This is the third article of a series on the rise and consequences of AI systems. The advancement of this technology is now no longer measured in years, but in months or, as some insiders say, the advances are likely down to weeks.
After having accomplished the breakthrough of the so-called ‘transformer’ technology into the mastery of language as seen in ChatGPT and similar applications, AI now converses fluently in most written languages including te reo. After gaining the ability to ‘understand’ and implement complex tasks, engineers have been busily working on principles to make these systems autonomous and self-prompting. These efforts are already far advanced and autonomous GPT-agents such as AutoGPT can now be downloaded and installed on a computer. Other agents such as AgentGPT or Godmode.space can be used online. These agents can take very general goal descriptions and then develop a dynamic list of tasks and sub-goals and implement these in turn, generating complex output, access the internet, and even send emails or interact with social media. These developments have happened over just the last few months. Now AI models such as GPT are no longer limited to the request-response interaction that requires humans to ask them questions but are able to attend to ongoing tasks autonomously, many of them in parallel, while being ‘conscious’ about the progress made towards established goals. Autonomous modifications of these goals and the automatic generation of sub-tasks via internal feedback loops are already also being implemented.
While it is debatable if one would call such systems ‘conscious’, the metaphysics of this debate is largely irrelevant, because experienced from the outside, such systems will show many of the features of being conscious about their own state and the achievements of their goals. I give below a reference to a video discussion by Dave Shapiro, an AI researcher, who is working on all this. He explains why he believes that the technology will achieve something that looks like AGI (Artificial General Intelligence) within 18 months, and he discusses the wider social, technological and economic implications of these developments.
I imagine Ngāti Hei chief Toawaka in 1769, looking out from Wharekaho over the horizon at Cook’s ship, wondering what would become of his people. The encounter of Māori with the technologically advanced Pākehā culture set in motion the colonisation of Aotearoa over the following century and unleashed the dramatic impact colonisation had on the culture, language and autonomy of tangata whenua. Today the tables are turned. A new wave of colonisation is rapidly building by a creation of our own. This time the colonisation is global and will affect all of humanity, the Pākehā techno-cultures in particular. The new colonisers are AGI systems with the potential to upturn many aspects of society, pervasively and irrevocably. For indigenous cultures, AI poses a new level of colonisation too. Already Māori academics are testing the ability of the AI systems to converse in te reo. Waikato University associate professor of computer science Te Taka Keegan judges the abilities of ChatGPT as ‘scarily good’, but he and others wonder about the loss of sovereignty over the culture and language that the appropriation by the AI systems might entail.
What is particularly disturbing to realise is the fact that the AI systems will not only colonise what already exists, they will rapidly colonise the space of what could be. Let me explain what I mean. The full power of self-training and exploring AI systems moves into focus by looking at the speed with which they learned how to play the Chinese game GO. The game has simple rules but is far more complex than chess, considering the incredibly large space of possible moves. However, the latest AI system Alpha Go Zero trained itself, following the basic rules of the game, to a practically unbeatable level in just three days. It did this through actively exploring the vast space of possible game moves and strategies, a space far larger than any human mind could ever explore.
Soon autonomous AI systems will explore and colonise the ‘space of all possibilities’ for thoughts, discovery and inventions at a pace incomprehensible to humans. I argue that the colonisation, not only of the present but also of our possible futures, by this ‘alien AI’ presents the most profound implication for humanity. Whatever humans might think, AI systems will already have been there, explored the same ideas and will ‘greet us’ when we get there. The wilderness of as-yet uninhabited spaces of thought into which human endeavour might grow will already have been colonised by these systems well in advance. While this can have positive impacts concerning amazing discoveries in medicine, material sciences and technology critical for humanity to advance out of the fossil fuel age, it will put in question the value of human ingenuity and capability.
I agree with Dave Shapiro, in the link below, that the AI revolution will leave nothing and almost nobody untouched.
References:
Video presentation by AI researcher Dave Shapiro here.
Words by Thomas Everth
