Artificial Intelligence: Future Starts Slow?

The future has come: artificial intelligence systems surround us everywhere and continue to evolve. Some emulate certain types of intellectual human activity. For example, chess computers beat a person or a more complex project AlphaGO plays go with itself choosing the best strategies. Others have turned from AI (artificial intelligence) and IA (intellectual assistance) due to the fact that the computer performs many routine things faster and better than a person.

For example, translation from language to language, including the ability of Skype to translate human speech in on-the-fly mode. Artificial assistants in smartphones, the first projects of autopilot drives, creating images from sketches, imitation of photos, videos, sound, future prediction in certain areas. Even now, my text editor stresses some words hinting at possible errors. All these applications include complex solutions such as machine vision, machine learning, decision-making algorithms, and self-study. Does this mean that Skynet’s shadow hangs over us?

These examples of AI have one thing in common. They find the best solution for the goals that have been set for them by humans. In his novel iPhuck 10, Russian science fiction writer Victor Pelevin described the top level of such AI. He called it “random programming”. A person sets a task as a final state and a program that works as a “black box”, constructs a code that best serves this task. By the way, generative-competitive networks are being actively used today choosing the best algorithms inside themselves without human. The robot in ‘Ex Machina’ movie, passed the test and escaped from the lab (sorry for the spoiler), but still reached the goal set by human. She had no motivation apart of the order. The last scene of the film can be interpreted in different ways: does the robot have any idea of what to do in the real world? Robots in ‘The West World’ movie behave in a similar way, existing within the framework of the “basic needs” set by the creator.

The next evolutionary step of AI that seems natural is the ability of AI to set tasks for itself. Here we face a very human question. What motivation will AI have for setting tasks? Suppose that such a system will be connected to the global network. What will shape the cognitive needs of the AI, how will these needs be prioritized? The human personality is formed gradually under the influence of various factors, both external and internal. Thus, the physiology of particular parts of the brain directly affects the nature, and therefore speed, completeness, accuracy and adequacy of the decision-making. In addition, the humans do not grow up at a sudden but develop gradually, and their willingness to analyze incoming information, whether it be text, sounds, images, sensations or experiences, plays a crucial role. Having ignored these elements in AI, we will create a new type of intellect with unknowable for human motivation and logic. This raises a new question: what characteristics should be used to conclude that AI is functioning correctly? At the output, we will get only the result of the work, but we will not know the initial premises and the sequence of reasoning. Well, if the result is music or an image, you can find logic in the real world. But what if the answer is zero or one word “yes” or just a strange character?

The second global question is what kind of motivation (apart from purely sports interest) does a human have to create a thinking machine. Do we have a need for questions that we ourselves even cannot imagine? After all, to solve the existing tasks IA, an artificial assistant, is more suitable.

Perhaps the only adequate goal of creating such a system may be the practical immortality of consciousness. But how is this possible? The problem of preserving personality is not much about byte-by-byte copying of the current state of neurons. It is likely that in the foreseeable future we will reach a level of technology when recording the state of the neurons of the brain will not be a technical problem. But consciousness is very closely intertwined with physiology, so the question of survival or adaptation of a copy in a non-native brain appears obvious. And, of course, the broader question is whether such a transplanted consciousness will be the same person as the donor.

Creation of a service operating system raises even greater difficulties. In order for the personality to continue to function “as living”, you must first determine all possible structural features of the human brain that affect our consciousness. Then the influence of sensory systems, or rather the conversion of signals from sensory systems into sensations. After the physical factors are determined, it is necessary to understand how cognitive distortions, which influence the formation of our ideas about the world, are formed, how short-term and long-term memory functions, and how the system of analysis and synthesis of information works. You also need to add a software implementation of the hormonal system, including those responsible for the formation of behavior and the emotional component. After creating all this environment we face the problem of the enormous difference in the speed of data processing by humans and computers. Too fast and the electronic brain can cause the electronic person to live countless lives in an instant and come to a state that we, who live once, also cannot interpret.

From this, it follows that the prospects of using AI for copying and preserving a person are very vague. This line of research can lead to the emergence of a new, at best, human-like, type of consciousness with practical objectives unclear to humanity.

The development of technology has led to the fact that the definition of artificial intelligence fits too many heterogeneous concepts that require individual categories. It is necessary to define what we create (definition of AI) and why we create it (criteria for evaluating success):

Helping a person in solving problems that require large calculations or special abilities, which, however, can be algorithmized. The output is an intelligent assistant or trainer: a chess computer, systems for modeling physical phenomena, a big data processor, a data converter (for example, a sketch from a sketch or on-the-fly video effects)
Self-developing system that selects the best solutions for your goals. These are solutions built on machine learning using competitive technologies.
High-level self-developing system that can set tasks for itself and solve them in the best way. The moment of creating such a system will become the horizon of events in the field of artificial intelligence due to the fundamental inhumanity of artificial consciousness.
Creating an analog of the human brain/copying consciousness/personality. The scope of factors that need to be considered today pushes the prospect of creation for an indefinite period.

The first two types of AI are actively developing and improving nowadays, but it’s probably too early to say that the terminator of artificial intelligence is close.