In recent years, the concept of the superintelligence captured the imagination of scientists, entrepreneurs, and people who like technology worldwide. Popular by British Physical Nick Bostrom in his influence Superinteritence: road, dangerous, strategyBostrom presented the idea of Ai beyond human ability, sparking both excitement and concerns about the future of humanity.
Fast forward decades, Sam Altman, the CEO of Openai has suggested that the wise can be a new company that focuses on secure superiniblelelligence. With these bold estimates, pressure questions become: we are close to Indeed this, and what does it mean for the future?
Understanding the level of Ai Intelligence
One of the most interesting way to break the complexity of the AI wisdom is from the work of the US ceedoes and her team at Google. They offer a six-level comprehensive framework: No AI, emerging, abilities, specialists, specialist, virtuoso, and superhuman. This classification also distinguished between ai narrow and common, dramatically focus on specific tasks, while the general system has a wide range of skills.
For example, a narrow Ai can act well in some domain – Deep Blue, a classical Garry of Virtuoso. Meanwhile, such as Alphafold, which predicts protein structures with prominent accuracy, the most narrow accuracy. These specialty tools are powerful, but their extent is limited.
On the other hand, the common AI is very challenging to move forward. Despite the recently, the language model, such as Chatgpt also falls in the category “of the universal human. This indicates that the levity, the common Ai is also distant – even technological advances are unpredictable.
Current AI Capacity: Reality vs. expectations
The current AI’s intelligence assessment depends on the indicator, such as a standard testing For example, Winograd Schema Challenge is designed to test the ability of AI to understand the skills and doubt in the university. On the other hand, the mathematical reasons are similar to resolve the problem in the mathematics or graduate courses. Model like Dall-e, which creates an image, may be seen as a quality of the art; Midjourneys, another AI model, has shown the ability to create a very natural character that was arguing with the current Midjourney. However, the two systems also face the restrictions and sometimes, no mistakes of human artists do, even more advanced manufacturers for updates. The debate continues among the researchers, by some people suggest that AI model, while others argue that true wisdom.
Despite reference to a complicated reasoning, the previous study shows that even more comprehensive modes will also take a comprehensive model instead of a real understanding. The difference between recognition and actual skills emphasizes that the highest wisdom may not be near that some are as close.
Will AI progress keep pace?
The rapid advancing of the AI technology in recent years has believed that many people believe that wiseness is accelerated. The company is investing significantly in the development of AI, and some are expected to appear within decades. However, the dependence from the information generated by humans to train these models to their evolution.
To overcome these restrictions, researchers are searching for more efficient use of data, generating data, syringement and transfer skills throughout different domains.
However, there are doubts about that AI now, such as Chatgpt, reaching the real ability – a major capabilities against superinterlligence. The latest mode, O1 Preview, there is significant advancement, demonstrates up to reasonable skills and understanding of many complicated tasks. However, it also lacks the ability to achieve full capacity, especially in the area that requires a deception and adjustment of adjustment. Open AI development, which can create a continuous new results and learn adaptation, may be the key to reach this vision.
Risk in extent
While superinterlinessence may also be more years, risks associated with increasingly effective AI systems cannot be overlooked. When the AI system become more independent, their potential for unintentional consequences will grow. Now, the main concern is not ai-obligation to come, but instead of risk associated with excessive AI or assignment.
For example, the introduction procedure that creates our digital experience or conversation format for instructions will be vulnerable to them. In addition, anxiety, masses, individuals to make sense of AI, and the occurrence of the encorptor of the researchers and policy identifies must be solved.
Ambitious balance and security security
If humans reached the point of creation of ai-viable AI, the next challenge is assured that it is in accordance with human values. It is possible to create automatic AI systems that also offer human control and control of a high level. Many in the AI community remain optimistic that the safe mode of disagreement is to be achieved – but the work can be done.
A sharp wise to the sharpness is unlimited, bring a great benefit, but also the risk of health, and the wrong issue. Going forward, researchers must search for new ways, integrated with these and ethical technology of the safest development of humanity.
Source: Conversation