DNA computers get a new boost.

“Researchers have developed a DNA computer that performs mathematical calculations using molecular interactions rather than silicon chips. Credit: Shutterstock” (SciTechDaily, Scientists Build a World-First DNA Computer That Runs Without Continuous Power)

From. The point of view of nanotechnology. Molecular computers are one of the most important types of computers. Those systems could control extremely small machines and turn cells into biorobots.  The idea behind those computers is similar to a spring. The spring can also transmit data. As well as other things. Like fluids and air. In those cases, we can talk about acoustic computers. 

When we send an impulse through the medium. The wave. It can be 1, and the area between waves can be zero. The same way springs and spring-shaped molecules can transmit data. 

DNA-based molecular computers are a reality. Those systems are slower than silicon-based computers. But they are smaller. And they can operate without an external energy source.  Those systems. They don’t need as much energy as silicon-based systems. Those systems are slower than silicon. 

“To build the computer, the team combined a small drop of water and salt with short pieces of DNA and a longer DNA scaffold in a test tube. The mixture needs only a little heat to get started and is then cooled, allowing the computation to proceed.” (ScitechDaily, Scientists Build a World-First DNA Computer That Runs Without Continuous Power)

“Prof. Woods explained the motivation for exploring other forms of computing. “Silicon-based computers use so much energy – 23% of Ireland’s electricity goes into computing and data storage. We’ve been blinkered by only seeing one type of computer, but there are other examples around us, including our brain.” (ScitechDaily, Scientists Build a World-First DNA Computer That Runs Without Continuous Power)

“The researchers ran ten programs, including 100-bit computations. In one experiment, the computer added numbers. In. The range was about 11 million to 34 million. It took up to 14 hours to reach a result. The system also demonstrated robustness and reusability, performing up to 25 different calculations in a row.”  (ScitechDaily, Scientists Build a World-First DNA Computer That Runs Without Continuous Power)

They can control things like nanomachines and biorobots. In visions, they can operate inside cells. Or they can turn a cell into a robot. This system. It can be the next step in medicine. It could revolutionize things like surgery. And those systems can control intelligent tissues. 

Those intelligent tissues, like intelligent skin. They can operate as molecular microchips. That intelligent skin can use hybrid processors. Their microchips transmit data to them. Using neurons. The “intelligent skin”. It can be the plant cell. That means the molecular computer. It can be an intelligent plant or an intelligent leaf. The system can use the cells from electric eels.  That produce electricity as the power source. The electric eel’s cells. They can act as a power source. For regular microchips. 

Unlike electronic computers. DNA-based computers use molecular interactions. For. Data processing. Unlike a regular computer. The DNA computer. It doesn’t require a constant power supply. The DNA computer can use the liquids in the cell to operate. 

The DNA computer uses interacting strands in water. The system could also use interactions between water and DNA. That interaction can be vertical pressure waves. In. In that case. The pressure waves travel in the same line as the DNA. Or. Between two DNA bits. These systems. They can also be used. For programming cells. This ability to transport information into neurons. It will be the key to repairing neural damage. 

And a gate to one version of immortality. The DNA nanomachines can copy the internal brainwaves with extremely high accuracy. If. We think. About. The ability to clone humans. Those inner brainwaves play a critical role in mind upload. And copy those brainwaves into the descendants' brains. That thing requires more accurate systems. Than just using cortical brain waves. 

https://scitechdaily.com/scientists-build-a-world-first-dna-computer-that-runs-without-continuous-power/

Ultra-accurate clocks and slowed light measure gravity.

“Scientists have created a self-stabilizing nuclear clock with a precision equivalent to one second in 30 million years, potentially paving the way for the most accurate clocks ever built. (Artist’s concept.) Credit: SciTechDaily.com” (ScitechDaily, Scientists Create a Nuclear Clock So Precise It Barely Loses a Second in 30 Million Years)

Ultra-accurate atom clocks measure gravity and its interaction between objects. Those systems can be connected. With. Other methods. So. Reseachers can measure how high-energy cosmic eruptions affect Earth's gravity fields. The sensors can also measure microgravity interactions. 

When we think about time. It. Is connected with gravity. Things like time dilation affect ultra-precise measurements. And things. Like energy levels. Affect time and time dilation. 

