Free Shipping on orders over US$39.99 How to make these links

AI without computers – Advanced Science News


Artificial intelligence, or AI, is ubiquitous and is integrated into almost any field or application. The progress made over the past few decades has been impressive, with achievements-such as DeepMind’s AlphaGo beating the world’s most Go player in 2016 and the application of LinearFold to predict the secondary structure of the sequence. of SARS-CoV-2 RNA in just 26 seconds — demonstrating the ever-growing capabilities of these systems.

There is one caveat though; Running AI requires a lot of power and data, and computer hardware can’t keep up. Integrated circuit chips come in capacity even when the structures of the chips and circuit components can be small. There is a limit to how far we can go physically.

“The semiconductor structures of computer chips are now approaching the size of several nanometers and quantum uncertainty is starting to erode the electrical insulation, causing the chip to fail,” explains Li Lin, a post-doctoral researcher in the Department of Mechanical and Aerospace. Engineering at George Washington University, focusing on plasma physics, plasma chemistry, and machine learning. “It’s very difficult to keep making small and small chips.”

Lin and his advisor, Michael Keidar, a professor of engineering at George Washington University, thought alternative hardware could be explored, and as a result, they returned to chemistry.

During a chemical reaction, the number of reactive species within a system is defined and carries the product sets. “The chemical system naturally has a network of all its chemical reactions, which is the‘ chemical pathway network ’,” Lin said. “This network can be trained like a regular artificial neural network. One can create a material with chemical behavior that can be used in AI.”

Anything with a temperature higher than -273 ° C has the potential to randomly move the molecules and if a molecular collision occurs, it can lead to chemical reactions. This happens all the time, anywhere in the universe, from star formation to our own biological processes.

“In order to build AI from these systems, we need to manipulate and maintain collision possibilities,” Lin said. “In other words, chemical reactions are a form of data processing and nature can continue to run it automatically, we just have to find a way to use it.”

It could hypothetically transform any kind of object with a lot of chemical complexity into an alternative hardware carrier for AI. In other words, AI can be built without digital computers.

A winning AI plasma

In a study published in Advanced Intelligent Systemthe team trained low-temperature helium plasma to play the tic-tac-toe game.

The concept is based on creating a data processing unit from a network of chemical reactions that take place within an isolated chemical system – such as a plasma. Scientists have to find a set of chemical parameters for the system, such as pressure or temperature, so that the system can release accurate information in real time according to a dynamic input, making the system a programmable analog computers that operate at a molecular level, and can process complex information in nanoseconds. Chemical parameter sets are the “software” of an analog computer, to determine chemical reactions. In other words, the computer thinking process.

However, writing the program for it is very different from coding on a conventional computer. Scientists consider the map of chemical reactions in the system, which is the “chemical pathway network”, as an artificial neural network. The chemical parameters are the network weights, and the concentrations of the species are the neuron values. Using modern machine learning techniques, hardware can be trained (programmed) for specific missions. The purpose of the training was to find the chemical parameter set for the mission. Once this is obtained, the programming is complete. If the system is used after training, users can transfer the chemical set to different missions, such as running different software on one computer.

To demonstrate this, Li and Keidar teach plasma to play tic-tac-toe. But how does a plasma actually “see” the board to play the game? Li and Keidar achieved this by feeding the plasma with a gas mixture. The 3 × 3 game board is set up using nine different gases – NO, N2O, H2Oh, N2or3He, O2H2and NO2 – each represents one of the game’s nine tiles. The status of the board can thus be represented by the mixing ratio of these gases: a lower ratio of one type of gas means the plasma marker of that tile, while a higher ratio means the marker of the opposite. Once the plasma has received that gas mixture, the chemical reactions in the plasma begin to work, producing excited atoms and molecules. These excited species emit light signals that represent the next movement of the plasma. Thus, a different state of the board means different mixing ratios of these gases leading to a series of different chemical reactions in the plasma. The plasma will thus release a different next step. “It’s plasma‘ thinking ’using the chemical pathway in its network,” Lin added.

The light signals are interpreted to update the status of the board so that the opponent, either a human or a computer player such as the random-move player used for training and testing, can play. Next, after the opponent has moved, the gas mixture, with the new mixing ratios representing the new state of the board, will flow into the plasma for its consideration.

Training was achieved by testing chemical parameters with small changes. A change to a lower score will be discarded, and the parameters will be restored. However, a set of modified parameters that will enable plasma play to be better recorded and the next modification (generation) will be based on it. This is a common evolutionary algorithm used in the world of machine learning.

Through training, plasma eventually achieves a high win rate against the random-move player, indicating that plasma does not play randomly, but has logic and develops its own strategies. “This is how plasma shows us a chemical -based AI,” Lin said.

During training, Lin and Keidar also learned that plasma became more aggressive and tried to win games with little movement. However, they did not design any scoring system to determine that “winning a game faster is better” for the plasma during its training.

“Plasma concluded that the loss of a winning move would signal more uncertainty in the game,” Lin said. “So, it’s learned not to miss any chance of winning, and to start winning a game as quickly as possible.”

The plasma even starts to show “fork movement”, making two winning positions available while the opponent can only block one of them. “It’s interesting to see a material that concludes that the best defense is a good offense, all on its own,” Keidar said. “These are typical behaviors of an AI who learns a game, but now, materials with complex chemistry can also reflect such behaviors.”

Processing information in this way consumes less power compared to modern digital computers where electrical charges are transferred around circuits, generating and transmitting digital signals. “Using a digital computer is an indirect way of processing data compared to a chemical system,” Keidar said.

Chemistry -based AI has many advantages. First, unlike other intelligent materials, such as soft or adaptive materials that reflect only basic logic and memory, the chemistry-based approach achieves a higher level of intelligence. “Playing a board game is not possible for other intelligent materials,” Lin said.

The level of intelligence in a chemical reaction network can also be controlled by manipulating the complexity of the system. This is similar to the theory of artificial neural networks, where multiple neurons with multiple connections mean a stronger ability to process information. “For the chemical system, one can only multiply the reactors to increase the complexity of the chemical system,” Keidar said. “Of course, the system can also be used as a data processor without a high level of intelligence, like a modern CPU in a personal computer.”

This theory can be applied beyond plasmas of any materials with a chemical behavior complex enough to make it suitable.

“We can’t limit ourselves to making computer hardware that rely on semiconductors, because this can limit their application,” Lin said. “We have a future idea of ​​integrating the self -organization of molecules with chemical -based AI, and could be a way, for example, to use any surface as a computer screen, which can be used if necessary. people.

“Creating self-organizing patterns is very common in plasma,” he continues. “The remaining technical gap is the self-organization of molecules with chemical-based AI, so we can use it anywhere and anytime.”

The next question, according to the team, is how far they can go with these systems. Maybe even far enough to make the thing not only intelligent but self-aware.

Reference: Li Lin and Michael Keidar, ‘Artificial Intelligence without Digital Computers: Programming Matter at a Molecular Scale’ Advanced Intelligent Systems (2022) DOI: 10.1002/aisy.202200157

Image credit: Josh Riemer of Unsplash



Source link

We will be happy to hear your thoughts

Leave a reply

Info Bbea
Logo
Enable registration in settings - general
Compare items
  • Total (0)
Compare
0
Shopping cart