In discussions, transformative AI is often seen as a certain point in time when an AI system achieves a certain level of performance or capabilities. For example, some people believe(d) that transformative AI will occur when an AI system can pass the Turing test, or when an AI system can create art or music that is indistinguishable from human-created art or music (yep, a few years ago people said that). More modern definitions have a higher bar (e.g. do all human jobs at low cost.)

However, it is also possible to see transformative AI as a process rather than a specific point in time. In this view, transformative AI is not a single event, but rather a gradual evolution of AI systems that become increasingly capable and sophisticated over time. This process of evolution is driven by advances in AI research, as well as by the increasing availability of data and computing power.

The view of transformative AI as a process has several advantages over the view of transformative AI as a specific point in time. First, it allows us to see that transformative AI is already happening, and that it is not something that will happen in the distant future. Second, it allows us to better understand the factors that are driving transformative AI, and to develop strategies for accelerating this process.

Finally, it allows us to better prepare for the challenges and opportunities that transformative AI will bring. As transformative AI systems become increasingly capable, they will have a profound impact on our society, economy, and environment. It is important that we start thinking about these impacts now, so that we can shape the future of transformative AI in a way that benefits all of humanity.

The long term impact of relatively small changes

The exact start and finish of a transformation can only be determined in hindsight. However, we are currently in a period of accelerated change, driven by a number of factors, including the rise of artificial intelligence, the increasing availability of data, and the development of new technologies.

If this rate of change were to persist for 5 to 10 years, we would see our economic, societal, and environmental situation changed dramatically. A transformation might not need as much growth as you might think. A short-term productivity growth change of 2% per year to 4% per year might not seem like much in the short run, but over a period of 5 to 10 years, the changes would be significant.

Here are some of the potential impacts of an accelerated rate of change:

• Economic growth: Increased productivity could lead to significant economic growth, as businesses are able to produce more goods and services with the same amount of resources. This could lead to higher wages, lower prices, and a higher standard of living for everyone.
• Job creation/losses: New technologies could create new jobs, as businesses need to hire people to develop, implement, and maintain new systems. However, new technologies could also lead to job losses, as some jobs become obsolete. It is important to invest in education and training so that workers can adapt to the changing job market.
• Social changes: New technologies could lead to changes in the way we live and work. For example information and impact on elections might change democracy.
• Environmental impact: New technologies could have a positive or negative impact on the environment. For example, new energy technologies could help us to reduce our reliance on fossil fuels and mitigate climate change. However, increased production could also lead to increased pollution and environmental degradation. 

A possible timeline

For a transformation, we do not need a very advanced AI. We just need an AI with sufficient overhang to have enough low hanging fruit to accelerate the rate of change. The AI has to keep up to be able to generate new use cases to implement. What is required is AI significantly advanced enough to keep pushing the advancements. Look at the possible timeline below. Nowhere does it require any remarkable innovations. 

• 2023: The introduction of GPT-4, great progress in open source models, significant algorithmic progress, and building the infrastructure to support AI like plug-ins and copilots. This would be a year of significant change, as AI systems become more capable and sophisticated. We would see the widespread adoption of AI in a variety of industries, including healthcare, finance, and manufacturing.
• 2024: An update on the foundational models, wide proliferation of human-in-the-loop models that generate a lot of data for further development, and progress in synthetic data. This would be a year of continued growth for AI, as we see the development of new and more powerful AI systems. We would also see the development of new applications for AI, as businesses and organizations find new ways to use AI to improve their operations.
• 2025: Worldwide production of cognitive work is greatly increased by this time, part by automation and part by freeing up cognitive workers to make better use of their human cognitive abilities. This would be a year of significant transformation, as we see the impact of AI on the workforce. We would see the displacement of some jobs previously hard to automate. But because learning from the human in the loop systems progressed to automate further and further. 
• 2026: And so on. The pace of change is likely to accelerate in the years to come, as AI technology continues to develop. We can expect to see even more significant impacts of AI on our lives and businesses in the years to come.

The AI will grow into the transformation

Although significant changes are taking place in the field of AI, it is not necessary to have a very advanced AI to keep pushing the advancements. What is required is an AI that is significantly advanced enough to learn from its mistakes and to improve its performance over time.

• Collaborative AI: Collaborative AI can be used to bring together people and AI systems to work together on problems. The AI will will in this process and the human part will become smaller. 
• Self-learning AI: Self-learning AI can be used to improve the performance of AI systems by allowing them to learn from their own experiences. This can help to reduce the need for human intervention in the development and maintenance of AI systems.
• Infrastructure : Once a proper AI infrastructure is in place, progress becomes easier. When people use Copilots, chatbots, plugins etc. it is easy to implement the next foundational model. The impact of gpt3.5 to gpt 4 is limited because the limited amount of users and not a lot of software was built upon gpt3.5. When we have software built upon gpt4 is would not be hard to switch the underlying foundational model. (e.g. you could easlily run Autogpt with gpt5 instead of gpt4)
• The unknowns : A lot is hard to predict, think of going from CD’s to Spotify, the impact of social media etc. The creative solutions can help to create new and innovative solutions that would not be possible without AI. Changes like this can help to break down barriers and to open up new possibilities.

Conclusion

Fixing on a specific date for when AI will surpass human intelligence or replace all human jobs is not very useful. The pace of technological development is constantly accelerating, so it is impossible to predict exactly when AI will reach a certain level of advancement.

When the last few thousand humans lose their jobs because AI finally is advanced enough to do every human task, that will be at the end of a process more than that being as specific point in time. It is unlikely that there will be a single point in time when AI will be able to do every human task. Instead, it will be a gradual process as AI systems become more and more capable.  

Being in the process it is easy to miss the transformation. If this rate of change continues the transformation has already started. It is possible another AI winter will come. But as it is now, there is enough new technology to deploy. We are probably close to the maximum adoption rate for new technologies (human have to learn/decide etc.). As I see it technological progress will keep outpacing human ability to implement. Although I think deployment speeds in cognitive processes can be very rapid.  

After we implemented the all the use cases of gpt4, enough advancements will have been made to make the next step. To summarize, I think gpt 3.5 (chatgpt) was the transformative AI. It kicked off the period of an increased rate of change. Now we need of enough innovation to keep the progress going. To me that seems very likely.