>[!idea] >Artificial intelligence has interesting practical applications as a means of accelerating the stochastic methods that facilitate the evolution of [[ordered complexity]]. Understanding the mechanisms of artificial intelligence informs our understanding of human intelligence, and may provide a means for incorporating non-human intelligence into the conversation. > Artificial Intelligence (AI) is the technocratic ambition to reproduce human learning and decision-making within a computer. The origins of the field lie within Alan Turing's 1950 essay, [Computing Machinery and Intelligence](https://www.cs.colostate.edu/~howe/cs440/csroo/yr2015fa/more_assignments/turing.pdf). However, it is widely agreed that the discipline itself was conceived at the [proposal](https://raysolomonoff.com/dartmouth/boxa/dart564props.pdf) of what is now known as the [Dartmouth workshop](https://en.wikipedia.org/wiki/Dartmouth_workshop). Early AI research gained momentum in the mid-1950's through '60's before losing government funding in the 1970's from DARPA[^2] and as a result of criticism[^3] in the Lighthill report. Another round of research and funding occurred in the '80's through mid-90's in the private sector as specialist AI developed out of domain-oriented approaches. Rather than attempting to achieve general [[learning]] through a symbols-based approach, corporations sought to automate manufacturing and logistical processes through brute-reasoning. Funding again receded as these approaches began to incur costs associated with maintaining large knowledge-bases as tasks grew in [[complexity]]. From 2000 onward, the internet provided a new data repository for AI development. Watson is the first iteration of an AGI machine. Task transfer and composability are the main challenges in current AI development. The ability to identify the broad conceptual movements for a given task to a new object, and the ability to identify the subject of "it" in general speech. ##### Thinkers - Darwin: adaptation and diversification - Turing: Proof of a universal computation machine. A device that can perform any mathematical computation - [John McCarthy](https://en.wikipedia.org/wiki/John_McCarthy_(computer_scientist)): Proposes the study of machine learning. >The study is to proceed on the basis of the conjecture that every aspect of learning or any other feature of [[intelligence]] can in principle be so precisely described that a machine can be made to simulate it. - Norbert Wiener: cybernetics - Frank Rosenblatt: perceptitron ##### Mechanisms - Good Old-Fashioned AI: formal logic [[Paradigms|paradigm]] - Physical symbol system hypothesis: Any formal system that represents physical patterns as symbols (eg. $t$ for time, or $Δ$ for change) and combines them into manipulable structures can produce rational behavior. >A physical symbol system has the necessary and sufficient means for general intelligent action.[^1] - Rule-based reasoning: Highly complex if/then statements. These are closer to the instruction sets in a universal Turing machine, whereas phys. system symbol hypothesis depends on higher-level language. - Nouvelle AI: [[Complex Adaptive Systems|complex adaptive system]] paradigm; adaptive intelligent systems. - Parallel distributed processing (PDP): "back-propagation algorithm" successful with pattern recognition. - Deep neural networks (DNN): DeepMind - Foundations models - ChatGPT [^1]: Allen Newell, A. and Simon, H.A. (1976). Computer science as empirical inquiry: symbols and search. Commun. ACM 19, 3 (March 1976), 113–126. [DOI](https://doi.org/10.1145/360018.360022) ### Species of Intelligence - Artificial General Intelligence: AI with [[Range - David Epstein|range]]. Can respond to wicked challenges. - [`LISP`](https://en.wikipedia.org/wiki/Lisp_(programming_language)): Early language used in AI development based on LISt Processing. Used to break down defined to intelligent behavior into functions that could manipulate symbols. - Domain-oriented, specialist/expert systems: Emerged in 1980's through '90's. Perform specific tasks beyond the ability of human precision. Later examples are Deep Blue (chess) or AlphaGo (Go). Can determine rules of the domain then iterate through best scenarios given a set of parameters. Require large knowledge-bases to maintain learning environment. - Writing for Time magazine in 1996, Kasparov observed: “I had played a lot of computers but had never experienced anything like this. I could feel–I