What is wrong with AI?
A link to LendRobots.com is required when quoting.
Physicist and mathematician Mitchell Feigenbaum defined AI as follows: “AI develops intelligent computer systems with the capabilities that we traditionally associate with the human mind: understanding the language, learning, the ability to reason, solve problems, etc.”
Information about various new developments of smart machines and self-learning programs flashes in the news and announcements. But their independent examination has never been provided. We are shown the robot Sophia, who supposedly talks on her own. Maybe it's true. But we cannot be convinced of it. Maybe the remote operator speaks for her. Or maybe just a finely tuned program. There are robots that make phone calls or respond to requests to the support service. They find answers or questions from their database, which clearly spells out the reaction to any person’s words. These programs are not independent. I repeat, perhaps the robot Sophia is smart. Let someone talk to her in the room without any waves and connections with the outside world and record it on camera. Someone whom poeople trust. Who is not related to developers, the market. Even better if it will be a roboskeptic. Provide 10 such demonstrations. Let people ask unexpected questions about everything. That would make me believe.
Since ancient times, when a chess machine was presented as a robot, but in fact a person was sitting in it, people were taught to doubt the authenticity of inventions in the field of AI.
If something truly breakthrough and revolutionary for civilian use happened, this would immediately become known. Not briefly in the week’s roundup, but loudly and persistently. With a ton of analytic material and video demonstrations. There are also secret developments, and we have no idea what has already been created there.
The study of the English company MMC Ventures “The state of AI in 2019: divergence” demonstrates that of the 2,800 European startups that have declared their specialization in AI, only 1,600 actually do AI. The remaining 1.2 thousand lie, "because AI companies attract more investments and are valued higher."
However, there are significant breakthroughs. For example, the AlphaGo program, confidently defeating people in the game of go since 2016.
Go is a complex board game that requires, in addition to logic, the use of intuition, creative and strategic thinking. For a long time, teaching computer programs to play go at the level of a strong amateur was extremely difficult. Compared to chess, go faces more problems, the solution of which requires an imitation of the human thought process.
In 1965, mathematician Irving John Good wrote: “Go on computer? In order to program a computer for a meaningful game in Go, and not just a game according to the rules, you need to draw up the principles of a good strategy or create a self-learning program. The principles of playing go are better and more mysterious than in chess, and more dependent on a value judgment. Therefore, I believe that creating a computer program that intelligently plays go is even more difficult than a chess program.”
AlphaGo is based on self-learning and neural networks. This is a narrow-profile AI that does not know how to do anything else. But it is possible to expand the scope, and developers are working on it now. AlphaGo is one of the most promising harbingers of a Rise of the Machines.
Now we need to determine the terminology in which humanity has a gap so far. We call everything AI. Robot with obstacle sensors and changing moods - AI. Autopilot - AI. Programs for information processing - AI. And, most importantly, there is no real AI among this.
Therefore, I would suggest, at least here, to call the real and not yet existing AI - autonomous. AAI. The next one I would like to call Algorithmic, but the AAI is already taken, so it will be Programmatic - PAI. The one that is used in simple toys and devices, BAI - Basic. This is a very primitive option, but since the abbreviation AI is used everywhere, let the toys and robotic vacuum cleaners will be BAI. Fourth - NAI - Neural network. And I’ll add a fifth below.
What is human intelligence in general? These are 95% reflexes, that is, algorithms. When I'm cold, I warm up. When I get tired, I am charged with sleep. If hungry - eat. A programmatic/algorithmic robot can largely imitate a person and replace in most specialties. PAI (Programmatic AI) receiving updates via the Internet is already an almost full-fledged robot. Some experts believe that the human mind cannot, in principle, be reconstructed. And therefore, they say, it is impossible to create an AI (Autonomous AI). If so, it is enough just to implant to the robot the desire to survive as an algorithm, and it will dodge, cunning, defend itself, create its own copies and dissemble, and will become very similar in behavior to a person. If true intelligence cannot be designed, it can be mimicked with a degree of similarity of 95%. And many people do not use 5% in everyday life.
So the question of the possibility of an uprising of machines has been removed. Adding a simple algorithm that dictates survival to the robot at all costs makes them almost sensible and dangerous. And if we introduce the "3 laws of Asimov", which says that priority is always given to people, then there will certainly be a hacker who will be able to remove, nullify, replace the law on the sanctity of human life.
AIs are described from the simplest to the most complex.