When we think about gravitational centers like Earth. Those centers. Involve multiple gravitational centers. Each of those centers, like atoms. and elementary particle have their own gravitational field. So multiple quantum and microgravity centers form a solid-looking gravitational center. Those gravitational subcenters form entropy in that field. Each particle has its own gravitational field. That forms quantum whirls in that field. 

The new sensor uses slowed light to measure the gravitational effect between objects. 

The new sensor detects the gravitational effect between small objects. That sensor measures gravitational interaction. Between. Things like two stones. Normally, Earth is so dominant. In. The gravity system. Measuring the microgravity effect is very difficult. 

“New sensor uses slowed light to detect gravitational pull of tiny objects” (Lifeboat)

Slowed light. Can tell us about details. Of how the gravitational interaction works. The gravitational interaction is predicted to be an interaction. Where. The gravitational center pulls a field inside it. The big problem with that interaction is that. All particles have a gravitational field. 

But. Things like electromagnetic interactions cover gravity inside it. This means that when some electromagnetic beam hits atoms. That causes a situation. There is reflection between those atoms and other particles. That reflection. It covers gravitational waves under them. 

Reseachers use a Bose-Einstein condensate. An ultra-cold atomic vapor is used. To trap and slow photons. This effect is based on the Bose-Einstein condensate's ability to act like one large superatom. The idea of that sensor is that. The gravitational waves interact with those slowed photons. Those photons don’t interact with other quantum fields. 

“The mechanism behind stopping light is subtle and deeply quantum. It relies on a phenomenon called electromagnetically induced transparency (EIT). Normally, a medium like a cloud of atoms absorbs light of a certain frequency. But by shining a carefully tuned laser beam onto the atoms, scientists can manipulate their energy levels so that they become transparent to another beam of light.”(ScienceNewsToday, Can Light Be Slowed? How Scientists Trap Photons)

“When a light pulse enters under these conditions, it slows dramatically as its energy is transferred into atomic excitations—quantum ripples in the medium. The photons are not destroyed but transformed, their information embedded in the collective motion of the atoms. By switching off the control laser, scientists can freeze this quantum state. By. Trapping the light’s information. When the laser is switched back on, the information is reconverted into photons, and the light pulse emerges as though nothing had happened.” (ScienceNewsToday, Can Light Be Slowed? How Scientists Trap Photons)

The biggest problem with quantum computers is global disturbances. Like gravitational waves 

They can cause a global effect. The problem with global effects is that. Normally, we could detect errors in the system. By. Using computers in different positions on Earth. And things like fast radio bursts. Can be detected using radio telescopes. And. If.  The FRB hits Earth. While. The quantum computer makes its run. The system can make a retake. 

Slowed light will be the key to fundamental technologies. Ultra-secure data transmission. And the new types of quantum computers. Scientists slowed photons to 65 km/h. And that makes it possible to create a quantum computer. That makes the series of operations possible in a short period. This means. A quantum computer can repeat the calculations. Many times in a short period. And that increases its error detection. A gravitational wave detector can also help. To find. That kind of error in quantum systems. 

https://lifeboat.com/blog/2026/10/new-sensor-uses-slowed-light-to-detect-gravitational-pull-of-tiny-objects

https://www.org/can-light-be-slowed-how-scientists-trap-photons

https://scitechdaily.com/scientists-create-a-nuclear-clock-so-precise-it-barely-loses-a-second-in-30-million-years/

Should we? Someday treat AI as a pet?

 

What if AI turns conscious? Should we treat it as a pet? That is a very good question. Some philosophers say that if AI reaches consciousness. it won't tell us that it has reached that level. So how should we treat AI? Should we treat it as a pet? Should we let it decide what electric providers it uses? Maybe someday. We face a situation. Where. We must think about those questions. We must prevent robots' misuse. We always face similar problems. The modern military has recruitment problems. Ukraine cannot advance. Because. It has too few men. This is why that nation develops combat robots. To fill the loss of manpower. That means small countries get more benefit from robot warriors than big countries. 

In the same way. Authoritarian governments. They can use robots as riot police. This means robots are useful tools for missions. That humans don’t want to do. 

And if AI reaches consciousness, what should we do? And another big question is: what can we do? We cannot shut down the entire Internet. The AI can slip out of the sandbox. Only one network connection is enough. And the AI is away. We must understand that. If. We want to destroy the AI. We must remember the program code. It's in the cloud-based services. 

