Researcher: Artificial Intelligence Currently Lacks Perception Capability

2023-08-31

There's a joke about a daughter asking her father why he speaks so softly at home. "Because there are artificial intelligence listening to us everywhere," the father replied. The daughter laughed, the father laughed, and Alexa laughed too.

Artificial intelligence seems to be increasingly permeating every aspect of our lives. AI brains have absorbed trillions of bits of data, equivalent to obtaining a million PhD degrees, and respond in a charming tone and manner that sounds as simple and human as your favorite university professor. This leads some to wonder: are computers becoming emotional?

Cynics would answer, "Of course not. Computers can solve problems that would take generations for humans to solve in a matter of seconds, but they cannot feel love and pain, see and appreciate the moon and stars, or smell the coffee we spill on our keyboards."

However, some argue that we need to define our understanding of "perception" more clearly. Are there varying degrees of consciousness as we understand it? Is there an overlap in subjective experience between humans, animals, and intelligent machines?

A group of 19 neuroscientists from the United States, United Kingdom, Israel, Canada, France, and Australia explored this question in a report published on the preprint server arXiv on August 22.

Last year, the chief scientist of OpenAI, the developer of ChatGPT, speculated that advanced AI networks might have "a mild form of consciousness." A year ago, a Google scientist was fired for claiming that the precursor to the chatbot Bard, called LaMDA, had perceptual capabilities.

However, after an extensive analysis of various theories of consciousness, the authors of the report, titled "Consciousness in Artificial Intelligence: Insights from the Science of Consciousness," concluded that AI systems do not possess consciousness, at least not yet.

Nevertheless, they proposed some methods for future researchers to consider.

"Our analysis indicates that current AI systems do not possess consciousness," said Patrick Butlin, the lead author of the report. "But it also suggests that there are no apparent technical barriers to constructing AI systems that meet these criteria."

They narrowed down the theories of consciousness to six, which they believe are convincing indicators of the existence of conscious entities.

One example is the theory of recurrent processing, which posits that the brain processes information through feedback loops. Through such loops, the brain can adapt to a constantly changing environment, adjust perception, and make wiser decisions. This iterative behavior is crucial for memory formation and acquiring new knowledge.

Another key concept is the higher-order theory, which some summarize as "consciousness of consciousness."

"The difference between the higher-order theory and other theories is that the higher-order theory emphasizes that for a conscious mental state to exist, the subject must be aware that they are in that mental state," Butlin said.

The third factor is the global workspace theory. This theory suggests that consciousness arises when information is widely accessed in the brain. Information is not limited to input from individual senses but is contained within a global center that can be used for various internal cognitive processes.

According to Butlin, the proposed tools for blocking perception "provide us with the best available means to assess whether AI systems could potentially have consciousness."

"One of the reasons we published this report is that we seriously consider the possibility of conscious AI systems being built in a relatively short period of time," Butlin said. "This possibility will raise profound ethical and societal questions."

RECENT AI NEWS

RECENT AI TOOLS