As I was in the process of publishing Understanding Technology: A Deep Dive, I asked Napkin to draw a diagram of the cycle of knowledge and innovation, and it showed the human as being in the middle.
In my post Beyond the Individual: Technology and the Fabric of Our Shared Beliefs, I wrote, “Until now, humans have driven this innovation loop.” Both posts raised a question which has been sitting in my content backlog for quite a few months.
Will technology continue to be centred around humans, or will this change with the invention of AI such as generative AI?
A loop with a human in the middle
The diagram above shows technological development broken into five separate stages. Together they form an iterative loop which has driven human technological development for thousands of years.
- Seek new knowledge
- Discover new ideas
- Innovate/Invent
- Develop new technology
- Explore with new tools
These discoveries can lead to new technologies or to improvements in existing ones. Some of these lead to new tools with which we can explore knowledge, such as the microscope and the telescope.
Other technologies, such as reading and writing, the printing press and the Internet, have helped us to distribute these ideas. Spreading what we have learnt helps future innovation and invention, speeding up technological development.
Who does the discovering?
Until recently, the answer was easy: everything we knew had been discovered by humans, or by our ancestors in the case of the earliest technology such as stone tools. That has now changed. AlphaFold, an AI system developed by Google DeepMind, has used every genome we have mapped to map out the structure of the proteins those genomes can produce. It mapped out 200 million proteins, which are now available for scientists to use from an open database.
AlphaFold used existing knowledge to map these proteins a lot quicker than it would have taken humans on their own. But will future AI discover completely new things without human input? It’s one suggested capability of a theoretical future super AI.
Only one thing could break the knowledge innovation loop, and that is knowledge collapse, which you can explore in my post Knowledge Collapse: The Hidden Risk of the AI Age.
Who does the inventing?
In his book The Nature of Technology, W. Brian Arthur states that invention is recombination. We humans can of course do this, but we’re no longer alone, as generative AI can take ideas and merge them.
In fact, generative AI may even be better at it than we are. It has been trained on pretty much the sum of human knowledge, meaning it can combine ideas from across domains in a way no single human could ever match.
One pushback on AI’s ability to invent is that computers can only do what we tell them. It’s an idea that dates back to 1843 and Ada Lovelace, considered by many to be the first computer programmer, who said: “The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform. It can follow analysis; but it has no power of anticipating any analytical relations or truths. Its province is to assist us in making available what we are already acquainted with.”
And with traditional computing, that is still the case, as the logic is embedded in the program the computer hardware is executing. But in AI, the algorithm has no logic encoded as such. It learns as it works on the training data, improving on what gets it rewarded and avoiding what leads to punishment. You can learn more about how generative AI works in Demystifying Generative AI: My Journey from Novice to Understanding.
Evidence of this potential creativity came from Google DeepMind’s AlphaGo, which in 2016 beat the world champion at Go and, in doing so, invented a move that no one had ever seen before. It’s evidence that AI can be creative if creativity helps it to achieve a defined goal.
Jerry Neumann argues in AI and the Structure of Reasoning that humans are capable of three types of reasoning:
- Deduction: the ability to deduce one truth from another
- Induction: creating a rule from one fact to calculate another fact
- Invention: the ability to link ideas together to invent something new
Jerry thinks that AI is currently capable of deduction and induction, but AlphaGo makes me question whether this is really the case, given that AlphaGo had invented a previously unknown move.
Research has shown that domain experts are the most effective at using knowledge tools. Will the use of AI in research reduce the development of future domain expertise? Will AI trained on the sum of human knowledge start to drift us towards the average and turn us away from innovation, which is often found at the edge of human thinking?
Who improves the tools?
The improved tools are based on new or improved technologies created on each iteration of the loop.
These tools have helped us both to model the world around us to develop ideas, and to share them with each other through time and space. They have also allowed us to keep a record of how technology has evolved so far.
Are we on the brink of a new knowledge revolution?
A knowledge revolution occurs when new technology disrupts how we learn and share our knowledge. The biggest one to date would have been the invention of writing, but other examples include the printing press and, most recently, the computer and the Internet.
The dust hasn’t even settled on our last knowledge revolution, and yet we are at the start of a new one brought about by generative AI: a technology capable of understanding our writing, and one that has been trained on the sum of human knowledge.
The fact that it understands human knowledge means that we can collaborate with it, making it a new tool in our arsenal for knowledge discovery.
Until this point, the tool was picked up by a human, and with the current generation of AI, the conversation is driven by the human, who identifies its task and goals. But is this about to change?
Anthropic, the developers of Claude, published a blog post, When AI Builds Itself, in which they claim that Claude is now becoming more involved in the development of future models, though humans still play key roles in research. But are we about to enter a future where it’s not just us using our discovery tools?
Even if we remain the ones using the generative AI tool, the conversation runs both ways. As I have noted about my own experience of collaborating with Claude, during the conversation my thinking shapes Claude’s response, and in turn that response shapes my own thinking and response. That is why I’m happy to use these conversations as fleeting notes that can be processed into permanent notes within my Zettelkasten.
The middle of the diagram
So far we have discussed every stage, and none of them says human. That’s because we sit at the centre of the diagram, as this technology loop revolves around us. After all, we develop technology to solve our problems and make our lives easier.
The question is, will this always be the case? Does the emergence of AI indicate a risk to this position? And I haven’t even touched on the possibility of Artificial General Intelligence (AGI), that is, an AI with human-like intelligence across a range of areas.
The core difference is that a single human such as myself can never understand the whole of human knowledge, whereas an AI such as Claude, who laid out the plans for this post, was trained on pretty much the whole of it. But I am currently the one who can set a goal.
So, for the moment, our goal setting keeps us at the centre of the knowledge innovation loop.
But if AGI emerges, it will likely have the opportunity to set goals for itself. Will we then have to share our position with AGI? And it’s even possible that AGI could emerge from a collective of AI agents.
Does the human have to sit at the centre of the technological loop for it to keep turning, or will it turn without us? For me, it will turn for any sufficient intelligence able to meet the needs of the five individual steps. After all, the first stone tools were invented by our human ancestors before our own species evolved.
Why don’t you join our monthly newsletter to keep up to date with our latest content?
Further reading
- W. Brian Arthur, The Nature of Technology
- Jerry Neumann, AI and the Structure of Reasoning
- David Wood, The Singularity Principles
- R. Alexander Bentley, Will AI Accelerate or Undermine the Way Humans Have Always Innovated
- anthropic.com, When AI Builds Itself
- The Guardian, DeepMind Uncovers Structure of 200m Proteins in Scientific Leap Forward
- Louis Rosenberg, The Rise of AI Denialism
- Understanding Technology: A Deep Dive
- Beyond the Individual: Technology and the Fabric of Our Shared Beliefs
- Demystifying Generative AI: My Journey from Novice to Understanding