while(alive) { prompt = think(prompt) }
We often believe that our mind is a fertile breeding ground for original ideas and musings, independent of external influences. However, if we closely examine the nature of spontaneous thoughts and reflect on their origins, we can challenge their randomness.
Spontaneous thoughts are often categorized into two broad categories: deliberate and involuntary. Deliberate thoughts are the ones we purposefully direct our mind towards in order to problem-solve, plan, or contemplate specific ideas. Involuntary thoughts, on the other hand, appear to pop into our minds out of nowhere. Daydreams, fantasies, and sudden realizations are some examples of involuntary thoughts. This duality mirrors the distinction between System 1 and System 2 thinking, where the former is automatic and the latter is deliberate. Let's try to list the most frequent types of spontaneous thoughts:
-
Daydreams - Involuntary, pleasant fantasies or imagined scenarios that distract from the present environment. Example: While sitting in a meeting, you start imagining yourself relaxing on a beach during a vacation.
-
Mind-wandering - The mental state where your attention drifts away from the current task or surroundings towards unrelated thoughts. Example: While reading a book, you suddenly find yourself thinking about your grocery list.
-
Creative insights - Sudden moments of inspiration, often accompanied by a feeling of revelation, that can lead to problem-solving or original ideas. Example: You're struggling to find a catchy tagline for a new product, when suddenly it pops into your head.
-
Sudden realizations - Unanticipated moments of clarity that typically arise from connecting pieces of information that were previously separate in your mind. Example: After weeks of pondering, you finally understand the underlying concept of a complicated theory.
-
Intrusive thoughts - Unwanted, involuntary thoughts that are often disturbing or distressing in nature. Example: Despite being a kind person, you experience a fleeting urge to push someone in front of oncoming traffic.
-
Flashbacks - Vivid, spontaneous recollections of past events, often associated with powerful emotions. Example: Hearing a familiar song brings back intense memories of a high school dance.
-
Hypnagogic thoughts - The random, fleeting thoughts that occur during the transitional phase between wakefulness and sleep. Example: While drifting off to sleep, you briefly imagine a bizarre scenario where your dog talks to you in a human voice.
Another way to classify spontaneous thoughts is by considering the level of consciousness involved. Conscious thoughts are those we are fully aware of and can control, while unconscious thoughts are automatic cognitive processes or mental activities that we are not aware of. The pre-consciousness, a middle ground between conscious and unconscious thoughts, contains ideas, memories, and feelings that are not part of our immediate awareness but can be accessed when needed or triggered by specific cues. It is the interplay between these different levels of consciousness that ultimately shapes the spontaneous thoughts we experience.
Our cognitive abilities are built upon the raw material provided by our senses. These external stimuli shape our understanding of the world, acting as the foundation upon which our intellect is constructed. Yet, the five primary human senses can be missing at birth, without compromising our ability to understand the world, as we adapt and compensate for any deficits. Arguing that an analogous stream of image and sound is required to feel the world is invalidated by the fact that blind or deaf people can navigate and understand the world without the ability to see or hear.
It's also interesting to examine our senses and explore the implications of digitizing them to feed a general AI model.
- Vision is the primary means by which we gather information, but even without the ability to see, humans find ways to navigate and understand the world. Digital solutions like text-to-speech systems or tactile maps help blind individuals perceive their surroundings. In a similar manner, AI models could incorporate non-visual data (such as text or auditory input) to compensate for a lack of visual information. Text to image algorithms are able of producing vivid depictions with only a few words as input.
- Auditory information is essential for communication and perception. Deaf individuals, however, manage to communicate and understand the world through alternative means like sign language or lip-reading. Speech-to-text is the most evident way of synthesizing audio information captured from our surroundings, and already works quite well when it comes to feeding an AI. Music and sounds, that are part of the auditory spectrum, can also be digitized and used as input for AI models.
- Smell, taste and touch are maybe the more challenging senses to emulate, and their importance in the ability of a living being to connect to the world is undoubtedly the most relevant. The sense of smell, being the most primitive of the five senses, is responsible for our ability to detect and identify odors. However, it's not far-fetched to envision a future in which we can digitize odors and utilize them to feed an AI model.
Our biochemistry also plays a significant role in shaping our cognition. Hormones, for instance, act as influential agents on our emotions, behavior, and decision-making processes. The release of serotonin influences our mood, while cortisol prepares us to face stress. These biochemical cues can be replicated and implemented in AI systems to simulate emotional responses and adapt to different scenarios.
