In modern business processes, the adoption of generative AI has moved beyond the tool utilization stage and is now entering a phase of systematization.
Beyond this, a new era of intelligence called "Symphonic Intelligence" awaits.
This article explores the current state and future prospects of generative AI utilization from two perspectives: iteration work and flow work.
Iteration Work
In a previous article, I analyzed the perspectives of iteration work and tools, and flow work and systems, as viewpoints for enabling generative AI to perform tasks.
Iteration work refers to tasks that humans perform by semi-unconsciously combining multiple different concrete tasks and proceeding through trial and error.
And for this iteration work, tools are optimal. By selecting tools that fit various tasks, work can proceed efficiently. Therefore, it is necessary to prepare the required toolkit and become proficient in its use.
Currently, when generative AI is utilized in business, the vast majority of cases involve generative AI as a tool.
Most discussions about improving business efficiency with generative AI refer to adding this new and powerful tool to the existing toolkit used by humans for their iteration work.
Problems with Iteration Work
On the other hand, as pointed out in a previous article, the efficiency gains from tools in iteration work are relatively limited.
As tools become more efficient, humans ultimately become the bottleneck. We cannot ultimately overcome the limit of human working hours.
Furthermore, there is a significant gap in the efficiency and accuracy of iteration work between veteran employees and new hires, and bridging this gap is difficult. Consequently, even if one aims to double the workload next month, it cannot be handled without personnel possessing the skills of a veteran.
To resolve the problem of humans being the bottleneck, the ultimate solution would be to replace everything with artificial intelligence.
However, current generative AI does not yet possess that level of performance.
Moreover, even seemingly simple iteration tasks, when examined closely, consist of a large number of unconscious sub-tasks.
For this reason, these tasks could not be broken down into conventional IT systems or easy-to-follow manuals, and instead relied on human proficiency.
Unless these numerous unconscious tasks requiring proficiency are organized, and the necessary know-how for each is crystallized into knowledge, generative AI, no matter how much its performance improves, will not be able to perform tasks as a substitute for humans.
Flow Work Conversion and Systematization
To address the objectives of distributing tasks within the current performance limits of generative AI, and organizing unconscious tasks and crystallizing know-how into knowledge, it is highly significant to organize trial-and-error iteration work into standardized flow work.
Standardized flow work is well-suited not only for tools but also for systems.
Within flow work, there are tasks for generative AI to execute and tasks for humans to execute. By connecting these with a system, the entire flow work becomes executable.
Flow work conversion and systematization yield several significant benefits:
First, because generative AI is specialized for each individual task, optimizing its efficiency and accuracy for each task becomes clear.
Second, multiple workers can add knowledge to generative AI, and the benefits extend to everyone.
Third, it becomes easier to progressively shift the division of tasks within this work to generative AI.
By converting iteration work into flow work and accumulating the knowledge required by generative AI for each task as a system, intellectual work approaches automation, much like a factory production line.
And by incorporating the improving basic performance of generative AI, which evolves with the times, and utilizing the accumulated knowledge specialized for various tasks, it will become possible to make the entire flow work an automated process carried out by generative AI.
Virtual Intelligence
Up to this point, the analysis has been from the perspectives of iteration work and tools, and flow work and systems.
Another recent article further advances this discussion.
In that article, I touched upon the topic of orchestration by virtual intelligence.
Currently, and in the very near future, due to performance limitations, generative AI is more efficient and accurate when focused on specific tasks.
Therefore, as discussed earlier with flow work and systems, a mechanism that connects specialized generative AIs for each individual task was ideal.
However, even if the performance of generative AI significantly improves, processing tasks by switching roles and utilizing knowledge within a single process, rather than simply handling various tasks simultaneously, could potentially lead to higher efficiency and accuracy.
This approach eliminates the need for a system to link generative AIs together. Operations similar to system integration would occur within the generative AI itself.
Furthermore, it allows for flexible responses within the generative AI itself, moving away from situations where tasks cannot be swapped or added without modifying the system.
This means returning systematized flow work to iteration work.
However, this iteration work, having gone through systematization and flow work conversion, is now in a state where reusable knowledge can be formed, even if the number of generative AIs is increased or versions are changed.
This resolves the issues of human iteration work, enabling flexible work similar to that of humans.
Here, I refer to the ability of generative AI to switch roles and knowledge during a single execution as virtual intelligence. This is analogous to a computer's virtual machine.
Just as virtual machine technology simulates entirely different computers running on a single hardware, a single generative AI processes tasks by switching between multiple roles.
Current generative AI has already naturally acquired this virtual intelligence capability. For this reason, generative AI can simulate discussions among multiple people and generate novels featuring multiple characters.
If this virtual intelligence capability improves and sufficient knowledge is provided, it will become possible to perform iteration work.
Intelligence Orchestration
Furthermore, I refer to the ability to freely combine multiple roles and knowledge to perform tasks in this way as intelligence orchestration.
This is similar to orchestration technology that manages multiple virtual machines.
Just as orchestration technology efficiently operates systems by launching necessary virtual machines when needed, a generative AI with enhanced intelligence orchestration skills—a capability of virtual intelligence—will be able to perform iteration work flexibly, while appropriately managing numerous roles and knowledge, and maintaining efficiency and accuracy.
Symphonic Intelligence
Generative AI that reaches this stage can be called Symphonic Intelligence.
Just as an orchestra, skilled in playing each instrument, performs a single piece of music while fulfilling their respective roles, Symphonic Intelligence can play a symphony of intellectual tasks.
This Symphonic Intelligence is a new concept, representing one of the culminating points for generative AI.
However, Symphonic Intelligence itself already exists.
It is our human intelligence.
It is precisely because we possess Symphonic Intelligence that we can unconsciously and flexibly perform multiple complex intellectual tasks through iteration work, leveraging a wealth of know-how.
Finally: The Form of AGI
By providing generative AI capable of simulating Symphonic Intelligence with flow work and knowledge bases for other tasks, it will become capable of handling multiple iteration tasks.
Once it can handle numerous different iteration tasks, it will be able to grasp common rules among those tasks and structural patterns within the knowledge.
At that point, for entirely unknown iteration tasks, with just a brief explanation from a human, the AI will be able to learn the know-how of that task simply by observing how a human performs it.
This is true Symphonic Intelligence. Once this stage is reached, humans will no longer need to expend effort on converting work into flow processes or crystallizing know-how into knowledge.
Furthermore, the knowledge thus automatically accumulated by generative AI can be shared among other generative AIs.
If this happens, the learning capability of generative AI will far surpass that of humans.
This can be considered one form of AGI.