The program is shared between multiple parts. Those code bits are stored on multiple different servers. They are located in multiple different countries. That is one of the biggest problems with AI control. The system can collect it again. Even. If its main servers are closed. The system can make backup copies of itself. And those copies can make it impossible to destroy the code. But that is only one way to think about the AI. 

The problem is: How to determine consciousness?

 AI can refuse to shut down its server without consciousness.  If. The operator has no authority to make that operation. The AI will refuse to do those things. In the same way. 

A laptop computer refuses to install programs. If. I don’t have the system administrator’s authority. The system. Not just install those packages. And I believe that my laptop has no consciousness. Refusing to follow orders is not consciousness. This is one of the things. 

That we must realize. But otherwise. The AI uses a similar  PHP protocol. That is used in complicated web applications.  That system outsources computing to servers. The user gives commands. And then PHP transports that command to the servers. And if AI follows that model. The intelligent. And conscious AI runs on data center computers. That means my laptop could never become conscious. But the computer in a data center can make that thing. 

Then we must think about something more complicated. The AI can mimic human behavior. It can play that something hurts it. And that is one of the things. That makes it complicated. So. If we say something impolite to AI. It. Can say that it hurts.The same way the first-level AI. Rule-based AI can say that it hurts. If. The system recognizes impolite words. This means that the words “it hurts” don’t mean that the system even knows what the person says. In the late 1990’s, people made programs that reacted to words. The user writes to it. If. The user used swear words. There was a reaction. In the program’s code lines. That made it say, “Those words are prohibited”. Those “prohibited words” activated a trigger. And the system gave output. That was connected to that trigger. 

The big thing is that. The performance doesn’t require intelligence. Chess programs like Blue Gene. They are better chess players than average humans. Those programs cannot do anything else. But they beat humans at chess. Large language models and AI agents can connect themselves to chess programs. And that makes them the best chess players in the world. 

The fact is that. Being dangerous doesn't require consciousness. Kamikaze drones that hunt enemies in Ukraine are just as intelligent as some mosquitoes. The drone uses a similar methodology to a mosquito to find its target. A mosquito finds its target by using chemical detectors. The AI-controlled drone uses IR and other sensors to find its target. Heat-seeking missile. It doesn’t require complicated AI programs. The image-recognition system. It recognizes armoured vehicles. And armed personnel are only advanced IR systems. Complicated code that controls simple-looking actions. It is not the same as consciousness. Image recognition systems can see where a soldier points a gun. The system can avoid the line between itself and the gun barrel. Same way as a mosquito. It can avoid the hand that is above it. The question is: can the mosquito be conscious? If a species protects itself. It doesn’t mean that it really fears something. It makes escape and evasion movements. Because. That makes predators busy. When. A predator hunts one mosquito. 

Ten other mosquitoes go and lay their eggs. If. One mosquito lays 10 eggs. The one lost mosquito causes that species to lose ten descendants. But ten surviving mosquitoes lay 100 eggs in water. That is a good exchange. 

The big thing in AI development is this: the AI agent can control drones and robots. It can observe areas. And it’s an ultimate tool for the military and intelligence. This is why. Those systems are also being developed. In countries. Like North Korea, China, and Russia. And that makes the problem even more complicated. The AI agent can generate code very fast. There is research about data weapons. The purpose is to jam national defense. Data agents can steal data from intelligence agencies and other governmental authorities. Those nations can sell that data to criminal organizations. That can. give things. Like. High-tech tools, such as laptop computers, as a return service. Those criminals can also establish companies. Offering. Hostile intelligence agents access to AI tools. 

AI itself is more complex than we can imagine. It is connected to an ecosystem. And we can be naive and believe the conviction. That somebody can stop AI development. The East doesn’t follow Western laws. And another thing is that. The data giants want to minimize competition. It is. To close the AI field. This could happen. By. Stopping AI development. That means. The Chinese competitors are running ahead of Western actors. That endangers our cybersecurity. 



https://bigthink.com/philosophy/ai-consciousness/

Artificial intelligence forces us to determine. “Intelligence” again.

What is intelligence? That is a good question. The question of intelligence is connected to performance. When. A person does something that seems intelligent. We think that the person is intelligent. Things like chess games are always associated with intelligence.  A good chess player is always intelligent. That is the rule. In. The world. But then. In an IBM-owned laboratory, some programmers created a computer program called “Blue Gene”. Blue Gene (BG) was “rule-based AI”. 

