older hardware. Because the DeepSeek doesn't need the most powerful microchips. Is easy to use in countries where the newest microchips are not allowed to transport. That new AI boosts the AI advance on all fronts. The Western companies must have something to offer or their Chinese competitor will take their marketing position. 

The Chinese AI leaves the Western competitors behind.

When we start to use some new product, we must read what it does. Then we must find out what else that product does than its makers tell in commercials. Many times things that are not listed are more important than things that producers of the product tell to customer. The Trojan Horse is the spyware that plays some utility programs. Those programs include a hidden algorithm that dumps the hard disk and net traffic into the net. 

The new and hot thing in the AI business is the Chinese DeepSeek. That artificial intelligence runs on lighter platforms than Western competitors. And that makes it interesting. The fact is that the Chinese AI runs on Chinese servers. Which means that thing has connections with Chinese official policy. DeepSeek has one similarity with the other Chinese innovations. That means the DeepSeek seems to come from nowhere. There was nothing before the publishing. 

And then suddenly the most innovative AI in the world is reality. Sometimes somebody can say that there is suspicion that the Chinese hackers stole the code for the AI, which is similar to other AI's. That thing means that the China took a big step in AI. But the thing that the Chinese seek might be a surveillance tool that can follow and map Chinese citizens. 

The DeepSeek might be only one part of the AI development program in the East. That AI might have the "evil twin" that makes the next-generation hacking tool possible. The advanced AI is a tool that can serve military purposes. The highly advanced algorithms can remove databases from the hard disks. That makes the AI itself is weapon. 

The new road for AI is that it turns very important thing. The AI can generate virus code very fast. And that's why AI can support other systems. But the fact is that. Chinese AI programs are not possible without the government's acceptance. 

The government in China wants to use AI as an intelligence and control tool. There are no limits to the information collection in China if it serves the government's and state interests. AI is a tool that can revolutionize the Chinese control policy and their hacking- or cyber espionage projects. And that is one of the things that we must realize. China and its government know its value and power in the world.  

https://edition.cnn.com/2025/01/27/tech/deepseek-ai-explainer/index.html

https://en.wikipedia.org/wiki/Artificial_intelligence_industry_in_China

If we want new ideas we must be open to them.

The science facilities play a vital role in the R&D process. Things like Los Alamos Scientific Laboratory, Oak Ridge National Laboratory, and AREA-51 offer technical support and new innovations for the USA in civil and military sectors. The thing with technological advancement is that nobody can predict its routes. The fact is that: we can deliver our innovations for other nations by using one thing. We can deny innovations if we want simply blaming engineers and other scientists. If we want to deliver all our products to Mr. Trump and Mr. Musk we must only deny the R&D development. 

R&D is the key to the new equipment in the civil and military sectors. Being open to new ideas gives the landscape where innovations can grow. That means we should not underestimate science itself. Science is the thing that people do just for themselves.  When scientists find some phenomenon, there is an analysis of how people can use that phenomenon. 

Things like altermagnetism and other kinds of stuff like stopped photons are things that can make communication more secure than ever before. New things like radio telescopes can make things like follow the radio signals that come from a distance. 

They can eavesdrop on opponents' communication antennas or they can destroy or drop enemy drones. Or they can help to call scientists to some country. The limits are the worst thing that people can do for the R&D process. Limits and rules are only papers or texts on screen. And in the worst cases, the nation that made those rules will not even follow them. 

The AI is coming and that is the thing that we must realize. AI is the thing that can make it possible to control many objects. In stock marketing that means thousands of investments. And AI-based stock trading is the thing that will not allow the investor to make defeats. The AI can sell those objects immediately when they start to go down. And that makes the AI a tool that can guarantee profits. 

The AI is a tool that can make life better or worse. It's the most effective tool in modern computing. It's possible. The EU and other countries limit the use of that tool. That can make our lives better. But in the worst case, some actors like China and the USA, or some South American countries will not follow those regulations. They can call innovative people to make the AI. And that is one of the things, that can break the economy in the entire EU.  

There are stories that AI is bad for science. People published AI-based writings and said that the AI made their texts. That thing is similar to cases that other human makes the texts that developers or scientists introduce as their own. The source in those texts is the AI. The problem is that there are always people who want to cheat. They choose easy ways to make their thesis. That is wrong. But the problem is that nobody does their work with the computers there their schools have access. Those things are normally made with students' own computers. 

https://bigthink.com/starts-with-a-bang/new-science-facilities-keep-america-great/

https://bigthink.com/business/why-ai-for-business-means-a-revolution-in-efficiency/

Occam's razor as a thinking experiment, or model.

