Scientist and inventor S. Wolfram argues that people are more attracted to visual means as a richer form of communication compared to traditional ones, speech and writing, due to the wider bandwidth - the visual channel [2, p. 371-372].
This suggests the conclusion that quasi-intelligence is, rather, possible not in technical evolution, but still, in our natural, but in the opposite direction, that is, in the degradation of natural intelligence (EI) to the level of machine execution of commands, and there is a reason for this. According to the physicist N. Gershenfeld, the next stage in the development of artificial (AI) is the fusion of artificial and natural intelligence [2, p.233]. The technologies of convergence of AI and EI have already been formed. For example, in one experiment, the brains of several rats (3-4) were networked to solve computational problems based on the received data, such as predicting rain based on information about temperature and atmospheric pressure, pattern recognition, storage and retrieval of sensory information. In this case, the test subjects received a reward in the event of a successful calculation. Actually,according to the scientists themselves, it was createdlearning neural network with reinforcement . In another experiment, scientists used memristors to connect a rat's brain to a neural network embodied in a flint (neuromorphic) microchip located hundreds of kilometers away from the subject via the Internet. It should be noted that the difference between these experiments is only five years (the first one was carried out in 2015), and this short period predetermined the signified transition from EI to the merger of EI and AI.
This direction has a rich perspective, both literally and figuratively. The project of the US National Institutes of Health - BRAIN Initiative (NIH BRAIN), launched in 2014, involves a comprehensive study of the brain based on modern technologies and the introduction of new knowledge into practice. In the official reportthe goal is highlighted: "Our task is to understand the schemes and patterns of nervous activity that give rise to mental experience and behavior."
The strategy for this initiative defines:
- describe the brain and make accurate maps of the connections between neurons and glial cells;
- to measure the dynamic activity of cells in the chain under various conditions and in a wide range of behavior;
- using this activity, test causal hypotheses about how the activity of the chain affects behavior;
- using powerful computational resources, analyze and understand the mechanisms by which dynamic patterns of activity in neural circuits generate behavior.
The last point is crucial in this project, namely: using powerful computing resources (see available neural network systems and supercomputers), learn to predict behavior . That is, to learn how to analyze natural intelligence as artificial and lay the foundation for their unification, organization and management.
I must say that this research initiative has a serious financial budget - $ 500 million until 2020 and then $ 500 million annually, until 2025 [3, p. 120], which means that the most serious scientific resources can be involved in this program.
So, the idea of uniting two systems - biological and mechanical - becomes more real if it turns out to simplify the EI, that is, to vulgarize human consciousness. It follows from this that the technical problem lies in reducing the rich sensory world of a biological subject, endowed with self-awareness, emotions and thinking, to technical communication (instrumental interaction), which determines the interaction between devices.
In this context, it is worth referring to the quote by D.K.Dennett: “AI parasitizes human intelligence. He shamelessly devours everything that people have created ... ”- including our vices [2, p. 83], continues the famous philosopher, author of the concept of consciousness, in which the self acts as the center of narrative gravity [4].
I must admit that this idea is very appropriate, since today AI is a statistical object, a mathematical set of functions that creates a working (or not working) model based on the data received - from people or about people. If distortions are embedded in the data obtained by a person, then the learning neural network will also react to distortions in a certain way, which may look like a bias of the system towards a person from the outside. For example, the AI, designed to evaluate the resume of candidates for "technical" positions at Amazon, deliberately discriminated against women . studied on ten-year data, in which jobs were predominantly given to men.
We can unequivocally say that the “simpler” a person himself is, in a subjective sense, the more accurately it is possible to determine, and therefore predict, his preferences based on the statistics obtained.
In the near future, we should be concerned with the problem of the opposite nature, namely, the movement of a person, or rather human consciousness, into virtual space. For those whose interests in the present and in the future are the development of AI and control of the AI, this is a more urgent task. “We have good models of images and texts, but we lack good models of people, human beings are the best examples of thinking machines,” says Tom Griffiths, professor of computer science, culture and technology at Princeton University [2, p. 178].