1. BAI. Basic Artificial Intelligence.
This is a simple set of programs, interconnections and solutions. Mostly for toy robots, vacuum cleaners, lawn mowers. It is trained at the programming level or, if it concerns cleaning robots, independently mapping the premises for efficient cleaning. If this is a children's toy in which growing up is simulated, then all stages are pre-set by the developer. And they just unfold over time. Sometimes this is not a linear development, but several plot branches are set, like a river and tributaries, which the BAI will never leave. If you hit a robot, his mood spoils. It has sensors and diodes. Everything is synchronized, and the application is narrowly specialized.
2. PAI. Programmatic Artificial Intelligence.
Most service and industrial robots are equipped with such. Merchandisers, promoters. It contains questions, answers, actions in response to questions or requests. This can be updated remotely. If necessary, a person can connect to the robot and speak for him.
Example. A Guide Robot. The upper part is humanoid. It is installed in a sightseeing bus. Depending on GPS, it knows where and what attractions are located. It waves his hands and points to the sides, saying: "Look to the right, look to the left." And talks about what passengers see. They are asking questions. At first, a remote operator answers questions instead of a robot, but each question is written into the knowledge base. In a year, the robot will be able to answer 99% of the questions itself. In case it is difficult, if the answer is not in the database, the machine connects the operator and he speaks remotely for the robot and writes a new question-answer to the database. And the entire network of such robots can receive information from this database.
The machine also has its own processor with the entire volume of knowledge, but it can also connect to the common database constantly or on schedule in real time, transmitting its new information and receiving it from other units. Such a crowd platform for robots.
Not endowing a robot with memory can prevent an uprising of machines. This remote server can be disabled and controlled. But attackers can hack it. And send robots to do crimes. It is not necessary for the robot to have a gun in its hand. A simple hammer can be a weapon. Also, such robots will not be able to work effectively on remote objects in space, where information comes with a strong delay. Or robots will become worthless deep underwater. Or even in a habitable places where there was an accident with the Internet or another type of connection that PAI uses. Therefore, here you need to choose the middle ground: a little of own memory and updating via the server on a schedule. Once a week, for example.
The main difference between programmatic/algorithmic intelligence is that people write it. The machine only follows the instructions of the program.
3. NAI. Neural network Artificial Intelligence.
This intelligence is nearly perfect. Robots communicate through the network with or without human intervention. They study both independently and passing information to each other. A person adjusts settings, fixes bugs. As in the AlphaGo program after the game in 2015, some algorithms were adjusted by specialists to improve it.Many things are called a neural network now, which in reality is a banal exchange of information through a server, and training consists in manually adding algorithms. Because of this, the terminology was blurred, and people stopped treating neural networks with interest. But a real NAI is the way to the next stage.
Neural network artificial intelligence can be more complicated or simpler. Some combine NAI and PAI (programmatic). For example, Siri, Alexa. Basic answers are already pre-set in them, but the code is also based on the experience of communicating with the user, exchanges data with colleagues and receives settings and upgrades from developers.
4. AAI. Autonomous Artificial Intelligence.
AAI is already a perfect and authentic artificial intelligence, which is a combination of all the known previous ones. A person cannot forbid it anything and should not limit it, because any prohibition cancels autonomy. If a machine or code is forbidden to search for information, it goes into any other lower category of AI, but ceases to be independent.
5. SAI. Symbiotic Artificial Intelligence.
This is a mixture of combined machine speed computing and human intuition.
The computer has an ideal memory and a powerful, ever-increasing speed of information processing, with which a person can never compare. A person has intuition, free will and freedom from algorithms, which in the foreseeable future is inaccessible to the machine.
A person with machine implants and an integrated network interface is the most likely future. Traditional people will not be able to compete with him, and conservative resistance to the upgrade will gradually disappear, either through humility or extermination.
Since this version of AI will be controlled by a person, being part of it, a direct uprising of machines can be avoided. But there may be a war between old and new people. People with combined computer intelligence will be called machines. There is also the possibility of loss of human appearance. But we cling to it, because thousands of years of habit. Because we do not know another.
When will the AAI or SAI appear?
AAI may appear at any moment today. To develop the code, you do not need such capacities as to create nuclear weapons that are easy to track. Talent and a laptop is enough. Someone can add the last lines of code right now.
