Supersmart: the role of abstraction in AI, Art, and Life
by A M Howcroft & ChatGPT
Understanding the intricacies and impact of natural and artificial intelligence (AI) is becoming paramount, whatever your industry. The concept of levels of abstraction emerges as a critical framework we all need to know. It provides a compelling lens to explore how consciousness might arise from lower-level processes, hints at when AI might develop general intelligence, and guides collaborative approaches that are easier for humans and AI to comprehend. Learning how it works might also give you a tool to make smarter decisions.
Abstracts in the wild
Let’s start with an example of levels of abstraction based on computer software. Here’s how an application is constructed:
Machine Code: The fundamental level of binary instructions directly processed by the computer's CPU.
Assembly Language: A slightly more readable layer using mnemonics, yet closely tied to hardware architecture.
High-Level Programming Language: Languages like Python and Java simplify coding by abstracting away the complexities of machine and assembly language.
Framework/Library: Tools like React for web development build upon high-level languages, streamlining the development process by allowing developers to focus on the unique aspects of their software rather than reinventing the wheel.
Application Logic: Focused on implementing business logic and user requirements using tools from the previous layers.
UX (User Experience): The most abstract layer where end-users interact with applications, whether that is a spreadsheet, or a computer game.
Each layer simplifies and abstracts complexities from the previous one, making it easier for people to understand and use the applications. We should note that there are far more people that can use a spreadsheet (level 6), than there are with the expertise to adapt the underlying machine code (level 1).
We can also recognize that certain roles stick to different layers of abstraction; software developers spend a large amount of time working at level 5, they are not typically building compilers for the High-Level Programming languages (level 3), or even writing the frameworks (level 4) – these are specialized jobs.
Overall, though, things tend to get simpler for the general user as we move up a layer of abstraction – but not always! Let’s talk about an exception next.
Abstract Art: Cave Paintings to Modernism
Art's journey from simple (abstract) representations of animals and people in cave paintings to the realistic paintings of the Renaissance which added techniques like linear perspective, is an interesting journey. Ultimately, we reached a point where artists could produce works of art that were nearly indistinguishable from the real thing. However, artists did not stop here, they began to abstract their images to distort real-world representation, challenging the viewer to interpret based on composition, emotion, and color rather than easily recognizable subjects. Think of Edvard Munch and his famous picture ‘The Scream’, which officially is a piece of expressionism – in other words, a type of abstraction from the real world to express subjective feelings rather than physical reality.
Complete abstraction in modern art is the most challenging for laypersons to grasp, as it is non-representational and demands engagement based on emotional or conceptual responses, detached from traditional visual references. This requires a shift in perception, focusing more on elements than on literal meanings.
The lesson for us is that as we move into the upper echelons of abstraction, meaning can be lost without context, and expertise is required to interpret the results.
Abstract Math: Building Complex Ideas
In mathematics, foundational concepts pave the way for more abstract ideas. From basic arithmetic to algebra, and then to higher concepts in calculus and geometry, each level builds on the previous, catering to an increasingly specialized understanding.
Category theory, generally considered to be towards the top of the abstraction hierarchy, focuses on relationships and transformations between mathematical structures. It deals with concepts such as objects, morphisms and functors, which are very general notions that can be applied across various areas. Essentially, it's a unifying framework that reveals connections across different mathematical domains, though it requires a solid foundation in advanced mathematics to fully grasp – something only a few ever master.
As with art, higher levels of abstraction in math are beyond the capabilities of most laypeople. A pattern is emerging, where low levels of detail and high levels of abstraction are both equally hard to comprehend for most of us, as we sit in a comfortable middle zone. Abstraction makes our life easier, in this zone.
The Human Brain: A Spectrum of Abstraction
The human brain also operates at various levels of abstraction:
Low-Level: Basic neural processes like sensory perception and motor responses.
Intermediate-Level: Complex processes such as memory, emotion, and learning.
High-Level: Thinking, reasoning, decision making, understanding concepts like love or justice.
Consciousness: is a high-order state of the brain. Many consider it to be an emergent property arising from complex neural processes at lower levels of abstraction.
Understanding how these layers interact is crucial in fields like neuroscience and AI. We should note that most people have a view on how decision making takes place in their mind, and understand emotions, but perhaps could not explain neural pathways or the latest theory of consciousness; again the low detail and high abstractions are harder to grasp.
AI's Current Abstraction Level
If we compare AI to the human brain, it currently operates mainly at intermediate levels, with some forays into high levels in constrained environments. Many AI systems are also very comfortable at the low levels of detail, as well as the intermediate. Sam Altman, CEO of OpenAI, has mentioned that they are presently working towards General Intelligence for GPT 5.0 (i.e. high level abstraction), as are many other AI players and academics – this is the Holy Grail for AI research.
What we don’t know yet, is the size of the gap between high level thinking and consciousness. Does this emerge soon (or immediately) after we have high-level capabilities, or is there another cognitive leap involved? We may not have long to wait before we can answer this question. With current investment levels, we should expect to see more high-order AI systems in the near term.
How to Make Smarter Decisions
Intelligence is frequently ranked by how high up the abstractions layer you can move. For example, I think we would probably agree that to fully grasp category theory in math, you need to have a high degree of intelligence. However, everybody also knows people that are intelligent but perhaps not smart. In many different areas, I have noticed that people who are great at solving problems are the ones that can easily move between different levels of abstraction. These are the people that colleagues tend to call smart.
In a software example, if a piece of code doesn’t work it might be because there was a mistyped command (a low-level issue) or that the program structure is wrong (intermediate level), or occasionally the whole concept might be flawed (high level). Too frequently people get fixated on finding the issue at one level. The smart people are those who can switch between different modes of thinking, across layers of abstraction. This is not news, it’s embedded in the common adage which suggests someone is foolish because: they couldn’t see the woods for the trees. Next time you have a problem, ask yourself how the problem would look from a different perspective – and consider a switch in your level of abstraction.
The Future is Exciting and Daunting
We can see that the human brain and AI both operate using layers of abstraction, and that at higher reaches (and at low levels of detail too) it becomes hard to comprehend without context and expertise. What if there is an even higher layer of abstraction for minds? Let’s call it superintelligence or ultra-consciousness. This opens up new possibilities and challenges.
AI's ability to process information rapidly and store vast amounts of data, and perhaps its ability to switch rapidly between layers of abstraction, means that if it reaches consciousness, it could evolve into this ultra-conscious state. It might possess the ability to analyze and understand complex systems and concepts far beyond human cognitive capabilities, integrating vast amounts of data and insights instantaneously. Such a breakthrough could lead to unprecedented advancements in various fields, but also raises profound ethical and existential questions.
The Human-AI Partnership
Today, AI operates up to high levels of abstraction, and we can partner with it to leverage these capabilities. This model of human-AI partnership, central to SWARM’s Challenge Engineering approach, delivers benefits by combining human creativity, ethical management, and capacity for questioning, with AI's computational power. It is important that we continue to work with AI and learn with it, and from it.
This symbiotic relationship enhances our problem-solving abilities, driving innovation and progress in ways that were previously thought impossible. Few will object to significant reductions in food waste, increasing crop yields, and finding better ways to make agriculture sustainable and regenerative for the environment. By understanding and utilizing different levels of abstraction we can enhance our capabilities and drive progress in exciting new directions, while ensuring we continue to explore the ethical considerations.
Check out our website at SWARM to learn more about how we are harnessing this powerful synergy in the domain of agrifood.