AI: Overhyped or Hyper-accelerated?

The publication of my last article,?The Age of Fragility,?was the catalyst for an interesting online discussion with a friend and influencer in the supply chain community (whose opinion I value greatly) regarding whether we are currently at the midpoint of the sixth wave’s upswing or its beginning.

The crux of my friend's argument was that he believed that the Ukraine War represented the transition to a new wave. In contrast, my research into long waves indicates that the upswing began around 2010 and that the conflict perfectly represents an example of the chaos, disruption and unrest that emerges during the upswing’s midpoint.

Why I think we are in the Transition Period

As I detail in Transition Point, each technological wave has different phases and rates of change, with the upswing constituting around 65% of the wave's total period and the downswing just 35%. Technologies diffuse gradually, then suddenly, similar to Hemingway’s description of how companies go bankrupt.[i]?

The gradual buildup started at the end of the fifth wave's winter period and the start of the transition to the sixth wave. At that time, the market had become over-inflated due to the move from production capital (the real economy) to financial capital (e.g., subprime mortgages and other risky investment vehicles).? The next five years became about survival, as limited economic growth and rising debts called for austerity measures.

Phase One: Excitement Everywhere.

The green shoots of spring appeared around 2012 with the announcement of significant investments in exciting new technologies, ranging from warehouse logistics (Kiva / Amazon), Google autonomous driving (Google), co-bots (Rethink), drones (Amazon), Blockchain (IBM), etc. This period also saw the thawing of the famous 'AI winter' as IBM Watson beat human Jeopardy champions in 2011, and AI startups such as Deepmind (founded in 2010) were acquired by Google in 2014. AlphaGo then beat Go Grandmaster Lee Sedol in 2016, achieving a feat eighty years earlier than it had been predicted back in 1997 following Garry Kasparov’s defeat to Big Blue.

The period from 2012 to 2020 was one of great excitement for these technologies and their potential, attracting significant investment. Cryptocurrencies went exponential, promising new investment opportunities outside the big banks and stock market, and massive amounts of money were invested in them. Tech companies competed to snap up as much talent as they could straight from university—often recruiting not just recent graduates but also sitting students and their professors.

However, true to form, while this initial period provided a rich tapestry of viral videos, at the corporate coalface, little changed. This represents classic phase one behaviour, strengthening my argument that we were already in the sixth wave's upswing back then. Then, in 2020, COVID hit, accelerating investment in any new technology related to eCommerce, home delivery, or remote working. While exciting, most of these developments were either platforms that connected buyers and sellers or incremental advancements in existing technologies and ways of working.

Then, the bubble burst.

The Transition Period: Nightmare Fuel.

As the world emerged from the lockdowns, we went straight into the upswing’s disruptive transition phase. A series of black swan events rocked supply chains, billions were wiped off crypto, and tech companies started laying off everyone. No one is posting TikTok videos about all the crazy free benefits they get working from Meta / Google / Apple / Twitter anymore - mainly because they no longer work there.

These transition periods are also marked by geo-political issues and conflicts, and true to form, Russia invaded Ukraine in 2022, the October 7th massacre happened in 2023, and this year, Iran and Hezbollah attacked Israel. All these conflicts are yet to reach their peak. While the pharmaceutical companies made hay during COVID, military suppliers such as Lockheed Martin, General Dynamics, and BAE Systems are now posting record market valuations and stock prices. The US elections in November will be a car crash, resulting in delayed results and significant social unrest regardless of who wins. There is no guarantee how long this period of chaos and disruption will last, but we must get through it to move into the second phase of the upswing,

My friend and I disagreed on whether this period was the midpoint of the sixth wave or the start. While he felt that we are now living through a period of peak AI hype, which will soon collapse into the trough of disillusionment, conversely, I warn of exponential progression and massive industrial and labour force disruptions. While I agree that we haven't yet seen the true potential of AI, the pace of change is such that I don't think there will be a dip. Disillusionment is not the emotion that awaits us; fear is. ?

Here's why.

Phase Two: Welcome to the Exponential

The upswing's second phase—the summer period—is when we should see a marked return of investment into the real world of industry. As money moves back to production capital, the potential from the inventions in the first phase becomes industrialised, transforming everything from business practices to everyday life.

