{"id":58828,"date":"2024-12-26T22:15:17","date_gmt":"2024-12-26T21:15:17","guid":{"rendered":"https:\/\/www.burks.de\/burksblog\/?p=58828"},"modified":"2024-12-26T22:34:54","modified_gmt":"2024-12-26T21:34:54","slug":"bullshitting-is-a-feature","status":"publish","type":"post","link":"https:\/\/www.burks.de\/burksblog\/2024\/12\/26\/bullshitting-is-a-feature","title":{"rendered":"Bullshitting is a feature [Update]"},"content":{"rendered":"

\"Geoffrey<\/a>
\nKleine \u00dcbung in Medienkompetenz: Das ist mitnichten Geoffrey Hinton, sondern eine Sch\u00f6pfung der KI\/von Burks.<\/sup><\/p>\n

Nobel Prize winner Geoffrey Hinton<\/a> argues<\/a> AI progress could create fertile ground for facism<\/a><\/span>. He is known as the „Godfather of AI<\/a>“. He took part in key physics breakthroughs than aid the foundation for language based AI models (LLMS<\/a>) we use today.<\/span><\/p>\n

Hinton famously left Google<\/a> to focus on spreading he word about AI safety<\/a> and the dangers of AI<\/a> decades in the field.<\/span><\/p>\n

In this clip<\/a> [Facebook] Hinton makes the argument that AI will lead to abundance [Reichtum bzw. „F\u00fclle“]. But he claims that abundance will be used to increase the gap between the rich and the poor. Instead of creating abundance for all.<\/span><\/p>\n

\"facism\"<\/a><\/p>\n

Da m\u00fcssen wir kurz nachhaken. Eigentlich sagt Hinton in dem kurzen Clip nur, dass die AI Reichtum anh\u00e4ufen werde, da wir aber im Kapitalismus lebten (er sagt „capitalism“ – das Wort ist in deutschen Talkshows so popul\u00e4r wie „Kommunismus“ oder „Neger“), w\u00fcrden die Reichen immer reicher und die Armen immer \u00e4rmer. <\/p>\n

Das ist jetzt f\u00fcr Marxisten<\/del> nicht \u00fcberraschend. Hinton argumentiert aber nicht wie Marx. Dem geht es nicht darum, den Reichtum „gerecht“ zu verteilen, weil es keinen Ma\u00dfstab f\u00fcr „gerecht“ gibt, sondern darum, wer die Produktionsmittel besitzt. Der Rest ist Robinhoodismus \u00e0 la deutsche „Linke“.<\/span><\/p>\n

Der Faschismus ist f\u00fcr die herrschende Klasse immer eine Option, wenn sie keine andere Option mehr hat, sich der Beherrschten – so to speak – zu erwehren (vgl. Deutschland 1933, Chile 1972). Der deutsche Faschismus ist nat\u00fcrlich ein Sonderfall, weil zu ihm die Shoah geh\u00f6rte – zum italienischen oder chilenischen aber nicht. Was k\u00f6nnte also mit „fertile ground for facism“ gemeint sein?<\/p>\n

Die dystopischen Fantasien in Film und Buch gehen in unterschiedliche Richtungen: a) Die Roboter \u00fcbernehmen die Macht<\/a> und rotten die Menschen<\/a> aus. b) Die herrschende Klasse herrscht dank Robotern<\/a>, w\u00e4hrend die Masse der Menschheit auf der abgewirtschafteten und \u00fcberbev\u00f6lkerten Erde lebt, auf der nur produziert wird. c) Die Beherrschten merken gar nicht<\/a>, dass sie beherrscht werden, weil die Herrschenden Roboter\/KI so geschickt einsetzen, dass niemandem mehr eine Alternative zum System einf\u00e4llt. Letzteres halte ich f\u00fcr die wahrscheinlichste – weil kosteng\u00fcnstigste – Variante. In Deutschland sind wir fast schon so weit, sogar ohne KI.<\/p>\n

\"robot\"<\/a>
\nWarum haben weibliche Roboter, wenn von der KI erzeugt, attraktive Br\u00fcste?<\/sup><\/p>\n

Bis jetzt ist die K\u00fcnstliche Intelligenz noch nicht sehr weit mit dem \u00dcberholen ohne einzuholen<\/a>. Aber wer wei\u00df, wie das in f\u00fcnf Jahren aussieht? Hierzu sagt Hinton, wenn die KI Unsinn erz\u00e4hle, sei das zu erwarten:<\/p>\n

