Synthesis and Generative AI

A child piles up sand. When does she decide that the grains have formed a heap? This decision isn't straightforward due to two simultaneous phenomena.

The first process is a self-contained system, building a sand heap analytically. Though precise, it only knows its own models. It struggles to decide when the heap is complete, asking, what number of grains makes a heap? It can't step outside its confines without sacrificing accuracy, as this decision can't be analytically optimized.

To balance this, a second process operates externally, observing the sand heap and questioning its context, purpose, and potential interest. This process accepts ambiguity as part of life. It sees the heap's emergence—a sudden shift in perception. The child understands that the growing pile of sand has become a heap.

Our minds seamlessly integrate these two processes. The left brain directs the first process, while the right manages the second. Despite differing thought processes, our brain unifies experience into one conscious self.

Brains aren't just tools for sensemaking or prediction, they help construct our reality. They must balance personal and communal needs. In the personal sphere, you and your needs compete with others, requiring focused attention.

In contrast, the communal realm necessitates viewing yourself as part of a group, connecting to something larger and existing "through" others. Here, attention needs to be broad, receptive, and aware of experiences beyond yourself, emphasizing alliances.

Both brain hemispheres are involved in our lives, yet psychiatrist Iain McGilchrist emphasizes the right hemisphere's role. It's aware of the left's functions, while the left is unaware of the right's. The right hemisphere provides intuitive oversight, using the left's analysis to create new intuitions and connections.

Our existing AI exhibits a "left-brained" characteristic, working within its own system of data and algorithms for optimization. It's similar to the left hemisphere, focused on internal coherence, while humans, akin to the right hemisphere, unify the world's forces and connect to reality.

This metaphorical tension between AI and human functioning highlights unique cooperation. AI uses language, abstraction, and analysis, while humans rely on experience, empathy, and metaphor. AI manipulates static elements well but struggles in a changing world.

This leads to two distinct design approaches for future AI and considerations for "artificial general intelligence" (AGI).

One approach focuses on "left-brained" AI—controlling facts with internal consistency, while relying on humans for context, meaning, and care. Here, machines serve humans. This path is popular due to the challenge of developing AI mimicking human right hemisphere functions.

However, we want machines that can correct contextual mistakes and understand our intended meanings. The design challenge here lies in connecting highly "left-brained" AI to holistic humans in a way that enhances human capabilities.

Alternatively, we could design AI with asymmetry, mirroring the human brain's evolution. Such AI would provide a synthesized perspective before interacting with a human, applying computational power to intuition and addressing human paradoxes. Some envision this as AGI—an all-knowing synthesis machine.

Whether we aim for this depends on AI's potential to reconcile conflicts and act in a complex world. We may prefer to keep this uniquely human, expecting humans to be accountable for societal problem-solving. We might focus on developing superior left-brained AI, valuing accuracy but resisting its limitation on adaptability. We may prefer AI that enhances our right-brained skills, favoring adaptive social changes over optimization.

As language models advance, the gap between left-brained AI and right-brained approaches could be narrowing. Models like GPT-4 are becoming more human-like in responses which changes how humans react and use AI. Language bridges the AI-human divide, helping AI understand the intricacies of human thought and improving AI-human interactions.

Humans have evolved by transcending the vagaries and randomness of natural selection: by working together to define our values and behaviors, McGilchrist reminds us. This idea may guide machine design—rather than machines synthesizing outside human systems, we should create machines that enhance our holistic, social right-brained capacities, fostering collective wisdom.