A (really) Short History of Artificial Intelligence From Early Dreams to Open Source Collaboration Harish Pillay @harishpillay:matrix.org, [email protected] AI Verify Foundation Slides at: https://tinyurl.com/8czuu59w

Sometime in 1948

Grey Walter's tortoises - https://www.youtube.com/watch?v=lLULRlmXkKo

No content

1950 ● Alan Turing proposes the Turing Test (1950) a. In the 1950 MIND - A QUARTERLY REVIEW OF PSYCHOLOGY AND PHILOSOPHY, Alan Turing asks “Can Machines Think”.

1950 ● Alan Turing proposes the Turing Test (1950) a. In the 1950 MIND - A QUARTERLY REVIEW OF PSYCHOLOGY AND PHILOSOPHY, Alan Turing asks “Can Machines Think”. b. He rephrases that statement instead to “The Imitation Game”. [0]

1956 ● The Dartmouth Workshop (1956) a. Lead by Prof John McCarthy, he invited Marvin Minsky, Claude Shannon, Nathaniel Rochester and together they coined the term "artificial intelligence"

1956 ● The Dartmouth Workshop (1956) a. Lead by Prof John McCarthy, he invited Marvin Minsky, Claude Shannon, Nathaniel Rochester and together they coined the term "artificial intelligence" b. Held over an eight week period, between 11 to 47 people participated [1]

1956 ● The Dartmouth Workshop (1956) a. Lead by Prof John McCarthy, he invited Marvin Minsky, Claude Shannon, Nathaniel Rochester and together they coined the term "artificial intelligence" b. Held over an eight week period, between 11 to 47 people participated [1] ● Early AI programs: Logic Theorist and checkers-playing program

https://www.scienceabc.com/innovation/ what-is-artificial-intelligence.html

No content

1960s ● The “perceptron”, an algorithm invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt, funded by the United States Office of Naval Research and was implemented, in software, on an IBM 704.

1960s ● The “perceptron”, an algorithm invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt, funded by the United States Office of Naval Research and was implemented, in software, on an IBM 704. ● That was subsequently implemented in custom-built hardware known as the "Mark 1 Perceptron”.

1960s ● The “perceptron”, an algorithm invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt, funded by the United States Office of Naval Research and was implemented, in software, on an IBM 704. ● That was subsequently implemented in custom-built hardware known as the "Mark 1 Perceptron”. ● It was one of the first artificial neural networks to be produced.

1960s ● The “perceptron”, an algorithm invented in 1957 at the Cornell Aeronautical Laboratory by Frank Rosenblatt, funded by the United States Office of Naval Research and was implemented, in software, on an IBM 704. ● That was subsequently implemented in custom-built hardware known as the "Mark 1 Perceptron”. ● It was one of the first artificial neural networks to be produced. ● First AI Winter kicks in as the systems have severe limitations and research funding began to dwindle.

Mark 1 Peceptron - at the Smithsonian Museum

1970s - 1980s ● Research focus moves to Symbolic AI and Knowledge Representation

1970s - 1980s ● Research focus moves to Symbolic AI and Knowledge Representation ● Expert systems - Solving problems in specific domains

1970s - 1980s ● Research focus moves to Symbolic AI and Knowledge Representation ● Expert systems - Solving problems in specific domains ● Limitations of knowledge engineering and reasoning hampered adoption

ARPANET - https://en.wikipedia.org/wiki/ARPANET

1980s - 1990s ● Statistical learning algorithms gain prominence

1980s - 1990s ● Statistical learning algorithms gain prominence ● Support Vector Machines (SVMs - a form of perceptron), and decision trees gain popularity

1980s - 1990s ● Statistical learning algorithms gain prominence ● Support Vector Machines (SVMs - a form of perceptron), and decision trees gain popularity ● Increased focus on data-driven approaches

1980s - 1990s ● Statistical learning algorithms gain prominence ● Support Vector Machines (SVMs - a form of perceptron), and decision trees gain popularity ● Increased focus on data-driven approaches ● Groundwork laid for future advances

https://www.infrapedia.com/

2000s ● Increased computational power and availability of large datasets

2000s ● Increased computational power and availability of large datasets ● Revival of neural networks - Deep learning architectures

2000s ● Increased computational power and availability of large datasets ● Revival of neural networks - Deep learning architectures ● Breakthroughs in image recognition, speech recognition, and natural language processing

No content

2010s ● Big Data revolution - explosion of data volume and variety

2010s ● Big Data revolution - explosion of data volume and variety ● Cloud computing platforms - Scalable and accessible resources

2010s ● Big Data revolution - explosion of data volume and variety ● Cloud computing platforms - Scalable and accessible resources ● Democratization of AI - Increased accessibility for businesses and researchers

2010s - current ● Specialization of deep learning architectures for specific tasks ● Deep learning applications in various domains: self-driving cars, healthcare, finance ● Ethical considerations of AI - Biasness, Fairness, and Transparency and 8 other metrics

No content

2010s - current - Open Source & Collaboration ● Rise of open source AI frameworks and tools (TensorFlow, PyTorch) ● Collaborative research and development efforts ● Fostering innovation and accelerating progress

2020s - current ● Foundational Models - Large/Small Language Models capable of generating text, translating languages, and writing different kinds of creative content ● Generative models for creating realistic images and other types of data ● Multimodal models that can process and understand different types of data (text, images, audio)

2020s - current ● Mixture Of Experts (MOE) - A framework for training and using large, efficient multiple foundational models ● Improves scalability and reduces training costs ● Potential for wider adoption of complex AI models ● OpenMOE [2] - an open source implementation to drive innovation

2020s - current ● RAG - Retrieval Augmented Generation ● Using open source (as per Open Source Initiative’s definitions [3]) foundational models to train against corpus of private data for subsequent enquiry

2020s - current ● Testing of AI solutions for fairness, unbiasedness etc, via open source testing tools ● AI Verify’s toolkit [4], is a community driven global effort of the AI Verify Foundation to create a commonly agreed to test framework ● Guiding principles of AVF is on AIVerifyFoundation.sg

The Future from Today ● Sovereign AI - nation states taking charge ● Personal AI - kwaai.ai [5] ● AI Governance Frameworks v2 - from Singapore’s Personal Data Protection Commission [6]

[0] https://academic.oup.com/mind/article/LIX/236/433/986238 [1] https://spectrum.ieee.org/dartmouth-ai-workshop [2] https://arxiv.org/abs/2402.01739 [3] https://opensource.org/deepdive [4] https://github.com/IMDA-BTG/aiverify [5] https://www.kwaai.ai [6] https://www.pdpc.gov.sg/-/media/Files/PDPC/PDF-Files/Resource-for -Organisation/AI/SGModelAIGovFramework2.pdf References

https://matrix.to/#/#ai-verify-foundation:matrix.org Harish Pillay [email protected] @harishpillay:matrix.org