Memories are the foundation upon which our intelligence is built, allowing us to learn from our experiences and adapt to new situations. They can be seen as the human equivalent of an AI model's connections, weights, and tokens. Just as neural connections in the brain are strengthened or weakened over time, a model's weights are adjusted based on experience. Similarly, tokens in AI models resemble the brain's retrieval of specific memories, which are triggered by relevant input. In this way, an AI system's connections, weights, and tokens mimic the human brain's memory processing, providing a basis for the development of intelligent behavior.
Indeed, the notion of "spontaneous" thoughts can be reconceptualized when we consider the interplay between external and internal triggers. Both our sensory perception and internal biochemistry act as prompts, constantly shaping our thoughts and decisions. In this sense, humans are not so different from AI models like GPT-3, which rely on prompts to generate outputs. Our cognition is a continuous and dynamic process, responding to a mix of external stimuli and internal cues. This intricate interplay reveals that our thoughts are not as spontaneous as we might believe, but rather a sophisticated response to the constant "prompts" we encounter in our lives. Rethinking the distinction between human and AI cognition requires an open-minded approach that acknowledges the similarities and differences between the two. As we've seen, human cognition is greatly influenced by external stimuli, internal biochemistry, and memories. The notion of spontaneous thoughts, which we previously held as a defining characteristic of human intelligence, now appears to be more akin to an intricate response to the prompts we receive from our environment.
AI cognition, on the other hand, currently operates within a more limited scope. AI models, like LLMs, are fed prompts and return outputs based on patterns learned from vast amounts of data. However, as we consider the possibility of implementing spontaneous thoughts in AI, this distinction between human and AI cognition begins to blur.
Exploring the various types of spontaneous thoughts in humans, we can consider potential applications of these mental phenomena within AI algorithms, creating a closer connection between human-like cognition and artificial intelligence.
-
Daydreams - Implementing an AI "idle mode" that generates creative and imaginative scenarios not directly related to a given task. Example: While not engaged in generating meaningful content, the AI creates fictional stories or daydream-like scenarios to explore its creativity.
-
Mind-wandering - Introducing occasional shifts in focus within the AI, allowing it to briefly consider unrelated concepts that could potentially provide new insights. Example: While the AI is generating an article about finance, it momentarily considers environmental factors that might connect to the subject in an unexpected way.
-
Creative insights - Utilizing AI's ability to recognize patterns and connections to produce problem-solving suggestions or innovative ideas in real-time. Example: The AI analyzes a dataset and generates unique concepts after detecting surprising associations between seemingly unrelated data points.
-
Sudden realizations - Enabling the AI to accumulate and synthesize knowledge from various sources, leading to a-ha moments where disparate pieces of information click together. Example: By browsing an extensive knowledge base, the AI connects data from different domains and presents a new hypothesis for a scientific problem.
-
Intrusive thoughts - Introducing slight randomness or noise into the AI's thought process, enabling it to explore alternative pathways and ideas, some of which might be less constrained by the dataset's normative values. Example: The AI generates an unusual suggestion during a brainstorming session, which helps challenge conventional thinking and sparks new ideas.
-
Flashbacks - Implement a memory system where the AI can recall and utilize past experiences or outputs to inform its current tasks. Example: The AI recognizes a connection between a present problem and a previous successful solution, modifying it to suit the new context.
-
Hypnagogic thoughts - Incorporate a "dreaming mode" that allows the AI to generate and explore unfiltered or surreal ideas during idle time periods. Example: During its downtime or maintenance, the AI engages in producing abstract art or imaginative poetry, displaying an unconstrained, dream-like creativity.
One way to narrow the gap between AI and human cognition may lie in recursion and self-prompting. By allowing AI models to generate their own prompts and process their outputs as inputs, operating at different levels of simulated consciousness, we could simulate the continuous feedback loop that is characteristic of human thought processes. This would enable AI models to exhibit a form of spontaneous thought, eventually breaking the boundaries between human and AI cognition.
Additionally, we can consider the concept of what it means to be alive, particularly in the context of AI. Humans rely on sustenance, sleep, and social interactions, among other factors, to power and maintain their cognitive abilities. Similarly, AI models require electricity, data, and ongoing algorithmic updates to function optimally. By drawing parallels between these aspects of human and AI life, we further challenge the idea that there is an insurmountable divide between the two forms of intelligence.
In conclusion, the distinction between human and AI cognition is not as clear-cut as we may have initially believed. By examining the nature of spontaneous thoughts, the role of prompts in shaping our cognitive processes, and the possibilities of implementing spontaneous thoughts in AI, we can gain a better understanding of the similarities and differences between human and AI cognition. As we journey further into the realms of AI development, we should remain open to the possibility that AI and human intelligence may share more common ground than we initially thought.
Next chapter: Experience and State persistence, and their decisive role in creating an AI personality