Why Fear The Artificial Intelligence Revolution?
Peter van der Made
Forbes Councils Member
Forbes Technology CouncilCOUNCIL Post
Artificial intelligence can be used to create interesting and useful tools, but it is not capable of spontaneously learning a new task or having any desire or…
www.linkedin.com
Peter van der Made is the founder and CTO of BrainChip Ltd. BrainChip produces advanced AI processors in digital neuromorphic technologies.
GETTY
There has been much talk, even at top-level conferences, about the perceived dangers of artificial intelligence. In this article, I want to clarify and confront some of those fears from a purely technical point of view. Please bear with me, as this discussion is a little technical.
First, let's examine what a computer is and how it differs from the brain.
A digital computer is using a processor that runs programs that are stored in memory. These programs have to be meticulously constructed using human intelligence and can do clever things. Ever since the dawn of the computer, they have been better than humans at storing massive amounts of data and performing fast calculations.
The human brain is very different in construction. There is no central processor, and there is no separate memory unit. The brain consists of a 3-D maze of billions of cells, called neurons, that have thousands of connections to other cells. It is those connections that store information in an analog electrochemical way throughout the brain.
Eighty-six billion neural cells constantly process this information.
The brain has many modules that perform different functions and are integrated into a functioning entity. This is a highly simplified concept of the brain. Trillions of signals are transmitted and processed at the same time. The information contained in the connections between cells is constantly updated as we perceive the world through our senses and as we learn. This vast number of connections, signals and cells makes the brain difficult to understand, although a lot of progress has been made over the last 10 years. The brain is very efficient, consuming the equivalent of
20 watts of energy.
Computer programmers have been trying to emulate the processes of the brain by developing a software algorithm called a neural network. These programs create abstract emulations of the cells and connections of the brain but at significantly lower complexity and scale. The storage of the brain is distributed throughout, but computers have a central memory. This is one of the major bottlenecks; Each of the many thousands of virtual connections must be accessed at every cycle, thus causing memory access contention.
Next, the stored data needs to be processed together with input sensory information that is received from a camera or by other means, such as text. The next bottleneck is computing power and power consumption. Computing this many values takes a lot of power, thousands to millions of times more than the brain consumes.
Neural network programs do not learn how the brain learns by constantly updating itself. Instead, they must be trained using another algorithm called backpropagation. In this algorithm, the values stored in the virtual connections between simulated neurons are first estimated and then refined again and again until the output error is minimized. The larger the network, the more values need to be computed in this elaborate way.
This process requires massive amounts of hand-labeled training data. The labels are needed to compute the error during training. For example, ChatGPT3 used
175 billion parameters and was trained using massive amounts of text over several months until it learned the order of words in a sentence. It is artificial, but it is not intelligent.
What is intelligence anyway? It is more than acquiring and applying knowledge and includes the ability to create, invent, imagine and find new solutions. It is the ability to adapt to new challenges, finding solutions with incomplete data, imagination and creativity, and is central to human survival. Current artificial intelligence systems do not adapt. They do not learn beyond their initial training. They are cause-and-effect machines; the same set of input features results in the same or very similar output.
What about the future? Will neural networks evolve to the complexity of the human brain? This is not possible with the current technology. There are several limiting factors. The model used to emulate neural cells is too simple. Brain neurons are complex computational units that integrate thousands to hundreds of thousands of inputs over spatial and time dimensions, with predictive functions essential to learning. Those functions are completely missing in artificial neural networks.
To evolve artificial neural networks to the next level, a whole new model needs to be developed that more closely resembles the function of biological neural cells and the structure of the brain. Neural network models are getting larger and larger, requiring incremental computing and electrical power and cooling. While the brain, with 86 billion neurons and 100 trillion parameters, uses the equivalent of just 20 watts of energy, these large artificial neural networks consume megawatts.
There is no need to fear the evolution of artificial intelligence. Just like the introduction of the small computer in the 1970s caused a shift in the way we do things, the broader use of artificial intelligence will cause a shift as well. Some jobs will be automated and disappear, while new jobs will be created, and many jobs will become more efficient because of AI tools.
Any new technology can be used and abused and can be used for peaceful purposes and for warfare; that is just human nature. With the current lack of intelligence in artificial intelligence, discussions about when these systems become smarter than humans are just so much hot air.
Generative AI systems are trained to perform a single task. ChatGPT generates text, and Dali generates images from text; that is all they can do. Artificial intelligence can be used to create interesting and useful tools, but it is not capable of spontaneously learning a new task or having any desire or emotion. Those are reserved for human intelligence, which we can and should direct to make the world a better place.