Artificial general intelligence can help us solve dead-end problems, speed up and optimize
How versatile is general AI? How will it work?
The term itself sounds like Artificial general intelligence (General AI, AGI) or “artificial general intelligence” (AGI). It is translated as general, true or universal artificial intelligence.
So far, it has only been possible to create a "narrowartificial intelligence "(AI): it can perform very specific tasks. According to Arkady Sandler, director of the MTS AI Center, it is incorrect to oppose AI and AGI, since the artificial intelligence that exists today helps us automatically perform any point functions at the human level or even surpassing it.
In the case of AGI, the problem solving process occurs differently. Such a system will be able to draw its own conclusions based on the information received, as well as self-learn and even become self-aware.
Why do we need common AI and where will it be used?
Generic AI will be able to use global knowledge aboutall types of problems, as well as related data to solve a specific issue. Such an AI will take into account the context when analyzing the text, as well as other types of information, for example, auditory and visual.
Formally, AGI is suitable for any task, sinceWith his data-driven approach, he can answer any question. Despite this, there will remain areas in which AGI will not find its application: the fact is that in some areas AGI is already sufficient for effective functioning.
But there are also serious differences.According to Daniil Kireev, a leading researcher at facial recognition product developer VisionLabs, general AI will expand the list of problems and issues that the algorithm will deal with, for example, complex compound tasks that require information about different areas of life.
This could be recognizing a person's emotions. To make an unambiguous conclusion about this, you need to process data on facial expression, heart rate, posture and spoken words.
Who is developing general AI?
Such a project was launched at Google.The company plans to train AI in a variety of skills so that it becomes universal. It is noted that the developers are going to use multimodal models that process and understand all incoming signals, regardless of the method of transmitting information.
The company is now only at the initialdevelopment stage. But the plans are to create an AGI that can solve millions of different problems, summarize data and combine them. The basic capabilities of the neural network have already been tested in projects such as Google Switch Transformer, Gshard and others. Tests of these models have already shown that such systems consume ten times less energy than conventional neural networks.
Also in China they created the Tianjic chip, in whichmethods of neurobiology and a scheme for simulating the human brain were combined. The chip has more than 150 cores: they repeat the functions of brain neurons, and can also simultaneously perceive information and convert it into nerve impulses for a pulsed neural network and vice versa.
Burst neural networks use biologically plausible neuron models to process information. This is similar to what happens in the human brain.
IBM also worked in the same direction.The company launched a convolutional neural network based on the TrueNorth neuromorphic chip, which mimics the human brain. It consumes thousands of times less power than modern processors.
TrueNorth is a self-sufficient computinga system that combines a processor and memory. Each chip consists of 4,096 neurosynaptic cores with 1 million programmable neurons and 266 million configurable synapses.
The company continues to develop in thisdirection and at the end of summer 2021 presented new chips for Tellum corporations - this is a new generation of z-processors that are designed to work with business-critical tasks.
What threats can there be from general AI?
General AI has the potential to profoundly change our relationship with technology. The idea is that machines will perform all functions as well as humans, and maybe even better.
This means that they will be able to create othersmore powerful devices. Once this happens, humans will forever lag behind the development of general AI, which will learn super quickly and produce more and more new solutions. This process is called "singularity".
As a result, it may happen that general AI will become the last invention of humanity, and then we will simply watch how the world around us changes, without realizing what is happening.
So far, humanity has not been able to create a commonartificial intelligence, but you can see how this area is progressing: the quality of communication of voice assistants, speakers, chat bots or the development of autopilot in electric vehicles.
But now it is too early to talk about full-fledgedinteraction, because the system is still imperfect, therefore, while making decisions, AI is only a tool that can be relied on, and not a full-fledged unit that can make decisions.
Read more:
Look at pictures of two merging galaxies, taken 9 years apart
A new way of delivering a substance makes cancer cells self-destruct
American scientists received a record energy yield during the experiment