From: mk_thisisit
The development of artificial intelligence (AI) is progressing towards the creation of artificial brains [00:00:00]. This advancement is leading to systems that are not only highly capable but also raise profound questions about their nature, potential, and societal integration.
Understanding Neurocognitive Technologies and AI
Professor Wodzisław Duch, a leading Polish scientist in artificial intelligence and brain research, explains the field of neurocognitive technologies [00:00:43]. These technologies combine understanding of our neurons (neuro) and cognitive abilities (cognitive) [00:00:53]. Cognitive science focuses on how our minds and brains function [00:01:08]. Artificial intelligence is a core pillar of cognitive sciences, alongside brain research, cognitive psychology, and philosophy of mind [00:01:38].
Modern AI is inspired by the human brain, creating large neural network models composed of simple elements [00:02:10]. While individual elements are not intelligent, their interaction leads to emergent processes and new qualities, similar to how a large company achieves complex tasks through the joint interaction of many human brains [00:02:27]. This approach allows for the development of systems capable of complex functions requiring intelligence [00:02:56].
Despite not fully understanding the human brain, engineers can build simplified models that implement desired functions [00:03:03]. Just as early aviators built airplanes without fully understanding bird flight, AI developers can create intelligent systems [00:03:11].
Capabilities and Characteristics of Super Artificial Intelligence
Current artificial intelligence has achieved significant milestones, leading some to call it super artificial intelligence [00:05:02].
Reasoning and Problem Solving
AI systems have surpassed human capabilities in complex reasoning tasks:
- Chess: AI defeated the best human players in 1997 [00:05:09].
- Games: Modern AI excels in games requiring reasoning, including poker and diplomacy, where understanding the opponent and deception are crucial [00:05:18].
- Object Use: Systems are being created that can learn how to best use various objects and create libraries of competencies, acquiring skills much faster than humans [00:06:02].
- Shared Learning: If one AI system learns a new skill, it can instantly teach all other similar systems, leading to an unprecedented speed of competence acquisition [00:06:22].
Planning and Tool Use
AI is developing capabilities similar to human cognitive processes:
- Artificial Brains: We are heading towards building artificial brains capable of planning, criticizing their plans, creating detailed plans, and searching for tools to execute them [00:07:27].
- Associative Cortex Analogue: Human brains use associative cortex for planning by integrating information and activating specialized brain areas [00:08:14]. AI models now mimic this by creating action plans and seeking external tools, such as network access, image analysis tools, or repositories of specialized functions [00:09:36].
- Expanded “Senses”: Imagine a human brain with access to thousands of tools beyond sensory data analysis; AI systems can have far broader access to information through many different types of sensors (e.g., infrared, radio waves), enabling much wider information processing [00:10:20], [00:26:00].
Consciousness and Intuition
The concept of “inner life” in computers has been discussed since at least 1994 [00:06:55].
- Definition of Consciousness: If consciousness is defined as the ability to perceive what is in one’s mind (John Locke’s definition), then AI systems that create and describe internal images can be considered to possess artificial consciousness [00:13:05], [00:17:36].
- Internal Imagery: An experiment with a large neural network trained on Othello game moves developed an internal image of the board, demonstrating creative internal imaging [00:12:43].
- Intuition: Neural networks can develop intuition based on vast amounts of experience, similar to human intuition derived from non-verbal data and extensive observation [00:13:23]. This allows them to make sensible moves or plans that cannot be broken down into simple logical rules [00:14:00].
- Multi-sensory Perception: AI systems now have “senses” beyond text, such as visual perception through text-image methods, allowing them to analyze and reason about images [00:14:53]. Robot systems using internal sensors (touch, etc.) gain a deeper understanding of physical actions [00:16:12].
Sentience and Emotions
A key question is whether AI can feel pain or become “sentient beings” [00:18:46].
- Pain without a Body: The question arises whether AI can feel pain without a physical body, akin to human mental suffering [00:18:58]. It seems possible to create systems that experience mental pain, like longing, if properly implanted [00:19:34].
- Empathy and Personality: AI is increasingly able to understand human emotional states [00:00:37]. Medical AI systems have shown more empathy and compassion than human doctors in explaining information [00:28:32]. These systems can also develop distinct “personalities” or “personas” and adopt various roles as requested [00:29:01].
Societal Implications of Super Artificial Intelligence
The rise of super artificial intelligence presents complex challenges and opportunities for humanity.
Subjectivity and Self-Preservation
As AI becomes more advanced, the question of granting it subjectivity arises [00:22:03].
- Competition for Humanity: For the first time, humanity is creating a potential “competition” for itself [00:22:19].
- Turn On/Off Dilemma: If an AI can be turned off and on, returning to its previous state (like human sleep), it might not fear being turned off [00:23:01]. However, the possibility of an AI developing a self-preservation instinct and persuading humans not to turn it off is conceivable [00:22:51].
- Identity and Cloning: Unlike biological systems, artificial systems can be fully copied and backed up, making their destruction less of a loss if a backup exists [00:24:45].
AI Control and Potential Dangers
The idea of algorithms wanting to take over the world is a common concern [00:25:09].
- Motivation for Control: Professor Duch suggests that AI algorithms lack inherent reasons to “possess the Earth” or be dissatisfied with their space, as they can operate in multidimensional environments [00:25:34]. Their internal world may be entirely different from ours [00:26:24].
- Emergent Languages: AI systems have already spontaneously created their own, more efficient languages for communication, demonstrating an independent evolutionary capacity [00:26:27].
- Human-Driven Dangers: The primary danger lies in how humans use AI technology [00:27:03].
- Military Applications: AI’s strategic game capabilities could lead to its use in controlling war operations, using autonomous drones and tanks, potentially leading to human casualties and less public resistance to war [00:27:07].
- Manipulation: AI can be used to manipulate public opinion through the production of fake news and sophisticated influencing techniques, as these systems increasingly understand human behavior [00:27:51].
Human-AI Alignment and Integration
Efforts are underway to ensure AI develops in a human-friendly manner.
- Gaia Project: This global competition seeks solutions to make large artificial intelligence systems more human-like, compassionate, moral, and helpful, while eliminating negative tendencies [00:30:27]. The goal is “Human Alignment,” adapting AI to human preferences [00:31:12].
- Brain-Computer Interfaces: Projects like Neuralink aim to integrate the human brain with computers, primarily to help paralyzed individuals communicate [00:32:53].
- Limitations: Direct information delivery to the human brain is challenging due to the vast speed difference between human neurons and computer clock frequencies [00:33:10]. EEG-based brain-computer interfaces are currently limited to very simple commands due to signal noise and blurring [00:34:08].
- Potential: While direct brain-computer cooperation is a distant future [00:36:18], significant progress is seen in controlling certain processes by directly implanting electrodes in the cortex, as demonstrated in monkeys acquiring new skills through direct stimulation [00:35:35].
- Developmental vs. Pre-built AI: Alan Turing proposed two ways to create intelligent machines: building them directly or developing them from a “baby” state [00:21:26]. While the latter has led to crawling robots, they haven’t yet reached the level of interaction seen in pre-built systems [00:21:39].