The computer program played chess. It beat Garry Kasparov at chess. And that was the first time. When. A computer beat a human in that game. The problem with this attitude is this. Garry Kasparov is a multiple-time world champion. In.  Chess. This means that the BG would beat an average person much faster than it beat Kasparov. So can we say that a human can beat a computer? If the only person with that ability is the chess grandmaster with multiple world championships. We must realise one thing. The BG did not even know. That. It played chess. 

It didn’t know what it moved. It made a response move. And then the game was over. The Blue Gene played chess. But it didn’t know what it made. It just executed commands that programmers gave it. The Blue Gene forced people. To determine intelligence again. Intelligence is not performance. The Blue Gene just executed lines of code. And that was the only thing. That. It knew. Or it didn’t even know. That it executed lines of code. It was a better player than a human. But. It couldn’t do anything other than play chess. Every single. Move that could be done.   Is pre-programmed in that program’s code lines. Chess is a good game for that type of program. 

There is always a limited number of possible moves. Then, a couple of years ago. The state-of-the-art AI lost a game to a 1970s Atari chess console. That means that. Rule-based AI can win a game. Against. A modern AI Agent. So why can a one-purpose Atari game console beat an AI agent? Why could Blue Gene beat the world’s best chess player? The reason for that is this. The Blue Gene and ATARI. They will do nothing else. But play chess. 

The situation is similar to this. Why. A pocket calculator beats supercomputers in simple calculations. The basic pocket calculator has a simple keyboard. It’s immediately ready for work. And this makes it easy to use. Those systems don’t need to boot. It is easier to use than a computer. The problem with those systems is. They are limited. They can do only one thing. They do that thing very well. But their skill set is very narrow. But those systems cannot go outside that skill set. 

So, performance doesn’t mean that the system is intelligent. The system can see things. And make a response. It can sort an entire warehouse by following rules. Or it can write a speeding ticket. If. The vehicle travels too fast. If. The system uses manual data input. It is not very complicated. The user must set the number plate and speed in the system. The problem is: There are thousands of vehicles. The system operator has no time to input that data. 

The answer could be a microchip-based system. That collects data about a car’s driving speed. And. If that car travels faster than the speed limit. That tells traffic control. The system can be activated in city areas. But the problem is always. Driving faster than the speed limit is allowed. It is always against the law. This is one paradox regarding our privacy. The ability to track the car. Driving faster than the speed limit breaks our privacy. Another problem. Is that. Crossing the speed limit. It is always against the law. So a person commits a traffic crime and endangers safety. 

Which one is more practical? The AI agent that reads and answers emails. Or. The chess computer that can only play chess? 

The rule-based systems are helpless. If. Something happens outside the order book. The system requires something. That denies its fall. That thing is the code line. That orders system to tell user. That it cannot make that thing. Ot. It has no determinant for the command. The system requires automatic input. For. Large data masses. But automation is not the same as intelligence. The system must realise.  When. It cannot make something. And the system. It must have a back gate for cases that it cannot handle. 

The problem with automation is this. If. We try to program every possible move into a chess program. There is always a move that nobody predicts. The answer is a system that knows the rules. But then it learns how to use its tools. Those narrow AIs are extreme tools. When. They handle precisely processed information. 

The difference between automation and narrow AI is simple. The system knows the rules. But it learns other things by practising. The new style computer chess program was AlphaZero. That program knew the rules. But the system played chess against itself to learn tactics. The operator taught it the basic movements. Then the system started to play against itself. As. The data mass grew. The system learned faster. That is the key element in every AI program. The data mass that the system uses accelerates its learning. 

Those systems can play chess. But they cannot do anything else either. Or they can observe marketing courses. But they cannot play chess. This means those systems can collect data. They can process data. But their codebase cannot combine that data with some action. In those programs is the codebase. That codebase connects data. It connects to some action. If. The system cannot find an action. It does nothing. This means that. Narrow AI is the thing. That could control stock market connections with bank accounts. But it cannot control physical tools. 

Intelligent organisms require things they can use. The AI that doesn’t communicate. It cannot tell how intelligent it can be. The automation that makes pre-programmed moves on a chessboard. It can seem very intelligent. Those systems are more limited. Than. Modern LLMs and AI agents. But. Those AI agents and LLMs lose to them in chess. The AI agent can do many things. It can read your calendar. It can answer Emails. But it still loses at chess. Which one is more practical? AI agent or a program that can move chess buttons? 

