Файл: Основные проблемы искусственного интеллекта в науке.docx

There are three types of artificial intelligence:

Weak AI (ANI)-an artificial intelligence that specializes in one area. This AI that can beat the world chess champion at chess, but that is the only thing it does. This is the only AI that exists today: Google Assistans, Google Translate, Siri, Alice.They all use Natural Language Processing (NLP). By understanding speech and text, they are programmed to interact with people in a personalized, natural way.

Strong(AGI) is a strong artificial intelligence, as close as possible to the abilities of human intelligence.

For comparison, the power of the human brain, according to data by I.A. Kalyaev, Academician of RAS, is 10 zettaflops, and the most powerful supercomputer today is Fugaru with productivity of 537 petaflops. That is, in order to simulate the human brain, we need to raise the performance of

supercomputers by three or four orders of magnitude more[5].
Super(ASI)-that's super artificial intelligence that outperforms humans in all areas of activity. It is the way of the future.

What is the future for humans with ASI? To this question, the opinions of scientists studying AI are divided.

Many, including Professor Vernor Windge, scientist Ben Herzl, Sun Microsystems co-founder Bill Joy, and futurist Ray Kurzweil share the opinion that ASI will appear soon, in the next few decades, and is capable of defeating mortality and expanding our consciousness, when humans and machines will merge into one. Humans will be able to upload their minds to the cloud and renew their bodies through intelligent nanobots injected into the bloodstream. In addition, superintelligence could help us get rid of incurable diseases and cope with global warming.

Others like Microsoft co-founder Paul Allen, research psychologist Gary Marcus, computer expert Ernest Davis, and technopreneur Mitch Capor believe that thinkers like Kurzweil seriously underestimate the scope of the problem, and think we are not that close to a tipping point. The Kurzweil camp argues that the only underestimation that occurs is ignoring exponential growth, and one might compare the doubters to those who looked at the slowly blooming Internet in 1985 and claimed it would have no impact on the world in the near future.

"Doubters" might paraphrase, saying that it is harder for progress to take each successive step when it comes to the exponential development of intelligence, which negates the typical exponential nature of technological progress. This is the illusion of rapid progress. Progress in solving easy problems does not equal progress in solving hard problems. Just because a computer beats a human being at mind games does not mean it is smarter.

The third camp, which includes Nick Bostrom, does not agree with either the former or the latter, arguing that all of this absolutely could happen in the near future and that there is no guarantee that it will happen at all or take longer.

Others like philosopher Hubert Dreyfus believe that all three of these groups are naïve to believe that there will be a tipping point at all, and that we will probably never get to ASI. In 2013, Bostrom conducted a survey in which he surveyed hundreds of AI experts at a series of conferences on the following topic: "What would be your prediction for reaching human-level AGI?" and asked for an optimistic year (in which we would have AGI with a 10 percent chance), a realistic guess (a year in which we would have AGI with a 50 percent chance ) and a confident guess (the earliest year in which AGI would appear with a 90 percent chance).[6] Here are the results

Average optimistic year (10%): 2022

Average realistic year (50%): 2040

Average pessimistic year (90%): 2075

The average pollster thinks there is a 90 percent chance of AGI by 2075.

A study conducted by James Barratt at the AGI Conference talks about the opinion of AI scientists as to which year we will get to AGI: by 2030, 2050, 2100, later, or never. Here are the results:

By 2030: 42% of respondents

К 2050: 25%

К 2100: 20%

After 2100: 10%.

Never: 2%

The results are similar to Bostrom's findings. Expected years of AGI appearance: 2030-2040.

AGI is not a tipping point for humanity like ASI. When is super intelligence expected to appear according to experts?

Most are inclined to conclude that a rapid transition (less than 10 years) from AGI to ASI will occur with a 10% probability, but in 30 years or less the probability of transition rises to 75%.

That is, experts believe that the emergence of ASI is possible by 2060-2070.

Of course, all of the above statistics are speculative and simply represent the opinion of experts in the field of artificial intelligence, but they also indicate that most interested people agree that ASI could appear by 2060.

