From: mk_thisisit
This article summarizes a debate between Professor Andrzej Dragan, a Polish physicist, and Jacek Dukaj, a Polish writer and futurologist, held on the 46th floor of the Skyfall Wars building in Warsaw. The debate, titled “The End of Man,” was organized by the University in USB Merito Warsaw and focused on the nature of humanity, its relationship with artificial intelligence, and the definition of life itself [01:07:09].
Defining Fundamental Concepts
The debate explored the elusive definitions of life, intelligence, and consciousness.
What is Life?
Jacek Dukaj offers two perspectives on life:
- A biological process defined by energy cycles [02:01:42].
- The subjective feeling experienced by every human or living being, contingent on their level of consciousness [02:04:13].
- This aligns with an ancient division between “bios” (biological processes) and awareness within the mind, which is a narrative of what happens in matter [02:15:36]. Humans have a concept of life and discuss it, while bacteria, despite living, do not [02:38:15].
Andrzej Dragan, drawing from physics, states that definitions are often irrelevant because there will always be creatures on the border of any given definition, making classification difficult [02:53:50]. He views philosophical discussions on this topic as “completely futile” [03:38:29]. He explains that while physics can describe elementary particles like electrons, describing a complex organism with similar precision is “basically impossible” [05:02:40].
What is Intelligence?
Jacek Dukaj defines intelligence as operations on symbols [02:54:33]. He argues that the history of homo sapiens’s technological advancements began when humans first used and exchanged symbols [02:44:26]. He highlights a point where knowledge was acquired through operations on symbols without any single human mind comprehending the entire complex process [02:59:04]. The recent introduction of models like Chat GPT-3 demonstrated to “most people” that humanity is living in the era of artificial intelligence [02:47:04].
Andrzej Dragan disputes Dukaj’s definition, using the “Chinese room” thought experiment. In this scenario, a person in a room translates Chinese to English using a rulebook, without understanding either language. Dragan argues that while this is an operation on symbols, “no one will call something like that intelligence” because no one in the room possesses intelligence; it’s a “mindless application of the algorithm” [02:56:56].
Dragan offers his own definition: intelligence is the “ability to see analogies” [02:29:29], which can be precisely measured by the ability to compress data [02:31:30]. He references the Hutter Prize, a half-million-euro award for the best data compression algorithm, explicitly linked to artificial intelligence research [02:33:37]. He asserts that intelligence and consciousness are independent, as a data-compressing algorithm can be intelligent without being conscious [02:56:06].
Dukaj views Dragan’s definition as a “special case” of his broader definition, as any process of carrying out analogies requires the use of symbols [03:19:15]. Dragan counters that Dukaj’s definition is “too wide” because it would imply even “idiots” or the Chinese room are intelligent, which he considers “nonsense” [01:19:18].
What is Consciousness?
Andrzej Dragan believes that talking about artificial consciousness is not necessary for discussing artificial intelligence [00:27:54]. He notes that one can have intelligence without awareness and vice versa [03:11:00]. He admits that current scientific understanding of consciousness and emotions is limited; no one knows how anesthetics work to induce unconsciousness or what neuronal structures are responsible for such experiences [03:18:14]. Dragan predicts that once these mechanisms are understood, experiments will involve “creating networks and inflicting pain on them and seeing what are the reactions” [03:54:19].
Dukaj argues that current “primitive AI” can convincingly claim to feel mental pain, and it is becoming increasingly indistinguishable whether one is talking to a machine or a human [03:58:31]. He suggests that distinguishing true feeling from imitation will only be possible through “autopsy of the structure of this network” [01:00:20].
The Information Society and its Value
Andrzej Dragan suggests that the idea of living in an “information society” is akin to saying “let’s live in the garbage” because information has become “worthless” [00:51:30]. He contrasts this with ancient times, where transmitting a single bit of information could “cost a life” (e.g., the marathon runner) [02:07:33]. Today, phones hold orders of magnitude more information, and its value has plummeted due to abundance [02:22:18]. He likens this to gold, which retains value because it remains rare [02:27:08].
Jacek Dukaj questions this, suggesting that information’s value has increased because technology allows for much greater utilization of it compared to ancient times [02:58:45].
Artificial Intelligence and its Future
Technological Singularity and Human Control
The concept of “technological singularity” is discussed as the moment when artificial intelligence can exponentially improve itself, rapidly “flying away” from human comprehension and control [03:31:07]. Jacek Dukaj views the predictions for this event (e.g., within 2-3 years) as “too optimistic,” though he wouldn’t bet against it [03:55:00].
Dukaj describes a future where humans live in a reality “completely subordinate to non-human beings,” no longer the “highest form of intelligence” [03:59:25]. While some silicon valley potentates view this as the “expected arrival of God,” others are “terrified” and try to delay it [04:09:50]. Dukaj believes that no methods can stop this process, and humanity must “mentally prepare ourselves for this moment” [04:26:06].
Andrzej Dragan agrees that just as humans dominated animals, AI will dominate humans [04:46:17]. He argues that human control over a superior AI would not be an “equilibrium situation” and could quickly reverse. He warns that when humans start blindly following AI’s advice because it seems to offer better outcomes, the situation is “obviously reversed,” and humans become the tool of the more intelligent entity [04:31:09].