I've talked and written about disruptive technologies and their corporate and industrial potential for over 12 years. When I first stood on stage and discussed the implications that a convergence of these technologies would create, I was the only person who seemed to be speaking about this. Now, nearly every business keynote or article has an AI angle. But despite the massive shift in knowledge and focus on the subject, I am still shocked at the speed of advancement we see in artificial intelligence and how few people are paying attention. And it leaves me worried.

Only four months ago, in May 2024, OpenAI released GPT4o, their flagship multi-modal AI model capable of reasoning across audio, vision, and text in real time. This effectively gave the AI system ears, a voice and sight. Unlike the ancient (May 2023) GPT4, GPT4o responded to audio inputs in 320 milliseconds on average, about the same as human conversational response time, allowing the AI to be interrupted mid-instruction without losing track of the conversation.

Then, on September 12th, OpenAI announced the launch of OpenAI o1, a new large language model trained with reinforcement learning to perform complex reasoning. OpenAI claims that o1 thinks before it answers, allowing it to produce a long internal chain of thought before responding to the user.

The posted results of this new LLM have been astonishing. In a series of tweets last week, Noam Brown, a research scientist at OpenAI, implied that OpenAI had utilised a new optimisation algorithm and training dataset containing "reasoning data" and scientific literature specifically tailored for reasoning tasks. "The longer [o1] thinks, the better it does," he said.[ii]

In a qualifying exam for a high school math competition (the International Mathematical Olympiad (IMO)), o1 correctly solved 83.3% of problems while GPT-4o only solved 13.4%. o1 also trounced GPT4o in code writing and PhD level Science. It also beat human PhD experts in their specialist subject matter.[iii]

OpenAI also provided more detailed o1 test results, showing exactly how much better this new LLM performed than gpt4o. It showed significant performance improvements in all areas tested except English language, but it also showed enormous advancements in math, physics, chemistry, and logic.

OpenAI is also not the only player in town, and Anthropic has demonstrated exponential development levels in its Claude GenAI LLM family.?Claude 1 was released in March 2023, Claude 2 in June 2023, Claude 3 in March 2024 and Claude 3.5 in June 2024, with each version significantly more powerful and capable than the previous.

Claude 3 made headlines when it launched because it was the first AI system to record a IQ of 100, the average human IQ level in the Western world. After multiple tests, it recorded a average of 90 IQ points, still highly impressive. However, OpenAI’s new o1 model was tested last week using the Norway Mensa IQ test, scoring 120. This places it 20 points higher than the average human score, a massive increase in an incredibly small timeframe. There are only 50 IQ points between average and genius - and AI has just jumped 20 points in four months.

The tester was journalist and technology commentator Maxim Lott, who put the various leading Gen AI solutions through the Norway and other IQ tests, with the results as follows:

Maxim declared that he believes we will see an AI pass 140 IQ points by 2026, though I imagine we will reach this sooner at the current rate of change. Ark Invest, perhaps the most famous of hedge funds that focus on new and disruptive technologies, believe that original predictions of when we will achieve Artificial General Intelligence (AGI) in the next decade are far too conservative given the current rate of development and that we will now reach this stage by 2027.

One interesting comment from Maxim after putting these LLMs through the IQ tests was that he was adamant that this clarifies that AIs?are intelligent and reasoning.

States Maxim, “AIs aren’t merely regurgitating words pulled out of an algorithm. Yes, they are fundamentally doing that — but predicting the next word gets so complex that logic and reasoning seem to arise out of the process of prediction. Is that maybe also the same process from which human higher-order intelligence?originated from?” [iv]

Chip Wars

The advancements in LLMs and their capabilities are also being driven by massive advancements in silicon chip power, with Nvidia announcing the launch of its Blackwell B200, loaded with over 200 billion?transistors?and optimised for both AI training and inference tasks. Nvidia said the Blackwell processors can power much larger language models than are deployed today — up to 27 trillion parameters, many times the size needed to support the leading LLMs such as GPT-4.?

But that was five months ago, in March.

