From Lily Pads to AI: Grasping the Exponential Future

As discussions about AI has become a daily topic in our discourse, it lead us to question the trajectory of this acceleration. How do we perceive this pace of innovation, and what implications does it hold for the future? This article delves into the nuances of our expectations and intuitions regarding AI progress, challenging conventional linear thinking with the concept of exponential growth.

The AI race has?begun

The sound of the starting shot has been heard and the AI race has begun.

AI has been here for a while, but it’s just during the last few years it has really been getting mainstream attention, and that new advancements seems to be emerging almost every day.?

So in this race, what are we moving towards in such a pace?

Whenever I try to challenge my thinking in these kinds of topics, I always imagine myself climbing a ladder. I imagine my self at the bottom of the ladder observing the small concrete details in this case the technology fueling AI. Then I mentally climb to the top of the ladder, observing the abstract topography, but at the top of the ladder I also try to peer at the horizon. This mental exercis for me, is a helpful and intuitive way to look at certain topics. But there is one aspect that I find the most challenging with this exercise and perhaps the most interesting, it's to grasp the pace of innovation.?

So back to the question what are we racing towards? This has to enviably be AGI or Artificial General Intelligence, where machine intelligence vastly outperforms any human cognitive tasks, this is the view that we can se from the top of the imaginary ladder, the abstract view. That horizon can be quite intuitive, but a more interesting question that perhaps have a non intuitive answer is, when will we get there?

Linear intuition

Imagining time as a linear progression is the intuitive state, there was a day yesterday, and there vill be one tomorrow. Our lives are filled with cause and effect relationships and our minds are tuned to think of relationships as linear.?

So imagine a pond with a single Lilly pad, this Lilly pad has a special ability to grow exponentially doubling in size until it covers the entire pond. In this scenario it will take the lilly pad ten days to cover the entire pond. At what day will it cover half of the pond? If your answer is less then 9 then you're not considering the exponential factor in the growth rate.?

As new technological advancements fules the coming technological advancements each progression increase the pace creating an exponential growth rate.?

So if we stay in this exponential mindset, how do we apply it to reaching AGI? When we will reach high-level intelligence and AGI is a topic of an ongoing debate with various reasoning and estimates. varying from 5 years to a few decades.?

So for the sake of simplicity let's go with the assumption that we’ll reach AGI in 10 years. Lets take a look by breaking down the milestones getting there in 10 years.

Predicting the future and aligning expectations

Fortunately Google Deepmind have done the work for us there, staging the different levels of AGI, from no AI to superhuman or ASI (Artificial Superhuman Intelligence). In the image below you’ll see that the authors of the paper has set that we reached level 1: emerging AGI at 2023 with chatGPT and similar models, this is a huge leap.?

Creating our scenarion for our timeline plotting the progression through the different levels of performance, by setting the target for level 5, ten years from 2023, and by reading from the table below setting level 0, using Amazon Mechanical trunk released 2005 as our starting point.?

Link to Paper

So if we set level 0 at 2005, Level 1 at 2023 and Level 5 at 2033 and fit an exponential graph to to these points we get the graph above. Looking at the graph we see a slower takeoff and an increasing rate, taking four years between level 1 and 2 but less than two years between level 4 and 5.?

Of course predicting the advancements towards AGI is not this simple, and it's not the focus of this article to do an accurate forecast. Let’s back up again to the Lilly pad and our linear intuition and our expectations for progression, and this is where it gets tricky.

An exponential growth-curve may not necessarily mean a faster growth. If we reach AGI in ten year we’ll reach it in ten years, no matter if we plot it with a linear or exponential curve the time frame does not change, but our expectations may. Let's look at the graph below.

If we plot a straight linear line from level 1 (2023) ten years forward to 2033 at level 5 and set our expectations towards that linear line, we’ll see that we would expect technology to evolve at a more rapid pace, reaching each level between 1 and 5 sooner that the exponential curve. In scenario we would experience the progress as slow until it eventually catches up with our expectations.?

On the other hand, let's look and another opposite scenarion below.

If we plot a new linear curve setting our expectations beginning at level 0 in 2005 and plotting a straight line from 2005 to 2023 and on. If we would align our expectations to this line we would expect reaching AGI 72 years from 2023, and in this case we would be swept of our feet and our expectations shattered.?

Conclusion

Accurately predicting the future is extremely hard, and aligning your expectations and intuitions towards it, is even harder.?

Today it feels like AI is progressing at a rapid pace, and for many these last advancements came seemingly out of nowhere, suddenly everyone is talking about AI. This could be an indicator that an exponential growth is talking off, but if we set our linear expectation to this recent rapid pace (Fig. B) we might expect AI to advance much faster over the coming decade than what would be reflected in reality, making technology advancements appear slow before catching up. What impact would this have on the investments, beliefs and adoption of the technology?

And on the other hand, if our intuition is that AGI is still something far far away in the future, we might get blind sided by sudden advancements (Fig.C). A surprise that many of us experienced with the releases from OpenAI.

So how do one navigate and manage this? The simple answer is to expect to be surprised and the act accordingly.

Balance Short-Term and Long-Term Perspectives: Recognize that while AI advancements may appear slow in the short term, they can suddenly accelerate in the long term due to exponential growth. Balance short-term expectations with long-term vision, ensuring investments and strategies are flexible enough to accommodate rapid changes.

Embrace Uncertainty: Understand that accurately predicting the pace of AI advancement is challenging, and aligning expectations with the dynamic nature of exponential growth is even more so. Embrace the uncertainty and be prepared to adapt strategies accordingly.

Stay Agile and Adaptive: Expect to be surprised by sudden advancements and be ready to pivot strategies swiftly in response to technological breakthroughs. Keep an open ear of developments in AI and remain open to reevaluating assumptions and plans as new information emerges.

/ Filip?