Here it is necessary to recall what was written above, namely: the creation of a systematic approach to the study of human behavior at the level of the neural structure within the framework of the BRAIN Initiative project (see above). In support of this, we can add the following statement by T. Griffiths that it is possible to bring a computer closer to human capabilities, “... by identifying human prejudices that form human cognition” [Ibid, p. 179].
So the study of behavior in the digital space concerns not only the study and quantification of consumption matrices for their implementation as training models of neural networks, but also, first of all, how a person thinks and why he does this at all. The difference between these two types of actions lies in the attitude towards the subsequent paths of development of AI: in the first case, it is a consumer model, in the second case it is epistemological. Conditional dichotomy - consume / philosopher. This dichotomy, in my opinion, is the main concern of experts about the supposed paths of development of the future superintelligence: aggressive actions against humans up to destruction or evolutionary development to an independent species and joint coexistence.
Conclusion
A team of researchers from various universities in the United States and Canada created a computer modelcapable of learning from single examples (one-shot learning). This ability is given to a person from birth and is available from an early age. The researchers conducted a series of "visual Turing tests" on various examples, where their model showed creative generalizations that in many cases are indistinguishable from human behavior. This is one of the basic criteria of the human mind: learning to learn ("learns to learn"), which was previously only available to natural intelligence. By starting to learn in this way, AI can create conditions for itself favorable for its development. The most favorable environment for AI, where it will have a significant advantage over humans, is virtual space, in which we, with pleasure, spend more and more time.
Moving human consciousness into a space where AI has advantages over EI, for example, an exorbitant gap in the speed of calculations, is not a question of the future and has nothing to do, at least for now, with the script of the famous movie "The Matrix". Although A. Pentland believes that it is possible to create a human network, according to the principle of a neural network based on machine learning, where selected individuals will play the role of neurons, the scientist does not have a scientific methodology that ensures transparency of such selection [2, p.263-279].
The fact that the unification of brain structures is technically possible, we have seen in experiments with mice. And the fact that a computer model is able to learn to learn is also a fact.
If you combine these two features, you can get a beautiful utopian vision of the future. A neuromorphic computing network based on living human brain substrates and a self-learning agent embedded in this system, which also has full access to external data of accumulated information. In this case, the effectiveness of the convergence of natural and artificial intelligences will tend to the maximum, whether at the same time a person will remain an independent subject, this is a question.
PS Utopia that may soon be a place
Somewhere in the sparsely populated areas of Southeast Asia, sharply discordant with the surrounding landscape riot with all kinds of vegetation, there are huge white hangars with a strange HBRT emblem on the walls. The territory, for many kilometers around, is surrounded by high - four-meter, lattice energized fences. And not a single living soul around, only autonomous observation systems - drones belonging to the private military company Black Rock, from time to time, patrol the area.
In hangars, in even rows in special conditions, there are thousands of human bodies with connected life support systems, for a single purpose: they are all connected into a neural network - a single living brain that plays the role of a supercomputer. This "computer" is owned by Human Brain Resources Tech., Whose operating office is located in the most fashionable skyscraper in Singapore. HBRT is a subsidiary of Goodle Corporation, previously infamous for its experiments with human consciousness. Now it is a leader in the cloud services and computing market, but the main secret of the company's success is forecasting and predicting probabilities in various fields of activity with the highest degree of accuracy in the world.
Footnotes:
1. By the way, this idea was visualized in popular culture in the Devs TV series, where in the first episode a group of developers demonstrated to the employer a predictive model of the dynamic behavior of a nematode based on the analysis of data from a living sample.
Literature
1. Rozin V. Semiotic studies. M, 2001, - 256 p.
2. Brockman J. Artificial intelligence - hopes and fears. M, 2020, - 384 p.
3. Brain Initiative 2025 braininitiative.nih.gov/sites/default/files/pdfs/brain2025_508c.pdf
4. Dennett D. Consciousness Explained. 1991, - 511 p.