Artificial Intelligence Problems

1.Energy and Cost

One of the main problems is the high energy consumption to execute algorithms. Machine learning and deep learning are the stepping stones of artificial intelligence, and they require an ever-increasing number of cores and GPUs to run efficiently. There are various fields where deep learning technologies are needed, such as asteroid tracking, healthcare, unmanned transportation, and more. They require the computing power of a supercomputer, which not only consumes a lot of power, but is also cost effective.

A Deloitte report states that about 94% of businesses face potential problems with artificial intelligence when implementing it.[7]

Small and medium-sized organizations have a lot of trouble when it comes to implementing artificial intelligence technology because it is expensive. Even large firms, such as Facebook, Apple, Microsoft, Google, Amazon, allocate a separate budget for the implementation and deployment of artificial intelligence technologies. A possible solution to the problem is to provide special programs and benefits from the state.

2. Data

Data is the basis for the development of artificial intelligence. The quality and quantity of data determines the speed of machine learning. Data quality considers many aspects, including consistency, integrity, accuracy and completeness. Modern systems must be aware of the quality of data input/output. They must instantly detect potential problems and identify, inaccurate, or incomplete data. Erroneous algorithms created using an inappropriate data set can leave a huge dent in an organization's profits. The solution to the data problem is to spend time evaluating and scaling with careful data management, integration and research until you have a clear data.

Current AI-based applications require not only quality data, but massive data. Amazon, Google, Facebook, etc., they lead the way because they have access to a lot of data. Not all companies have access to massive data.

3.Lack of skilled resources and knowledge

Deep analytics and machine learning in their current form are still emerging technologies. Thus, there is a shortage of qualified employees. Representatives of the Russian IT industry say bluntly that domestic universities pay almost no attention to training in the field of artificial intelligence.[8]

Also, in addition to AI enthusiasts, students and researchers, there are only a limited number of people who are aware of the potential of AI. For example, to integrate, deploy and implement AI applications in the enterprise, an organization must be aware of current AI advances and technologies, as well as its shortcomings. The lack of technical know-how hinders the adoption of this niche area in many of the organizations. Therefore, enterprises need a specialist to identify obstacles in the deployment process.

As a consumer and developer of artificial intelligence technology, we need to be aware of both the merits and the challenges of AI deployment. Knowing these details of any technology helps the user/developer reduce the risks associated with the technology as well as take full advantage of it.

4.Ethical.

One of the major AI issues yet to be addressed is ethics and morality. We need data to train our AI algorithms, and we need to do everything we can to eliminate bias in that data.

For example, the ImageNet database has many more white edges than non-white edges. When we train our artificial intelligence algorithms to recognize facial features using a database that does not include the correct balance of faces, the algorithm will not work well on non-white faces, creating a built-in bias that can have a huge impact. It is important to eliminate as much bias as possible as we train the AI.

As we use more and more artificial intelligence technology, we are asking machines to make more and more important decisions. Who will be responsible for these decisions? How will the responsibility shift to the system? Will the system be responsible for the decisions?

Privacy (and consent) for data use has long been an ethical dilemma of AI. We need data for machine learning, but where does that data come from and how is it used, will it be protected?

Huge companies like Amazon, Facebook, Google are using artificial intelligence to crush their competitors and become virtually unstoppable in the marketplace. Countries such as China also have promising AI strategies backed by the government. Russian President Putin said: "Whoever wins the AI race is likely to become the ruler of the world."

The inequality between AI leader countries and developing countries will only grow, it is a serious problem in the world.

5. Trust

Algorithms are becoming more complex every day, an issue of trust is emerging, regarding its ability to make fair decisions and improve the quality of humanity. One of the most important concerns for AI is the unknown nature of how deep learning models predict outcomes. How a particular set of inputs can develop solutions to various problems is difficult for a layman to understand. In addition, the algorithms of most AI-based products or applications are kept secret to avoid security breaches and similar threats. For these reasons, there is no transparency about the internal algorithms of AI products, making it difficult to trust such products.

