Chomsky dismisses current "AI" as merely industrial scale plagiarism (or equivalent words). Is he right to be so dismissive? If he IS right a great deal of time and effort is being wasted!
Chomsky's critique of AI as "industrial scale plagiarism" warrants a nuanced examination, especially considering the capitalist landscape and geopolitical dynamics surrounding microchip production. Because, the real word is the construction base to all inquiries, and this are there rock base. Technological foundation: Microchips serve as the physical building blocks of AI systems, enabling the complex computations and data processing required for AI algorithms to function. Production bottlenecks: The current global chip shortage has highlighted the fragility of the supply chain and its impact on AI development. Geopolitical tensions: The US-China rivalry in chip manufacturing further emphasizes the strategic importance of microchips in the development and deployment of AI technologies.
Besides this topics, there are others capitalist considerations, to be described. Profit-driven innovation: The private sector, driven by profit motives, plays a significant role in AI research and development. This can lead to a focus on short-term gains and market-driven applications rather than long-term societal benefits. Intellectual property and control: The ownership of AI-related patents and algorithms raises concerns about intellectual property rights, data privacy, and the potential for AI to be used for surveillance and control. Access and equity: The high costs associated with AI research and technology can create barriers to entry for smaller companies and developing nations, exacerbating existing inequalities.
But, in the end of the FREE world who controls whom?
Let´s start with some points. Data dominance: The companies that possess large amounts of data have a significant advantage in AI development, as data is essential for training AI models. This raises concerns about data ownership, privacy, and the potential for AI to be used for discriminatory or manipulative purposes. Algorithmic bias: AI algorithms can reflect and amplify the biases present in the data they are trained on, leading to discriminatory outcomes. It is crucial to address these biases through careful data selection and algorithm design. Ethical considerations: The development and deployment of AI technologies raise ethical concerns such as job displacement, privacy violations, and the potential for AI to be used for malicious purposes. It is essential to establish ethical guidelines and regulations to ensure responsible and beneficial use of AI.
In this way, probably, Chomsky's critique of AI as "industrial scale plagiarism" is a valid starting point for a broader discussion about the economic and geopolitical factors shaping the development and deployment of AI technologies. It is crucial to consider the role of microchip production, the influence of capitalist dynamics, and the question of control in shaping the future of AI. By engaging in these discussions and addressing the associated challenges, possible we can strive for a more responsible and inclusive development of AI that benefits all of society.
Very depends on context, I would say. If you use AI for image generation, it might be true, as it is a partial copy of multiple works.
However, in research field, AI is used to recreate data for a reason, so I would say NO, it is not plagiarism in general. I mean, you're trying to teach a model on your own data to make predictions faster or solve some problems.
I'm almost sure AI already had plenty of AI WINTERS, no need one more :D
I do share Chomsky's criticism of modern LLMs for text generation. Though, I do so for more nuanced reasons. Given the theorem proofs in [1], transformers (from which LLMs such as ChatGPT are made) could be considered "universal" approximators (quotes are mine due to reasons posted in [2]). This is due to the transformers capacity to absorb group information and capture context based on sequence length. What has not been proven is that the space generated in the embedding space (a combinatorial space) is equivalent to creativity. I challenge any computer science and philosopher (or anyone else for that matter) to come up with such proof that a combinatorial space is equivalent to creativity(and therefore avoid plagiarism).
When this is done you can be sure that we can put speculation to rest and see if Chomsky's position and mine is correct or incorrect.Though I would not put my hopes up.
BTW: If someone has any idea or sketch on such a proof, please write to me since I will gladly serve as reviewer.
[1] ARE TRANSFORMERS UNIVERSAL APPROXIMATORS
OF SEQUENCE-TO-SEQUENCE FUNCTIONS?
available at: https://arxiv.org/pdf/1912.10077.pdf
Let's go, but let's break it down into parts first.
What are these "Transformer" models? Transformer models are a powerful type of neural network, like a complex brain for computers, designed to handle sequences of data, especially for language-related tasks like translation, question answering, and text generation (think of tools like ChatGPT).
How are they so powerful? Transformers are surprisingly good at what they do, even though they use relatively few parameters (the settings within the network) compared to other models. This effectiveness and their ability to process complex sequences were somewhat of a mystery.
The development of "universal approximators" is likely a result of ongoing research and advancements in the field. But, why now and not later?
How does this "universal approximator" work? It doesn't create a "universal language" but rather has the ability to learn and represent complex relationships between input data and the desired output. Imagine it as mimicking very intricate connections, but there are real?
I would suggest that you read reference [1] of my post and we can take it from there and I would also suggest reviewing Horniks paper[3,4] and references on embedding spaces so that you have proper context.
References
[3]Hornik, K., Stinchcombe, M., & White, H. (1990). Universal approximation of an unknown mapping and its derivatives using multilayer feedforward networks. Neural networks, 3(5), 551-560.
available at http://robotics.caltech.edu/wiki/images/5/53/HSH.pdf
[4]Multilayer feedforward networks are universal approximators