Much Cheaper and More Practical than AI... People Exploring the 'Intelligence of Bugs' [Tech Talk]

AI Implemented Through 'Kkulbeol Intelligence Reverse Engineering'
Thousands of Times Cheaper Than Mega AI
But Much More Efficient for Industrial Applications

The goal of artificial intelligence (AI) scientists generally culminates in creating 'human-like machines.' The development of 'general artificial intelligence,' pursued competitively by renowned AI research companies such as OpenAI and Google DeepMind, ultimately aims to create machine intelligence that resembles humans yet surpasses them.

Honeybee. [Image source=Pixabay]

However, there are inventors who draw inspiration from entirely different sources. Is it really necessary for AI to catch up to human intelligence? Much lighter, cheaper, and even practical, 'insect intelligence' can also be an alternative.

Big Techs Racing Toward '100 Trillion Parameters' AI

Currently, artificial intelligence (AI) aims to reach the level of the human brain.
[Image source=Pixabay]

Today, many AI technologies target the 'human brain.' Our brain consists of 86 billion neurons, each connected by 'synapses,' which serve as information transmission gateways. The total number of synapses in the human brain reaches 100 trillion.

In AI neural networks, parameters play the role of synapses. So far, high-performance AI has increased parameters as much as possible to catch up with human intelligence. Among these, the latest AI models possess parameters in the quadrillions.

However, as AI size grows exponentially, training and inference costs also increase. The number of high-performance processors that must be integrated into data centers rises, leading to higher electricity and heat management costs. AI data centers are already causing headaches worldwide over electricity consumption and land costs.

Does AI Have to Be a Giant Model?... 'Insects' Are Enough

But does AI really have to reach the 'intellectual level of humans'? Humans are highly advanced beings among animals living on Earth. We are not only capable of simple hunting or reproduction but also skilled at forming complex social relationships and abstract thinking. If the robots we need right now are only for automating specific tasks, AI does not necessarily have to catch up to humans.

Opteran developed an AI model aiming for intelligence at the level of small insects like honeybees.
[Image source=Opteran homepage]

For this reason, some have shifted their focus from 'human intelligence' to 'insect intelligence.' A representative example is the startup 'Opteran,' founded in 2020 at the University of Sheffield in the UK. This company is the world’s first to reverse-engineer the brain of a honeybee and apply it to autonomous drones.

As mentioned earlier, humans have 86 billion neurons. The brain of a bee has only about 1 million. Despite having a brain that is extremely primitive compared to humans, bees can select desired goals in complex environments and possess a homing instinct to return home.

In other words, creating AI at the honeybee level can meet most industrial needs. Therefore, if the principles of how an insect’s brain operates can be implemented on a computer, the cost of developing and integrating artificial intelligence would be thousands of times cheaper.

Simpler Than Humans, But Much More Agile Insects

The advantage of 'insect intelligence' is not only in cost. The ability to visually select desired goals and recognize locations, as exemplified earlier, belongs to the hot topic in computer vision research called 'SLAM (Simultaneous Localization and Mapping).' This technology is applied from autonomous vehicles to robotic vacuum cleaners.

Traditional SLAM creates a 'high-resolution map' of the surrounding environment using expensive advanced sensors, then uses AI to recognize its location and set routes within that map. Both the hardware and the dataset size required for training, as well as the AI model size, are very demanding high-performance models.

Example of SLAM mapping. Using sensors, a robot's surrounding environment is created as a 3D map. The process itself is complex and requires learning a vast amount of data.
[Image source=YouTube capture]

However, the honeybee intelligence developed by Opteran is much smaller than SLAM yet delivers similar performance. Thanks to this, it is much more 'agile.' In other words, robots equipped with SLAM are constrained in their actions due to processing large software. Their 'response speed' decreases. But honeybee intelligence is much smaller and simpler, allowing immediate movement. This makes it much more suitable for narrow logistics centers or factory interiors where unexpected situations can occur at any time.

Reverse Engineering Nature... The Value Shown by Basic Scientific Research

Honeybee intelligence demonstrates the efficiency of 'natural algorithms' created by ecosystems over hundreds of millions of years. Insects can perform almost all survival behaviors sufficiently with brains smaller than a pair of tweezers.

Moreover, insect intelligence is a case that reminds us how important 'basic science' is in creating new industrial fields. Opteran started from a unique project in the Department of Computer Science at the University of Sheffield. They have studied for ten years how insects respond to stimuli in the world, and as a result, have reached a level where the information processing of insect brains can be mimicked by computers.

Innovation often begins with a small curiosity. David Razan, CEO of Opteran, explained in an interview with the engineering media 'The Engineer' why he said, "If you want to see cutting-edge autonomy, you should look at the garden, not Silicon Valley."