AI in the Workplace and Scientific Discovery
Professor Aleksandra Przegaliński is quoted saying that artificial intelligence “changes the nature of work and does not replace it” [04:31:26]. Dukaj partially agrees, stating that AI changes work but also “replaces” in some areas, which will become wider with each passing year [04:44:48]. He considers current discussions about AI’s immediate impact “short-sighted” and “naive,” as what is happening now is disconnected from what will happen in three years [04:54:19].
The debaters discuss the prospect of an “automatic scientist” – an AI capable of creating scientific statements [04:56:56]. Dukaj confirms this has “already happened,” citing an instance where an AI (Funsearch) found solutions to scientific problems that humans could not, thereby proving false the thesis that “only people can practice real science” [04:47:43]. He notes that “another barrier has fallen” [04:08:08]. “Soft sciences,” requiring human interaction, are expected to be the last to be automated [04:46:27].
Limitations of Current AI
Despite advances, current AI models like Chat GPT-4 have notable shortcomings:
- They are “terrible” at adapting their actions to an opponent [03:19:59].
- They struggle to formulate strategies [03:24:43].
- They cannot cope with “abstract thinking” and have trouble creating plans and evaluating opponents [03:26:03].
- While capable of solving complex physical puzzles (e.g., tracking a stone’s location through a series of movements), they lack the ability to form strategies based on long-term interaction [03:41:20].
Dragan distinguishes machine learning from mere data repetition, stating it’s for “obtaining answers that go beyond the scope of training data” [05:08:14]. He argues that to skillfully predict the probability of the next word in a text, an AI must “understand this text” [05:22:20].
Economic Impact of AI
Jacek Dukaj presents a future scenario where AI agents are given cryptocurrency wallets and a goal to “earn money” by performing transactions and hiring other agents or people for tasks [05:42:58]. These AI agents “explore the phase space of all the needs of humanity,” optimizing for what people will pay for most [05:31:07]. This could lead to a “complete Game Changer” in the economy and culture, as AI might make “inhuman strange moves” that human entrepreneurs would never consider [05:54:14].
Human Evolution in the Age of Technology
Jacek Dukaj asserts that “evolution in the traditional sense,” as animals evolve, has “de facto ended” for homo sapiens [00:58:34] and [01:59:17]. This is because cultural pressure and technological advancements directly impacting human biology are now more influential than natural evolution (random gene exchange and extinction of less fit carriers) [01:02:00]. The “evolution of technology” is “much stronger, faster and has a greater impact” [01:02:29].
Andrzej Dragan believes that before humans achieve technology to significantly modify their organisms, they will have already transitioned beyond the “meat stage” [01:03:17]. He suggests that humanity will be “replaced by some creatures that are already devoid of this meat” [01:03:20]. The only way back to traditional evolution would be a catastrophe that sends humanity back to the Stone Age [01:03:46].
The Nature of Scientific Understanding
Unanswerable Questions and the Limits of Knowledge
Andrzej Dragan emphasizes that physicists often use concepts they don’t fully understand, like “electron,” “time,” or “space” [03:35:46]. He states, “I don’t know what an electron is at all, but I know how to describe it” [00:40:48]. He argues that many questions science previously asked are now considered “pointless” (e.g., whether light is particles or waves) because reality is more complex than binary choices [02:11:05]. He predicts that current questions will likely be seen as pointless in 100 years [02:17:59].
Laws of Physics and Predictability
Dragan notes that while physics has found simple laws for 400 years, this might be an approximation of a “much deeper” and more complicated reality [01:11:00]. He cites Holger Nilson’s hypothesis: simple laws like Hooke’s law emerge from extremely complicated quantum electrodynamics interactions of vast numbers of atoms [01:13:00].
Crucially, Dragan highlights situations where things happen “completely illegally” in physics, meaning there is no underlying law determining the outcome [01:01:04]. He uses the example of a photon hitting a piece of glass: it’s impossible to predict if it will reflect or pass through [01:17:09]. This randomness is a “fundamental law of physics,” not mere ignorance [01:47:00]. He provides this example to counter the assumption that everything must have a cause [01:05:00].
Challenges and Future of Technology
Data Storage and Processing
The debate touches upon the challenge of coping with immense data resources. Dragan mentions using electron spins for information storage as a new technology development [01:41:20]. He also points out that current electronic systems are mostly two-dimensional, meaning there’s “still a huge space to use” by moving to three-dimensional processor structures, which would significantly increase processing power in a finite volume [01:43:08].
Andrzej Dragan notes that large-scale machine learning algorithms, like Chat GPT 3.5 with its 175 billion parameters, succeed by efficiently searching a very high-dimensional space to find optimal configurations [04:40:53]. He highlights that research is showing smaller networks can achieve similar competence levels through “clever optimization procedures,” implying that powerful AI might soon run on mobile phones [01:06:00].
Energy Consumption
A major barrier for future technological development is energy consumption [01:10:00]. The increasing energy demands of computing centers following the path of “huge amount of data” will necessitate moving servers to colder climates for cooling and securing vast energy sources, with Microsoft even building its own nuclear reactors [01:17:00].
Ethical Considerations and Future Outlook
When asked if he would stop the development of artificial intelligence, Andrzej Dragan states, “no, but I am irresponsible.” He clarifies that his interest in physics is to “satisfy my curiosity,” not to “save the world” [04:41:35].
The discussion concludes with the主持人 suggesting two conclusions: that the debate’s title “The End of Man?” should be changed to “The End of Man.” (a statement), and that the debaters disagree on the definition of artificial intelligence [01:57:40].