could smell–a new kind of intelligence across the table.” - Animats: complete systems with distinct behaviors and sensors that store computational demands within the form of the machine rather than in a CPU. - **note:** Cf with idea of [[extended cognition]]. - The deep family: Deep Blue, Deep Mind. Brute force + heuristics. - these capabilities are a function of the relationship between processing power and data sets. It is a relatively simple model but does not translate to human or biological intelligence. - Watson: first AGI iteration developed by IBM. >Why not turn Watson into a physician? Once again, task transfer and generalization have turned out to be very difficult. * **note:** Cf with [[Our world is a black box#^6141bb|DeepMetab]] - Generative Pre-trained Transformer 3 (GPT-3): chatGPT - AlexNet, a neural network for image recognition published in 2012 by Alex Krizhevsky, Ilya Sutskever and Geoffrey E. Hinton. ### Blueprints of Intelligence How emerging technology is represented in media is important. In the case of AI, images of disembodied personalities inside the machine ([SIRI](https://en.wikipedia.org/wiki/Siri)) and isolated brains ([Krang](https://www.youtube.com/watch?v=qgc8swYrcpM)), cultivate distrust of how the technology can contribute to human flourishing. Popular images of AI are largely an attempt to anthropomorphize the technology so it is [[Legibility|legible]] to a lay audience, whereas the inner workings are less immediately accessible. However, these processes provide insight in to how we might better define intelligence itself. The study of AI represents efforts toward a [blueprint of intelligence](https://www.noemamag.com/blueprints-of-intelligence/). > [!quote] Samuel Watterworth > AI does not have a skull to autopsy, which some may consider unfortunate, but it could also be seen as an opportunity to encounter intelligence in isolation. What does it look like? Much in the same way that ancient structures predate the development of architectural drawing, the observation of intelligent behavior predates a coherent model. According to the metaphor of a blueprint, species of artificial intelligence are less entities in themselves, and rather *representations* of how we understand the phenomena of intelligence. The best illustrations to date are neural networks. Intelligence emerging from building blocks of logic controllers. The first neural nets drew from biological references, since then they have become more precise and mechanistic. Neural networks lack the immediacy of recognition of intelligence because they lack the form we have been trained to recognize, but that is their strength. They break the phenomenon of intelligence into its constituent components, allowing us to understand that it isn’t based on heuristics such as attire, skin color, or a manner of speaking. The challenge with this mental model is avoiding the trap that intelligence is entirely mechanical or electrical. #### Types of Blueprints - neural network architecture diagrams - Pitts/McCulloch Artificial neuron  - Frank Rosenblatt’s “Perceptron,” both a simple visual classification algorithm and a room-sized machine, is richly illustrated with beautiful drawings in organic strokes, often reminiscent of cell-like structures.  - Oliver Selfridge’s “Pandemonium” (1959), a **pattern learning** system that learned to translate Morse code into text, is another creatural take on neural networks.  - Corinna Cortes and Vladimir Vapnik developed “support-vector networks,”  - **Modern deep learning** neural networks are capable, in some cases, of doing things that humans do, and in a few cases even surpass our abilities. At the same time, the [[field]] **has distanced itself from relating to human example and established itself as a separate, if contested, branch of intelligence.**  - We see this distancing reflected in diagrams: **Modern neural network architectures are rendered in an increasingly abstract, flowchart-like diagrammatic similar to engineering. Earlier biomorphic connotations have disappeared from both the linguistic and visual vocabulary.**  *Adji B. Dieng, Chong Wang, Jianfeng Gao and John Paisley built TopicRNN, a neural network that models topics in language data, in 2016*. - **a poetic reading of metaphor and symbolism in AI diagrams prompts us to consider how their creators think about cognition. This may lead to more productive conversations about the risks and opportunities of artificial intelligence research than we could have amid sci-fi tropes.