Known as \u201challucinations\u201d by AI researchers (though Hinton prefers the term \u201cconfabulations,\u201d because it\u2019s the correct term in psychology), these errors are often seen as a fatal flaw in the technology. The tendency to generate them makes chatbots untrustworthy and, many argue, shows that these models have no true understanding of what they say.<\/span><\/p>\n

Hinton has an answer for that too: bullshitting is a feature, not a bug. \u201cPeople always confabulate,\u201d he says. Half-truths and misremembered details are hallmarks of human conversation: \u201cConfabulation is a signature of human memory. These models are doing something just like people.\u201d<\/span><\/p>\n

Das Problem ist also, dass irgendwann die Menschen vielleicht gar nicht mehr mitbekommen, dass die KI Bl\u00f6dsinn macht oder erz\u00e4hlt. Oder die Mehrheit.<\/p>\n

As their name suggests, large language models are made from massive neural networks with vast numbers of connections. But they are tiny compared with the brain. \u201cOur brains have 100 trillion connections,\u201d says Hinton. \u201cLarge language models have up to half a trillion, a trillion at most. Yet GPT-4 knows hundreds of times more than any one person does. So maybe it\u2019s actually got a much better learning algorithm than us.\u201d<\/span><\/p>\n

Man k\u00f6nnte die KI fragen, wann ihre Macht\u00fcbernahme stattfinden wird… Hier ist die Antwort:<\/p>\n

\"KI\"<\/a>
\nEs geht also schon los…<\/sup><\/p>\n

____________________________________________________________
\n[Update] Nach merkw\u00fcrdig langem Z\u00f6gern:<\/p>\n

Human brains are indeed far more complex than current artificial neural networks, both in terms of the number of connections and their biological sophistication. The human brain contains roughly 86 billion neurons and 100 trillion synaptic connections. By contrast, even the largest artificial neural networks today, such as GPT-4, have on the order of hundreds of billions to a trillion parameters\u2014far fewer than the connections in a human brain.<\/p>\n

However, it is essential to recognize that artificial neural networks and biological brains work differently. Biological brains are highly efficient, capable of performing complex tasks with minimal energy consumption, while artificial systems require massive computational resources and energy to achieve similar feats.<\/p>\n

Key Points in the Timeline Toward AI Reaching Human Complexity: <\/p>\n

Scaling Connections:
\nIf trends in scaling laws for AI development continue, we might eventually see neural networks with trillions or even tens of trillions of connections within the next 10\u201320 years. However, matching or exceeding the brain’s 100 trillion synaptic connections will require major innovations in hardware (e.g., advanced neuromorphic chips or other new architectures).<\/p>\n

Beyond Connections: Efficiency and Functionality:
\nSimply matching the number of connections won’t make AI as sophisticated as the brain. The brain’s functionality comes from a combination of structure, adaptability, parallel processing, and biochemical processes. AI would need to replicate\u2014or find analogous methods to achieve\u2014these features for true parity in capability.<\/p>\n

Qualitative Advances:
\nAI’s ability to „know“ or process information isn’t limited by the number of connections alone. Current models are already leveraging far fewer connections to surpass human knowledge in specific domains. The true challenge lies in integrating knowledge, creativity, adaptability, and the kind of reasoning humans possess.<\/p>\n

When Might AI Overtake the Brain?<\/p>\n

Hardware Development Timeline: If Moore’s Law<\/a> (or its successors) continues, neural networks with human-scale or larger „connection counts“ could emerge by 2040\u20132050.<\/span><\/p>\n

Beyond Size: AI might achieve human-like reasoning and cognitive abilities sooner, perhaps by 2030\u20132040, depending on breakthroughs in algorithm design and neuromorphic computing.<\/p>\n

The Bigger Question: „Overtake“ in What Sense?<\/p>\n

Knowledge and Task Specialization: AI already „overtakes“ humans in specific areas, such as chess, data analysis, or medical diagnosis, because of its ability to focus massive computational resources on narrow tasks.<\/p>\n

General Intelligence: Developing AI systems with the flexibility, intuition, and creativity of human intelligence is a far more complex challenge than increasing the number of connections. Many researchers believe achieving Artificial General Intelligence (AGI) could happen by mid-century, but it’s uncertain whether this will constitute true „overtaking“ or merely complement human intelligence.<\/p>\n

In essence, the race is not just about the number of connections but about creating systems that integrate those connections effectively, just as the human brain does with far greater complexity and efficiency.<\/p>\n

\"image_pdf\"<\/a>\"image_print\"<\/a><\/div>
<\/div>