When AI Tells You What You're Missing

Dear friend,

As I find myself grappling with the relentless tempo of our age, I must share, with you, my thoughts on the AI zeitgeist.

Previous generations of artificial intelligence, those specialist-trained models, wrestled with static situations, grappling with data in a multitude of forms, searching for known truths. Is there a tumour lying in wait? A potential drug candidate? An impending danger? A fraudulent transaction? Or a need for maintenance? You could claim protein folding as an exception, but peer closer, and you find the models were instructed on the art of folding, and the measure of success—the known.

The world has seen the arrival of ChatGPT and its kin, the artificial general intelligences (AGIs)—a new breed. I have touched upon the emergence of new behaviours in these large language models (LLMs) in one of my previous letters, and one can't help but wonder if it is language, rather than the overwhelming number of parameters, that is the driving force. Some contend that language is the very essence that enables conscious thought. I will return to this topic in a later letter. But emergence aside, I sense a more fundamental, radical shift in the works.

Cast your eyes upon today, and you will see the migration from static to dynamic situations, courtesy of AGIs. These AGIs aren't just cogs in a chain of events; they're capable of being the engine, driving processes by providing input and making decisions. Training has become passé, as simple or less simple prompts are all it takes to get these AGIs going. That is one way to look at "general."

More importantly, and this is where things get downright revolutionary, ChatGPT and other LLMs, no doubt, have the power to divulge the unasked, reveal the unknown. Take a friend of mine, for instance, who is revamping his garden. He asked ChatGPT to design a harmonious arrangement of plants and flowers, and it complied without missing a beat—does it ever? But, here is the kicker: it threw in a warning, unasked for, some plants don't fare well in the sun, be careful!

Now, put two and two together—processes and known unknowns—and you have got AGIs that allow you to build sturdier processes, demanding less initial setup, parameterisation, and training. And, those AGIs could tip you off about things you haven't even considered — the known unknowns or the unknown knowns, depending on how you see it. I am not talking about the AI admitting its ignorance, "I don't know"; I am talking about the AI perceiving your ignorance, "You don't know."

In the context of using AI to help us grasp our unknown, a study by Grace, Maher, Wilson, and Zhu (2016) proposes a process model for computationally aiding the discovery of creativity. The study shows how AI can be employed to boost creative problem-solving and generate novel ideas.

The next step, the exhilarating, transformative, hell, even frightening moment, is when AI tells us about the unknown unknowns because it is capable of setting up the discovery, creative, exploratory processes on its own. Maybe autonomous agents like godmode.space are the first step in that direction. Somewhere, something — as opposed to someone — in summertime is "emergent" enough to uncover the unknown unknowns.

As AGIs grow more sophisticated and interconnected, they might display emergent properties that allow them to forge surprising connections, cook up innovative ideas, or suggest groundbreaking experiments. Think about open-ended AI systems designed to explore, learn, and generate new content ceaselessly, unbound by specific goals or predefined endpoints. Wang's work on the POET system is one such example, as is curiosity-driven AI, which focuses on crafting algorithms that are intrinsically motivated to explore and learn about their environment, potentially unearthing new knowledge.

This makes me believe, like many of you, that we are on the verge of... something that some fear, while others embrace. Wherever you stand, let's stay curious and open-minded as we journey through this era of accelerated time in human history.

Yours,

L.

NB: As is customary, I present to you a compendium of sources for further exploration.

1. Argüelles, J. (2002). Time and the technosphere. In S. L. Gardner & F. J. Schmitz Jr. (Eds.), The Postmodern Turn: Essays in Postmodern Theory and Culture (pp. 71-87). Ohio University Press.
2. Callaway, E. (2020). 'It will change everything': DeepMind's AI makes gigantic leap in solving protein structures. Nature. https://www.nature.com/articles/d41586-020-03348-4
3. Grace, K., Maher, M. L., Wilson, D. C., & Zhu, F. (2016). A Process Model for the Computationally Aided Discovery of Creativity. International Journal of Design Creativity and Innovation, 4(2), 66-84. https://doi.org/10.1080/21650349.2015.1026941
4. Karmakar, B., & Pal, N. R. (2018). How to make a neural network say “Don't know”. Inform. Sci. 430-431, 444-466. https://doi.org/10.1016/j.ins.2017.11.061
5. Mack, D. (2019). Language could be an important part of artificial general intelligence and synthetic consciousness. https://medium.com/octavian-ai/language-could-be-an-important-part-of-artificial-general-intelligence-and-synthetic-consciousness-457a977cebc
6. Pathak, D., Agrawal, P., Efros, A. A., & Darrell, T. (2017). Curiosity-driven Exploration by Self-supervised Prediction. In Proceedings of the 34th International Conference on Machine Learning (ICML). https://arxiv.org/abs/1705.05363
7. Service, R. F. (2020, November 30). 'The game has changed.' AI triumphs at solving protein structures. Science. https://www.sciencemag.org/news/2020/11/game-has-changed-ai-triumphs-solving-protein-structures
8. Wang, G., Liapis, A., & Stanley, K. O. (2019). Paired Open-Ended Trailblazer (POET): Endlessly Generating Increasingly Complex and Diverse Learning Environments and Their Solutions. arXiv preprint arXiv:1901.01753. https://arxiv.org/abs/1901.01753
9. Wei, J., et al. (2022) Emergent Abilities of Large Language Models. Google Research. Stanford University. UNC Chapel Hill. DeepMind. https://arxiv.org/pdf/2206.07682.pd
10. Uhres, L. (2022). Intertwined Intelligences. https://www.dhirubhai.net/pulse/intertwined-intelligences-laurent-uhres
11. Uhres, L. (2022). I don't fear AI. https://www.dhirubhai.net/pulse/i-dont-fear-technology-laurent-uhres