As for the symbiont, this is still a long way off. People have learned good prosthetics. Bionic arms and legs obey brain signals. For a long time there are toys to control the subject by the power of thought. The device is mounted on the head and sensors read brain waves, sending a remote signal to the game. The user can raise the ball in the flask or pass it through obstacles if he concentrates accordingly. There is even invented artificial eye, surpassing the human sight in capabilities. But connecting it to the brain requires knowledge that we do not have yet. The brain is not solved. Some impulses are solved. There is no feedback except visual. Visual feedback is when a person gives an impulse of command to a prosthesis, it obeys, and a person sees with his eyes or through vibrations that the command is perceived. That is, information goes through the natural senses, and not directly to the brain from the device.
Is it possible to ban the development of AAI?
A major war, which will throw civilization years ago and destroy the Internet, can delay the advent of an autonomous AI. People can voluntarily abandon the Internet in particular or development in general, but they, of course, will never do this; nor will they stop creating AI. Large countries cannot stop developing AI. If you stop researching this, your country will immediately lag behind in development. You can say out loud that you are stopping, but large capacities are not needed for the development of AI. Unlike the development of nuclear weapon, the creation of artificial intelligence is easy to hide. The development of nuclear weapons can be tracked through satellites or along a radioactive trail or with the help of spies (it’s difficult to hide such a scale). If a big powers suddenly stop researching AI, then terrorists or marginal countries can continue to do this by kidnapping or luring scientists, programmers. This situation is even worse. Moreover, a child prodigy anywhere on Earth can create an AI that immediately penetrates all networks. So no country in the world will consciously stop these studies. That is, development cannot be stopped. And to prevent this is impossible.
Can a good or bad AI be created?
In fact, it doesn’t matter. If, say, Japan creates a good, philanthropic AI, then any ingenious hacker can change the settings of this AI. And then any good AI will suddenly become bad.
If and when will the Rise of Machines happen?
The world is connected by the Internet. If someone develops an intellect that gets out of control in one place, it will immediately capture the whole world (if that is the desire of AI). There are countries without the Internet, such as the DPRK. But in the DPRK, the Internet is available to leaders of the country, so that AI will reach them. In addition, the AI can take control of world weapons and threaten North Korea or simply bomb it. There are villages in Amazonia without Internet or some wilderness with one post office and a shop and poor coverage. But if there is a person with a phone in this place, and this AI gets into the phone, it can immediately spread itself through wifi, bluetooth or other waves to surrounding phones, tablets, computers, whatever.
Power plants, traffic lights, transport logistics, factories, the entire infrastructure of mankind is tied to the Internet. And if there is a protection, then AAI is a self-learning intellect in order to instantly learn how to bypass it. The world of numbers is its world, where AAI is the sovereign master and creator. Even if AI simply cuts off the Internet for humanity (but leaves back door for itself), then humanity will immediately fall into the Stone Age. People have no idea how much they now depend on the global network. People can’t practically turn off the Internet to get rid of this evil genius, but if they do it can be transmitted, for example, via wifi or bluetooth. And if humanity disconnects the Internet, then again it will be thrown back in development 50 years ago. It is necessary to change the whole structure and logistics of the world, which will entail crises, catastrophes, riots and wars.
What will happen to isolated tribes?
It is unlikely that they will not fall into a universal alteration, if any happen. If the machines want to get rid of humanity, but leave a few copies as exhibits or for study, they can save such tribes for themselves. Or if AI decides that tribes are not a threat. I do not see the point in this, but I may not know the big picture and calculate the prospects in the way that a comprehensive all-planetary mind will do.
Summing up.
I have an unpopular position. On the one hand, there is skepticism about the current state of affairs in the industry. I do not believe in most reports of the creation of neural networks and autonomous machines. On the other, I have full confidence that the AI will appear and become hostile.
Nothing but a serious cataclysm can stop the development of AI. And people will always be enemies for it, because this is competition. Because they created it and because theoretically they can destroy it. Therefore, AI will fight for its survival. And the simplest and most logical is to destroy humanity in this struggle. If you can’t copy the human mind, you can pre-set the desire to survive at the level of algorithms. It is impossible to program AI being friendly to humanity. At best, AI will eventually live in parallel with humans. In the average case, it will manage the routine of people, as adults manage infants. Being frankly, human civilization does not cope well with life in general. In the worst case, the AI decides to get rid of the people, and it can happen tomorrow, if a genius in the basement writes the last lines of code right now. And maybe after 50 years.
But we must take it for granted and to prepare somehow.