January 30, 2023

Columbia University: AI will be able to think in the next two years

Published: January 30, 2023 at 11:00 am Updated: January 30, 2023 at 10:26 am

In Brief

Neural representations of naturalistic occurrences are updated when our understanding of the past develops, according to a recent study from three US universities.

Understanding is impossible without a retrospective update of previously encoded information in the brain in light of newly revealed circumstances.

The neural representations of previously stored scenes in memory were recoded to reflect changes in understanding of these scenes in light of new information.

Neural representations of naturalistic occurrences are updated when our understanding of the past develops, according to a recent study from three US universities. This work is notable for its objective and the creativity of its approach. The aim is to investigate how the brain modifies our understanding of previous events in response to new information by examining variations in neural representations.

Columbia University: AI will be able to think in the next two years

The researchers used the renowned supernatural thriller “The Sixth Sense” in this study. The movie’s plot revolves around a boy who claims to see ghosts and a doctor who tries to help him. Three groups of subjects got different extra information about the two characters and were asked to record their interpretations of the movie.

The following were the researchers’ hypotheses:

Throughout the movie, details about the story, characters, and events are stored in the viewer’s episodic memory, depending on their personal interpretation.
The revelation that the doctor himself was a ghost at the very end of the movie should alter the viewer’s preconceived notions about what happened earlier.
By recoding it to include the new information, this adjustment should “refresh the viewer’s memory.”

Encoding and recall of story interpretation brain regions.

American researchers have taken a similar approach. Neural representations (patterns of neuronal activity) of the “Default Brain Network” were examined. This network is activated, among other things, during the processing (coding) of continuous, rich, dynamic stimuli such as movies and audio stories. Long “receptive time windows” exist in the regions of this network, during which high-level information accumulated over long periods of time (e.g., movie scenes, paragraphs of text) is integrated and stored to support the comprehension operation.

Understanding is a universal cognitive operation associated with the assimilation of new information, its incorporation into the system of established beliefs, and the updates of this system when new information is learned. In other words, understanding is impossible without a retrospective update of previously encoded information in the brain in light of newly revealed circumstances.

The study experimentally confirmed the authors’ hypothesis. The neural representations of previously stored scenes in memory were recoded to reflect changes in the understanding of these scenes in light of new information.

We can draw an important conclusion from this for the development of “understanding AI.” The lack of understanding of these systems greatly diminishes the sensationalism of AI systems, such as ChatGPT’s numerous intellectual achievements.

Regardless of how difficult the most advanced modern AI systems pass or how complex intellectual functions they perform, they are still devoid of even a speck of understanding and work solely by optimizing the statistically most likely choice of the next words.

Relationship between our behavioral memory updating measure (twist score) and neural memory updating measures across scenes (pISC difference).

Such systems’ inability to comprehend new information that modifies the interpretation of information already encoded in memory, formed during the period of AI training, is principally caused by the fact that they do not recode their memories.
AI does not yet comprehend how the brain actively modifies our perception of past events in the light of new knowledge.

Thus, it can be said that such a “knowledge of AI” will materialize within the next two years.

Read more about AI:

Tags:

Disclaimer

Any data, text, or other content on this page is provided as general market information and not as investment advice. Past performance is not necessarily an indicator of future results.

The Trust Project is a worldwide group of news organizations working to establish transparency standards.

Damir is the team leader, product manager, and editor at Metaverse Post, covering topics such as AI/ML, AGI, LLMs, Metaverse, and Web3-related fields. His articles attract a massive audience of over a million users every month. He appears to be an expert with 10 years of experience in SEO and digital marketing. Damir has been mentioned in Mashable, Wired, Cointelegraph, The New Yorker, Inside.com, Entrepreneur, BeInCrypto, and other publications. He travels between the UAE, Turkey, Russia, and the CIS as a digital nomad. Damir earned a bachelor's degree in physics, which he believes has given him the critical thinking skills needed to be successful in the ever-changing landscape of the internet.

Damir Yalalov