"A new Oxford University study reveals that deep neural networks (DNNs) naturally favor simpler solutions when learning, acting as an inbuilt form of Occam’s razor that balances the exponential growth of complex solutions. This simplicity bias allows DNNs to generalize well on real-world data but limits their performance on more complex patterns, hinting at deeper parallels between AI learning and natural evolutionary processes. Credit: SciTechDaily.com. (ScitechDaily, Scientists Discover AI’s Hidden “Occam’s Razor” for Simplified Solutions)

A couple of formulations of the Occam's razor. 

"This principle is sometimes phrased as Pluralitas non est ponenda sine necessitate ("Plurality should not be posited without necessity").In his Summa Totius Logicae,"

"To quote Isaac Newton, "We are to admit no more causes of natural things than such as are both true and sufficient to explain their appearances. " 

"Therefore, to the same natural effects we must, as far as possible, assign the same causes." In the sentence hypotheses non fingo, Newton affirms the success of this approach."

"Bertrand Russell offers a particular version of Occam's razor: "Whenever possible, substitute constructions out of known entities for inferences to unknown entities".

Quotes are from Wikipedia. (Wikipedia Occam's razor)

The origin of the Occam's razor is unknown. But Ptolemy once said: "We consider it a good principle to explain the phenomena by the simplest hypothesis possible." So here I don't write about the history of Occam's razor. I will write about principle that the Occam's razor represents. 

There are a couple of formulations about Occam's razor. There is an idea that the simplest solutions are the most common because complex solutions form through the simple solutions. Or the simple solution becomes before the complex solution. 

A simple solution must exist. So that it can intertwine with other simple solutions. So the complex solution requires more than one simple solution. And we can call solutions as models. So the complex model is the sum of simple models. That means there must be more than one simple model (or solution). That can form a complex model. (Or complex solution is the sum of simple solutions)

The Occam's razor is the thinking model. Where the thinker should select the simplest models if there are multiple offenders. The most usual models are the simplest models. And that means the AI should select the simplest model. Complicated models require more time and more structures to form. So everything begins from the simplest possible models. When we talk about Occam's razor and the principle that the simplest things are the most usual things we can make a model that if we choose the simplest model the probability that we hit the right point is more usual than if we choose complex models. 

That is one way to handle complex problems. And problems are systems. In the beginning is the simple model. Then that model collects layers on them. And those layers form complexity in the model. That is an important point in the thinking. The fact is that Occam's razor does not always give the right solution. That means sometimes the complex structure can be the right. But then we can think so simple things as the line. We can make lines that seem very complex. But if we look at those curves and loops we can see that even if that thing looks very complex. 

There is only one line that involves those things. If that line is woolen yarn that thing can form very complicated-looking structures. But if we have time we can clean up that mess and then use that woolen yarn. So everything that seems complex can turn simple if we think and try to fix that thing hard. Knots and curls don't make the woolen yarn useless. If we use little time and brains we can remove those knots and make the woolen yarn usable. Sometimes there is only one knot that holds very complicated curls. 

If we want to solve the problem we must know everything about that. If we lose pieces of the problem we cannot fix it in its entirety there we can solve it. When we think of woolen yarn and a razor. A thinker that we can call Occam can use the razor as a wire coil. This person uses a razor as the rack around which he wraps a woolen thread. So that means Occam's razor can be used as an example that there is always more than one possible use of the tools or subjects. 

And if we follow the woolen yarn from the end to begin, we can find the shape of that thing. Checking the answer for the calculations happens by calculating those things backward. So that means the mathematicians calculate all calculations using mirror calculations. When they check simple calculations for example 7+2=9 they can check that calculation by calculating 7=9-2 (or normally 9-2=7). That thing gives the second meaning for the Occam's razor. 

That means people must not use violence or destroy structures to find the key. If Occam uses a razor and cuts the woolen yarn in pieces, that means they turn carbage. There is no use for things that are cut into pieces. For making the woolen yarn people must have every part of it. And that things must be sorted into the right order if Occam wants to put possible colors in the woolen yarn into the original order. 

https://scitechdaily.com/scientists-discover-ais-hidden-occams-razor-for-simplified-solutions/

https://en.wikipedia.org/wiki/Occam%27s_razor