<aside> ☕ coffee with chAI

</aside>

The hype around AI makes it easy for laymen to imagine world-conquering robots, but that is still a distant goal. While certain tests aim at understanding consciousness in AI, I believe that we don’t understand it enough to devise robust tests at all. Douglas Hofstadter provided the most interesting hypothesis for consciousness: the idea of strange loops. It revolutionized maths, music, and art, and paved a new paradigm for viewing ourselves as conscious beings. Hofstadter talks about self-referencing and a hierarchical structure that moves upwards, always returning to the origin. In agreement, I believe humans can never be completely intelligent, and in that way, we would always climb the hierarchical ladder open-mindedly. At the end of the day, we would reach back to who we are. Taking off from this, I definitely believe artificial consciousness is far from where we are because such a self-referencing ability is not programmed. Until we grasp what our minds are made of, progress will remain limited in this area.

The field of artificial intelligence (intelligent behavior), started from a little experiment called the Turing Test, where a human attempted to guess whether they were talking to a robot. Interestingly, Turing noted that it is easy to appear intelligent even though one might actually not be. The language model won’t know what it’s saying, but the human's ability to make sense of what the AI says is what attributes intelligence to it. In some sense, we are projecting a certain type of intelligence and intentionality on them based on exhibited behavior, quite like the Braitenberg vehicles. However, not only are our current methods for testing intelligence too simplistic, but they are also heavily anthropomorphic. Testing intelligence would require a great deal of understanding, not only of humans but other animals as part of an ecosystem. With the ubiquity of ChatGPT, it’s important to discuss language models in the context of intelligence. Noam Chomsky believed that language can give insights into the human mind’s structure, suggesting that language shapes our understanding of the world and that we have an innate ability to communicate. It only follows that one might question the relationship between language and cognition in AI. Language models, however, do not have the capacity of intelligence as discussed above, and whether language hones cognition or cognition gives us language, this theory doesn’t hold true for artificial models. In fact, Chomsky believes that the current statistical learning methods of AI should shift focus toward the cognitive structures of language instead. Therefore, reaching human-like intelligence goes beyond just understanding and communicating in language; it requires advancements in cognition which AI has not yet achieved. In summary, a complete understanding of AI demands collaboration between fields like neuroscience, linguistics, psychology, mathematics, and computer science. The study of AI involves using AI itself to gain insights into human cognition and intelligence: a paradoxical approach to drive technological and scientific innovation.

References: https://www.theatlantic.com/education/archive/2015/02/what-animals-teach-us-about-measuringintelligence/386330/https://www.nytimes.com/2023/03/08/opinion/noam-chomsky-chatgpt-ai.htmlhttps://stories.clare.cam.ac.uk/will-ai-ever-be-conscious/index.htmlhttps://www.youtube.com/watch?v=4MGCQOAxgv4https://www.technologyreview.com/2021/08/25/1032111/conscious-ai-can-machines-think/https://www.technologyreview.com/2020/10/15/1010461/artificial-general-intelligence-robots-aiagi-deepmind-google-openai/https://www.theguardian.com/culture/2019/mar/09/bach-escher-godel-douglas-hofstadter-consciousness-ai-revolution-mathematical-idea-art-music#:~:text=The%20strange%2Dloop%20concept%20may,itself%20back%20at%20the%20beginning. https://medium.com/the-sophist/douglas-hofstadter-strange-loops-and-the-enigma-of-intelligence-f8f1dc9d377c https://gizmodo.com/chatgpt-openai-ai-contractors-15-dollars-per-hour-1850415474https://ubiquity.acm.org/article.cfm?id=3459743http://justinhart.net/files/publications/SHORT-HRI-10.pdf