6. Unemployment

According to a Mckinsey report, artificial intelligence could replace 30 percent of the world's current workforce. According to AI expert Kai-Fu Lee, 40% of the world's jobs will be replaced by social bots in the next 10-15 years. Low-skilled workers will be the hardest hit by the changes. With the development of AI, even for high-paid, highly skilled workers, they become more vulnerable to job losses because companies make higher profits by automating their work.

Transportation is one area where the above AI problems exist and require attention. Decisions in neural networks are made based on large amounts of data, the logic of decisions itself is hidden and is not clear not only to ordinary users, but also to experts. The more we rely on neural networks, the more important it becomes to understand the principles by which they make decisions. After all, when AI interprets images that are obvious to humans incorrectly, such as a picture of a baseball with shaving foam for a cup of coffee, or does not recognize a black car at night at all, as happened in an accident involving a Tesla[9]. These errors are critical if we are going to hand over vehicle control functions to systems. Many people overestimate the power of deep learning and its reliability. People think if an unmanned car can drive on the highway, it will successfully handle city traffic after refinement. However, there is a huge technological chasm between the two possibilities.

Deep learning is devoid of human intelligence, and speech and image recognition are not intelligence, but only the smallest fragments of it.

To create artificial intelligence similar to human intelligence, we need to move away from systems that use learning on statistical correlations and move to systems with basic human-like understandings to comprehend the world. We need to put into AI the principles that help people learn and explore the world around them: the ability to abstract, composition, causal representations, understanding that objects exist over a period of time. We need to lay the foundations in artificial systems in the form of an understanding of time, space, and causality. Therefore, when creating AI, computer science must be enriched with knowledge from other disciplines, including cognitive sciences.

While there is no strong intelligence (AGI) and super intelligence (ASI), one way to solve problems in transportation and other areas is human-machine tandem. This symbiosis mutually controls, complements each other and helps make the human-machine-environment system more efficient and safer. AI cannot yet replace humans and must complement them.

Conclusion

AI can bring enormous benefits, for example, in medicine, education, food distribution and aid, more efficient public transportation, and the fight against climate change.It is a technology that, like the Industrial Revolution, will change human history. However, there are serious ethical, social, and security concerns associated with the rapid spread of AI technology.

AI poses serious regulatory challenges because of the way it is funded, researched and developed. The private sector drives AI progress, and governments rely heavily on large technological companies to build their AI software, contribute their AI talent, and make AI breakthroughs. In many ways, this is a reflection of the world we live in, as large technology firms have the resources and expertise.

However, without government oversight, further application of AI's extraordinary potential will actually be given over to attracting commercial interests. This result provides little incentive to use AI to solve the world's greatest problems, from poverty and hunger to climate change.

Currently, governments are running a race to develop and deploy AI applications. Despite the international nature of this technology, there is no single policy approach to regulating AI or the use of data. It is imperative that governments provide "guardrails" for private sector development through effective regulation. But this does not yet exist in the U.S. (where most development is taking place) or in most other parts of the world. This regulatory "vacuum" has important ethical and safety implications for AI.

Some governments fear that imposing strict rules will discourage investment and innovation in their countries and deprive them of a competitive advantage. This attitude runs the risk of a "race to the bottom," with countries competing to minimize regulation in order to attract large investments in technology.

Perhaps the most promising approach to public policy on AI is the risk-based approach proposed by the EU. It would ban the most problematic uses of AI, such as AI that distorts human behavior or manipulates citizens in subliminal ways. It is equally important that the use of AI by governments be clear, consistent, and ethical, with respect for human rights obligations. Opaque practices by governments can contribute to the perception of AI as a tool of repression.

It will be difficult to trust AI until there is a regulatory framework governing its funding, development, and use, but global action to establish clear and effective regulation and accountability can help build confidence in the safe and ethical use of AI.

AI is not a "magic pill" for all of humanity's global problems. But it can be a major aid to humanity's transformation into a new era. The era of smart cities, safe roads and cars, space travel and new discoveries.