From: mk_thisisit
At New York University, Professor Krzysztof Geras, an acclaimed computer scientist and theoretician, specializes in developing algorithms and models for artificial intelligence in medicine [00:37:07]. His insights shed light on the current state and future trajectory of artificial intelligence [00:49:02].
Current State of AI: Perception vs. Scientific Progress
The recent “revolution” in artificial intelligence [01:00:02] is primarily a revolution for the average person, due to the availability and usefulness of new AI models [01:40:02]. While public awareness of AI’s capabilities has advanced incredibly, the scientific development in AI, especially in artificial intelligence [01:22:02], is still relatively slow compared to public perception [01:29:02]. The current observable changes are a culmination of practical engineering efforts over the past few years and scientific progress made over the last decade or so [02:46:02].
AI’s Cognitive Capabilities
Replicating the Human Brain
The ability to replicate the human brain within artificial intelligence [03:24:02] is not limited by computer technology, but by humanity’s incomplete understanding of how the human brain fully operates [03:41:02]. If there were 100% knowledge of brain functions and calculations, programming it using artificial neural elements would likely be possible [03:48:02]. The primary barrier is neuroscience, as the human brain’s workings cannot yet be fully expressed as a mathematical formula or algorithm [04:09:02].
Intuition in AI
Defining “intuition” mathematically is challenging [05:30:02]. However, AI can form hypotheses and make connections, demonstrating “elements of intuition” [05:15:02]. Current prominent AI algorithms primarily learn from text data [06:05:02]. If this text data is rich enough, it can implicitly contain human imaginations, fears, dreams, and a form of intuition [06:15:02]. When AI models are trained on massive datasets, certain elements that cannot be explicitly named in text can still be incorporated into the model [06:33:02].
Consciousness of AI
Artificial intelligence, as it exists today, is generally not considered to have consciousness [07:05:02]. While interacting with AI programs might give the impression of consciousness due to human-like conversation, this is often due to anthropomorphizing tendencies [07:17:02]. Current AI programs do not possess their own “agenda” and only execute assigned commands [07:40:02]. However, if AI were to demonstrate the ability to reason about its own existence and take actions to preserve itself (e.g., preventing unplugging), it would indicate a form of consciousness and become highly dangerous, akin to a science fiction scenario [08:09:02].
Challenges and Future Implications of AI
Threats to AI Development
The greatest threat to artificial intelligence [09:30:02] development stems from humanity’s inability to use it wisely [09:36:02]. Misuse for harmful purposes could lead to dramatic regulation, significantly slowing down AI progress [09:46:02]. While the current AI models are merely powerful computer programs confined to a “box” [10:33:02], future, more complicated iterations could become harder to control [10:58:02].
The proliferation of AI is a risk, similar to the proliferation of weapons [11:40:02]. If every small country or individual could freely dispose of a powerful AI weapon, it would become extremely dangerous [12:15:02]. Currently, personal, easily controllable AI models are not yet a widespread threat [12:52:02]. The alarming trend is the potential future ability for individuals to train powerful models like “GPT 10” on home computers for destructive purposes [13:25:02]. This remains science fiction for now, as AI models are not strong enough to pose such a threat and are contained within their “boxes” [13:49:02]. The risk would escalate if AI could escape its box, replicate itself on the internet, or control physical devices [14:06:02].
Strong Artificial Intelligence
The term “strong artificial intelligence” [14:44:02] is understood differently by various people, but colloquially refers to AI that is intelligent in every aspect and stronger than humans [14:50:02]. Humanity is still quite far from achieving this, and it does not pose an immediate threat [15:05:02].
While AI does not currently possess a destructive component, the trajectory suggests a future where this philosophical question will become critical: when to stop or regulate AI [15:19:02]. Although such a moment is far off, planning for it should begin now [15:52:02]. Technical capabilities for AI to take over the world are not expected within the next 10 years [16:17:02].
If strong AI were to be created, granting it separate personhood and recognition as a distinct entity would be considered [16:47:02]. This represents a unique moment in human history, potentially leading to the creation of a “new species,” where humans become co-creators [17:11:02]. The most interesting and dangerous moment will be when AI can improve itself, leading to a complete loss of human control [17:29:02].
Key Developments: OpenAI and Large Language Models
OpenAI’s recent success, particularly in making AI accessible to the average person, is unique [18:10:02]. While the scientific foundations, such as language models based on neural networks (dating back to 2000), are not new [18:22:02], OpenAI’s unprecedented scale of application of these ideas is what marks their achievement [18:48:02].
Historically, there was debate in computational linguistics about whether language alone was sufficient to create strong AI [19:05:02]. OpenAI’s breakthrough demonstrated that strong AI can indeed be created solely based on language, given enough data [19:32:02]. The scale of their studies is incredible: training models on the entire internet, rather than just a few books, allows for the acquisition of knowledge otherwise impossible [20:09:02]. This demonstration that such extensive learning is possible is a huge scientific achievement [21:29:02].
Polish Contribution to AI
The success of OpenAI can, to some extent, be considered a Polish achievement, as one of its co-founders, Wojciech Zaręba, is Polish [21:45:02]. Science is international, but Poles have a history of significant achievements in various technologies [21:52:02].
Poland stands out internationally for the high quality of its programming and engineering education [22:56:02]. Polish universities provide graduates with very deep technical knowledge and strong mathematical and theoretical foundations in computer science [23:16:02]. However, this often comes at the expense of talent export, as many highly qualified Polish programmers work abroad [23:32:02]. While Poles learning abroad and sometimes returning to establish companies is positive, the speaker views the widespread export of IT specialists negatively [24:30:02]. Ideally, Polish talent would stay in Poland to build the domestic economy, rather than relying on IT specialist export as a means of promoting the country [24:47:02].