** And it elevates the question: ** ## Notes - Wikipedia [summary](https://en.wikipedia.org/wiki/History_of_artificial_intelligence) discusses ancient models of AI, such as golems, Talos (guardian of Crete), and *[Takwin](https://en.wikipedia.org/wiki/Takwin)*, synthetic life created by Muslim alchemists #### Responses - Is there a difference between [[Consciousness]] and intelligence? We speak as if the two are interchangeable but I am beginning to wonder if each is distinct and yet intimately related. - Samuel Watterworth discusses the non-relatability of neural networks, is this a fault of cultural conditioning? Does this abstraction help us recognize intelligence in a non-humanoid form? - **Note the parallels between the building block model of the neural node for intelligence and the cellular model for consciousness.** - We might carry this observation forward into ==an essay exploring the spectrums of intelligence and consciousness.== How mitochondria form a relationship with cellular life, which congregated into bodies. Comparisons could be drawn between this model and social models that include the BE - **How does a mechanistic portrait of the neural network fit in with a regenerative paradigm?** - Is bit-based intelligence superior to carbon-based intelligence? Do different bases of intelligence permit different bases of life? Is the way we define life too narrow? - "**The disembodied demon may be a better stand-in for artificial intelligence than most other creatures, as its appearance shouldn’t be distracting.**" - **Note:** cf: the Discworld demon inside a watch. - These images connote a tendency toward [[Panpsychism]]. The machines *are* alive, but not in a way we can adequately represent through their mechanisms alone. > [!quote] Samuel Watterworth > It seems we lack the vocabulary and imagery to think and talk about intelligence without either invoking the animate or the spiritual. #### Challenges - Task transfer: specific tasks cannot be generalized then adapted to similar but different tasks. - Explainability: reasoning for a successful outcome is opaque to human understanding. - Alignment: ai goals and human goals may be different. - Bias: the training data contains implicit human bias - Distribution shift: when the test or field data differs from the training data. - Truth Evaluation: LLMs [famously](https://themarkup.org/news/2024/03/29/nycs-ai-chatbot-tells-businesses-to-break-the-law) provide inaccurate information. #### Projections - Data convergence: AI will connect disparate data sets at scale to create new forecasting and diagnostic models across a number of domains. - Intelligent infrastructure (II): an internet-of-things interoperation approach. Will likely utilize a form of [artificial swarm intelligence](https://en.wikipedia.org/wiki/Swarm_intelligence#Artificial_Swarm_Intelligence_(2015)), in which many devices cooperate to adapt to and directly modify their environments. "The real-time [instrumentation](https://www.amacad.org/publication/so-simple-beginning-species-artificial-intelligence) of our environment." - [Physical AI](https://arxiv.org/pdf/2105.06564.pdf) (PAI): an expansion of the current paradigm of digital AI, which emphasizes the data and processing within the machine. PAI expands the field to include machines that leverage chemistry, comp. science, biology, mechanical engineering, and material science to design and construct themselves. Basically it means robots. #### Questions - Did AI theory occur in parallel development with behaviorism in [[psychology]]? - Is neuroscience and AGI occurring in parallel development? - How does the physical symbol system hypothesis respond to the discussion of [[Plant Cognition]]? - From Samuel Watterworth: "What other conceptions of intelligence are desirable, and how can we represent them?" [^2]: Crevier, D. (1993). _AI: The Tumultuous History of the Search for Artificial Intelligence_. Basic Books. [^3]: Lighthill, J. (1973). Artificial Intelligence: A General Survey. _Artificial Intelligence: A Paper Symposium_. Science Research Council. >[!cite] [Nigel Shadbolt](https://www.amacad.org/publication/so-simple-beginning-species-artificial-intelligence) >The history of AI has constantly intertwined the discovery of new ways to reason and represent the world with new programming languages and engineering [[Paradigms]]. Computer science, in turn, has been enriched by these cycles of development.