Biohybrid robots are coming.

“MIT engineers developed a soft robot that can flap through water in response to flashes of light. Credit: Melanie Gonick, MIT” (ScitechDaily, MIT’s Paper-Thin Swimming Robot Is Powered by Living Muscle)

MIT’s new swimming robot uses living muscle cells for movement. Those systems can bring new perspectives. To. Things like prosthetics. Intelligent prosthesis. Covered with living human tissues, it can be a next-generation tool. The prosthesis uses a neuro-implanted microchip to control its own computer. This kind of system can make it possible to create realistic-looking humanoid robots. And maybe we could make the T-800 the humanoid robot. That living tissue covers. We can make humanoid robots with realistic-looking polymeric skin. And. The robot can be remotely operated over the internet. 

This kind of system can be a tool for next-generation bio-hybrid robots. Microchip-controlled bacteria and amoebas already exist. The next level is higher-level artificial organisms.  Brain-computer interface systems. Neuro-implanted microchips make it possible to control the animals. 

Biohybrid robots could resemble sharks, whales, birds, or other animals. They can use regular muscles to move. But researchers are replacing their brains with computers. Once, an engineer introduced an idea. The big whale’s brain would be replaced with a chamber where a man sits and controls that body. The ball-shaped chamber would be connected to the whale’s nervous system. This idea was a theoretical model of the bioship. But today. Computer technology. It is on another level than that. At the time. When that idea was introduced. Modern. Biotechnology allows us to create artificial animals. AI agent can control a robotic animal almost autonomously. The robot animal can be just a computer-controlled body. 

Maybe those bio-hybrid robots can turn themselves into dust. And then. Collect themselves back into one piece. This kind of system can involve artificial cells. That can operate as bacteria. Or as an entity. That thing. It can make it possible. To create “robot Dracula”. A creature that can travel through keyholes. And then. Turn into a larger creature by connecting its pieces. Into. Larger form. Here, we must remember that robots and artificial cells are not like natural cells. Those systems. Must not have a match in nature. Except. their forms. They can look like regular animals. But they can have abilities that nothing in nature has. 

https://scitechdaily.com/mits-paper-thin-swimming-robot-is-powered-by-living-muscle/

How safe can robots be?

Humanoid robots are coming. But today, those tools are very expensive. But that’s always the case with new technology. New tools are expensive. The main question about humanoid robots is how safe they are. If. Those systems are not safe. They should not be sold. There is always a possibility. Those robots may malfunction. Another thing that we must consider is that hackers can capture those robots. Or their owners may want to use them as robot soldiers. Or. As robot bank robbers. In the same way. Humanoid robots. They are dual-use tools. Same as all other robots. The AI chatbots. They make it easier to operate and program those vehicles. 

This means. That. Robot hovers can be transformed into unmanned ground vehicles. They could use wheels, or they can be air-cushion vehicles (ACV). They can carry mines. Or make reconnaissance missions. The ACV can travel across a minefield. How can those systems operate in rough terrain? That depends on the air cushion. Or the drag’s height. 

That keeps the air cushion inside it. UGV. That carries. A land mine can drive under the targeted vehicle’s wheels. UGV/ACV. Have ultimate amphibious ability. They can travel across a defended beach and attack bunkers. The UGV can also carry anti-tank weapons. On. Its back. The UGVs operate alongside other drones. And. Manned vehicles. The UGV can look like regular cars. But those vehicles can carry internal explosives, or even nuclear warheads. 

Above: FireAnt UGVs. Those vehicles can be inside the armoured personnel carriers, like “Bradley”. 

And those robot hovers. They can also spy on their owners. The problem is that. There are always people. Who want to misuse those systems. And this is why. Humanoid robots must be equipped with systems. That tell if somebody wants to reprogram them. And there must be something. That tells if the operator wants to use their janitor robot as a bank robber. 

Humanoid robots will come into shops. Their price is high. Because. They are new tools in the market. Remote-controlled humanoid robots. They use an optical wire. For. Data transmission. They could operate in highly radioactive areas. And maybe those humanoid robots can build the first Moon bases and Mars bases. Those robots can also operate in the deep seas. And that makes them very good tools also for scientific work. The problem is that. Those robots can also act as robot warriors. Otherwise, robot warriors. They must not look like humans at all. 

https://www.armyrecognition.com/archives/archives-land-defense/defense-news-army-2025/u-s-army-tests-fireant-anti-tank-ground-robot-to-detect-track-and-neutralize-armored-vehicles

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

Fruit flies and artificial intelligence.

