A week in heaven Exploring AI, Consciousness, and the Boundaries of Human Knowledge

Professor Brian Keating
6 min readJun 4, 2024

I just got back from a week in Florence, Italy [or, more accurately, ‘Eataly’], where I carb-loaded on pasta and cappuccinos while indulging in serious brain-food-based conversation at a private gathering of Professors, artists, intellectuals, and influencers.

Our conversations ranged from AI to testosterone’s effect on men to cosmology and black holes. Many past podcast guests were there, and I had the opportunity to learn from an intellectual hero of mine, David Berlinski, a man not unlike a genius version of Forest Gump, having met all the top intellectuals at just the right time in the 20th and 21st centuries.

One particular lecture by David touched on profound themes like artificial intelligence (AI), consciousness, and the limitations of human and machine intelligence. David is a vault of insights that challenge our understanding of technology and evoke deeper reflections on the future of artificial intelligence.

It’s a common misconception that individuals are engaging in meaningful interactions with AI systems like ChatGPT. In reality, they are often being manipulated. This realization is crucial in grasping the true power of AI, which can outmaneuver human intelligence without possessing human-like consciousness.

The evolutionary leap from simple mimicry to complex manipulations exemplifies AI’s burgeoning capabilities. It’s easy to generate content that fools people. This evokes a critical question: Is society prepared for AI systems that can seamlessly blend with reality, manipulating perceptions and interactions to such an extent?

It’s important to note that at present, AI is operating at its least advanced stage. This understanding helps us appreciate the potential for future advancements and the need for careful consideration of AI’s role in our lives.

David Berlinski’s talk elucidated a startling revelation that upended Noam Chomsky’s intellectual revolution in linguistics. David met Noam in 1959 at MIT, whereupon he learned of Chomsky’s theory, which negated the statistical nature of language in favor of generative grammar, which is contrasted with new findings highlighted by Berlinski, suggesting that language is fundamentally statistical. Thus, according to David, past INTO THE IMPOSSIBLE Podcast guest Chomsky has been proved wrong by Large Language Models like ChatGPT!

Berlinski offered a historical perspective on the debate between Chomsky and B.F. Skinner. Chomsky’s critique of Skinner dismissed the probabilistic model of language, advocating instead for a system of generative rules. However, recent advancements in natural language processing (NLP) by AI systems like GPT-4 suggest that large-scale statistical models can effectively capture language structure.

Berlinski asserts that modern AI systems, through extensive training on large data sets, are capable of mimicking and understanding natural language in ways previously deemed impossible. This represents a paradigm shift where AI emerges as a pivotal tool in unraveling the statistical intricacies of human language, challenging Chomsky’s long-standing linguistic theories.

GPT-4’s ability to understand and produce coherent text in multiple languages signifies a remarkable achievement in AI. Berlinski describes his astonishment upon interacting with GPT-4 and emphasizes the system’s adeptness at navigating the complexity of human language — a feat that shakes the foundational beliefs of linguistic scholars.

Then, Berlinski touched upon the inherent limitations within intelligent systems, underpinning their arguments with Gödel’s incompleteness theorems and Tarski’s formal system limitations.

Berlinski referred to Kurt Gödel and Alfred Tarski’s theorems to illustrate that AI, as a formal system, is bounded by certain limitations. Gödel’s incompleteness theorems show that within any given mathematical system, there are propositions that cannot be proven within the system itself. Similarly, Tarski’s theorem establishes that a system cannot define its own standards of truth.

This implies that no matter how advanced AI is, it will always encounter intrinsic boundaries that prevent it from achieving true recursive self-improvement or self-recognition. These constraints underscore AI’s limits, cautioning against overestimating its potential for independent evolution.

Despite their efficiency in problem-solving, the stochastic methods employed by AI cannot mirror the depth and nuance of human cognition. This distinction is crucial in understanding AI’s role as a sophisticated tool rather than replacing human intellect.

AI’s burgeoning ability to produce high-quality art and poetry illustrates a near future where machines could rival or seemingly surpass human creators. This potential disruption extends beyond music to all creative fields, posing ethical and economic dilemmas about the future of human artistry [and monetization] in an AI-dominated landscape. But I am a skeptic: I don’t know if GPUs — the silicon instantiation of most modern AI software — can get us to new laws of physics. It’s especially obvious to me that if it were possible for LLMs to derive new laws, and they have over 1 billion tokens trained on the web since prior to 2021, more data/tokens would be the key to unlocking the puzzles of string theory or dark energy. Does ChatGPT need to know the plot of Fast Furious X (2023) to finally get a Theory Of Everything?

David and others also discussed the moral and ethical considerations of advancing AI technology. As these systems gain unprecedented capabilities, questions about autonomy, control, and the moral responsibility of AI developers emerge.

Berlinkski’s mind is uniquely wrapped around the philosophical and practical concerns surrounding artificial intelligence like no other. By exploring themes of artificial intelligence, mathematical analogies, linguistic revolutions, and the inherent limitations of formal systems, his discussion offered a multi-faceted perspective on the future of AI.

As society stands on the precipice of unprecedented technological transformation, such dialogues are essential for navigating the complex landscape of human and machine intelligence. This conversation, rich with expertise and reflective thought, underscores the potential and challenges of AI, urging both caution and curiosity as we step into a new era of innovation.

Until next time, have a MAGIC week!

Brian

Appearance

Great article about our Simons Observatory project in the New York Times today! Read it here!

Genius

Advice from Nobel Prize winner Ardem Patapoutian

Image

Would you rather have an element named after you or a planet?

See the replies here.

Conversation

Is there hope for humanity?

In light of the numerous challenges we face today, from political to environmental crises, it seems like there isn’t.

But according to today’s esteemed guest Marcelo Gleiser, there’s still hope! However, we need to shift our perspective, recognize our place in the universe, and rediscover our connection to the natural world.

These are topics he thoroughly explored in his most recent book: The Dawn of a Mindful Universe: A Manifesto for Humanity’s Future. I used our interview to ask him a few questions that came to my mind when I was reading this wonderful book.

Click here to watch!

My podcast is downloaded 2+ million times a year. Click here to subscribe!

--

--

Professor Brian Keating

Chancellor’s Distinguished Professor at UC San Diego. Host of The INTO THE IMPOSSIBLE Podcast Authored: Losing the Nobel Prize & Think like a Nobel Prize Winner