"If I gave AI absolutely every test you can imagine, if you make a list of tests and present them to the computer industry, I would guess that in five years we will do well on every one of them," Huang said.
AGI is often referred to as an artificial intelligence that has the ability to understand, learn and apply knowledge across a wide range of tasks, and thus functions very similarly to the human brain. But often the same term is used to refer to theoretical AI that is capable of perceiving itself, or AI that can do anything. There is simply too much that can be hidden under this term - and it is no wonder that Wikipedia, for example, has a very general definition, stating that an AGI is "a hypothetical type of intelligent agent that could learn to perform any intellectual task that humans or animals can perform".
The OpenAI charter, on the other hand, describes AGI as a set of "highly autonomous systems that outperform humans in most economically valuable activities". And AI expert and Geometric Intelligence founder Gary Marcus defined it in his own way as "any intelligence that is flexible and general, with ingenuity and reliability comparable to (or exceeding) human intelligence".
According to Huang, the answer to the question of what AGI is largely depends on how one defines the goal. If it is to be the ability to pass tests that humans must pass in various fields, then general AI will emerge soon, he says.
Currently, AI can handle tests such as bar exams, but it still struggles with specialized medical tests in fields such as gastroenterology. However, Huang said that within five years, AI should be able to pass any of these as well.
Artificial intelligence similar to the human mind
By other definitions, however, achieving AGI may take much longer, according to Huang. In fact, scientists disagree on how to describe the workings of the human mind. "So from an engineer's perspective, it's hard to reach the goal," Huang noted. Indeed, software engineers need clearly defined goals.
Huang also addressed the question of how many more chip factories are needed to support the expansion of the AI industry. OpenAI startup chief Sam Altman, for example, thinks many more are needed than are now. Huang takes a similar view, but he also noted that the chips themselves will improve over time and the number of chips needed will decrease.
"We will need more plants. But don't forget that we are also improving (artificial intelligence) algorithms and processing tremendously over time," Huang said. Computing power will improve a million times in 10 years, he said.
Source: ct24.ceskatelevize.cz