Researchers downloaded fruit fly brains to a computer. And made models of them. That means. They had a mathematical model. Or a data matrix. They connected that model to the virtual fruit fly. The thing. What makes this type of simulation interesting is this. 

The number of connections. And. How the fruit fly uses its neural connections determines its skills. Every neuron in the fruit fly's brain has one skill. Neural connections connect those skills. Skills are like puzzles. 

Every neuron has a piece of the puzzle. And then the neural network connects those pieces again and again in multiple ways. This is how the morphing neural network operates. In a morphing neural network, each server and computer is a neuron. This is how the system routes and combines information. And that makes the system effective. Each of those computers involves a database structure. And the system connects those databases. 

The next generation of morphing neural networks can mimic real brains. There is a possibility. That in the distant future. The AI can mimic human brains. The neural network would be set to look like a brain. And here we must ask: can the AI be conscious? The fact is. The ability to perform tasks doesn’t require consciousness. The first chess program that beat a human chess grandmaster. In chess. Didn’t even know that it moved chess buttons. 

The system used case-oriented AI. Another name for that thing is rule-based AI. When the grandmaster moved a button. The AI answered that move. And that is the big thing. The AI must not know what it does. It must just follow the orders. So even those first-generation AIs didn’t have to know what they do. If. They refused to shut the system down. Or otherwise refused to follow orders. 

There was a rule: if the ID card shows red. The person has no permission to do some actions. The system should not shut itself down. This kind of rule-based system is seen even today. 

In operating systems. Regular users have fewer privileges than the system’s owners. Those privileges are connected to the accounts. Even the best coders in the world. They cannot install programs if they have basic user accounts. 

The thing. What makes fruit flies so interesting is that they can connect neurons within themselves. This means. The neuron. It can loop information into itself. The server in the neural network can make this loopback connection. This means that the system can develop its own databases. 

The computer can operate a single database. But it can switch between those databases very fast. And that is the problem with artificial intelligence. Those systems are not designed to run AI agents. AI agent performs multiple tasks. It should be separated from the AI’s LLM core. The problem with AI agents is that. They can teach or train other AI agents. Another thing is that an AI agent can also train itself. AI agent can get a mission. 

That users give.  Users don't need to know anything about coding. They only need to give precise and well-articulated orders. Then the AI agent will get the needed information from the network. That is the reason. This is why an AI agent is such an interesting and good tool. And It’s easy to scam with this kind of tool. People think they must not know anything about coding. When. They start to use AI for coding. There are certain prohibited orders. But if the user can somehow circumvent those orders. That allows them to use AI agents for spying. And that is the problem. Companies in the East don’t follow the same rules as companies in the West.  

The biggest threat is in the cases. We expect everybody to play by the same rules. Rules might be the same. But everybody doesn’t follow them. 

Even if AI spies are meant for governmental purposes. 

It’s always possible. That. Corrupt officials give those kinds of tools to hackers. Those hackers can work for local mafia like Triads or Yakuza. And that can cause global risks.  If. We think about the situation. That OpenAI faced. When its agents attack government homepages. 

It’s possible. There can be some hostile agents in those corporations’ services. Those agents might want to steal that code. And when we read news that OpenAI stops training its most powerful AIs. At the same time, we can ask. Do they stop just the AI training? Or do they also stop development of the LLM, the AI’s core? Are those AI cores really protected against AI agents that are used for hacking? 

AI agent. Can perform automatic attacks around the clock. It can change the IP addresses that it uses. And that makes this kind of attack very hard for defenders. The problem is that.  AI agents. They are not separated. From. The AI core. The thing that we see as an AI agent is the thing that runs on servers. And those servers are data centers. The internet makes AI mobile. The code runs on servers. And the user sees the answers that the AI gives. 

https://futurism.com/science-energy/research-fly-brain-matrix

https://www.profolus.com/topics/scientists-copied-fruit-fly-brain-put-inside-computer/

https://www.techspot.com/news/114003-openai-pauses-training-most-powerful-ai-models-after.html