Artificial General Intelligence (AGI) goal has been seen as a happy target for a long time in the world of technology. However, there is an increasing consensus between artificial intelligence researchers: the strategy of reaching AGI with more processing power may be wrong. Although the definitions differ, AGI is seen as artificial intelligence that can successfully perform any mental task that a person can do.
More power may not result
Association for the Advancement of Artificial Intelligence (AAAI) and 475 artificial intelligence researchers participated, according to a recent survey, 76 percent of the participants think that the existing artificial intelligence models do not reach AGI.
In recent years, technology giants have spent billions of dollars to get ahead in the AGI race. Only last year, initiative capital investments in the field of productive artificial intelligence exceeded $ 56 billion. In addition, with the increase in demand for chip manufacturers and AI accelerators, the semiconductor industry has reached $ 626 billion in 2024.
The energy required to run these huge models is increasing rapidly. Companies such as Microsoft, Google and Amazon are turning to nuclear energy agreements to feed data centers. However, despite these huge investments, improvements in the performance of the latest technology artificial intelligence models are gradually decreasing.
Priorities are changing
The results of the survey also indicate that the priorities of artificial intelligence researchers have changed. While 77 percent of the participants focus on the risk-perfect balance of artificial intelligence systems, only 23 percent aim to develop directly. In addition, a segment of 82 percent believes that AGI should be claimed to be the benefit of the public if it is developed by private companies. However, 70 percent opposes the stopping of AGI research until full security mechanisms are provided.
Cheaper and efficient solutions are also being investigated. OpenAI is called the “Test-Time Compute” technique that allows artificial intelligence models to “think more” before producing response. Although this method provides performance increase without the need for large scale, there is nothing magic. Obviously, in the world of artificial intelligence, eyes have been translated into the next revolutionary progress. Until this, we will see the evolved state of the existing models.