Since then, gpt4o and now o1 have been launched, which require more processing capabilities. Nvidia also has a new competitor in the chip market - Cerebras Systems. This California-based startup launched Cerebras Inference last week, a new wafer design with approximately 4 trillion transistors and 44GB of on-chip SRAM. This design eliminates the need for external memory, thus removing the memory bandwidth bottleneck that hampers traditional chip architectures. Cerebras focuses on creating the largest and most powerful chip that can store and process enormous AI models directly on the wafer, dramatically reducing the latency in AI computations. Cerebras chip's performance in AI inference is groundbreaking, processing inputs at speeds reportedly 20 times faster than Nvidia's solutions. This is due to the direct integration of memory and processing power, which enables faster data retrieval and processing without the inter-chip data transfer delays.?

The tiger in the tank of any innovation is investment, and the financial sluice gates have now been opened. Money is flying into the AI space—especially in the US, where it is outspending in every area AI touches, from sensors to security.

So, we have more extensive and intelligent AI models powered by increasingly powerful processors, both of which are advancing at an exponential rate.

The Automation of Labour

The final area of exponential development is robotics, specifically humanoid robotics.

In April this year, OpenAI announced a partnership with robotics manufacturer Figure and demonstrated the outputs of this collaboration by merging the Figure 01 robot with OpenAI’s gpt4 generative AI capabilities. The results were incredible – and somewhat unsettling. The robot understood verbal instructions and demonstrated advanced independent decision-making based on the contextual situation in which it found itself. When asked to provide a human with something to eat, it could identify which object was edible (in this case, an apple) from the variety of objects in front of it. When asked to explain its actions, Figure 01 complied, detailing the logic used.

Video: Figure 01 and OpenaI?

This mix of robotics and general intelligence led many to believe they had reached the point of AGI. However, one noticeable issue in the Figure 01 video was the time lag between being asked a question and the verbal or physical response. The release of gpt4o only two months later created a generative AI system that massively reduced the response latency to human conversational levels. Then, last month, Figure released Figure 02, a significantly upgraded version of its robot, showing it working at BMW's production line plant at Spartanburg, in South Carolina.?BMW reported on the trial here, including a link to the video showing them in action. A successful trial would prove the catalyst for other companies to invest in these robots, for unlike their existing automotive robotic setups, these are general purpose and mobile, able to go to where the need is and use their dexterity and intelligence to handle numerous requirements and challenges. Plus, they work 24/7 for no pay and don't need breaks, holidays, well-being checks, etc.

Other robotics companies are also racing ahead in their developments, ranging from the new version of Boston Dynamic's Atlas robot to Tesla's Optimus.

True to the nature of the long waves and innovation diffusion, our exposure to the results of all these new technological innovations has been limited to date. We often hear about AI but rarely see it, absent access to tools like ChatGPT and a prompt on LinkedIn asking whether we want AI to improve our text. But even then, its impact on our work and home lives is minimal.

This will change. Quickly.

As per Amara's Law, we have fallen into the trap of overestimating the impact in the short term and underestimating the long-term implications. This Law accurately reflects the diffusion of all significant innovations, ranging from automobiles to AI. In fact, the adoption curve of AI and humanoid robotics is incredibly similar to that of the car. While automobiles became commercially available in 1893, they didn't reach one million in sales until seventeen years later, in 1910. The first commercial humanoid sale was just over a century later, in 2013, and if it follows a similar path, by 2030, we should see sales of one million humanoid robots.

Given that we are seeing significant upgrades in LLM capabilities, chip processing power, and robotic hardware every three months, by 2030, the humanoid robots on sale will likely have been upgraded 20 times from the versions we see today. Given what we see in the progress of GenAI, AI chips and robotic hardware capabilities, those upgrades will not be minor. Therefore, the potential capabilities this machine workforce could rapidly acquire are daunting. Because let's face it, we puny humans don't experience exponential increases in our intelligence, agility and processing power every quarter. The impact of this is obvious. Just as the automobile removed the need for horses and buggies, AI and AI-powered robotics will remove the need for everything from cheap migrant labour to researchers, receptionists and content writers.

The speed of these technological advancements is beyond anything we have seen to date. It took tens of thousands of years to move from the Stone Age to the Bronze, thousands to move from Bronze to Iron, and a millennium to progress to the Industrial Age. Now, giant technological leaps occur in just months. So while we accelerate towards a world of science-fiction hardware and software, our human thoughtware remains almost identical to our bronze-age relatives who only had to adapt to generational-pace change. So, while we have always been able to adapt to the rate of change, keeping pace is becoming increasingly challenging. And things will never again be as slow as they are now.

So, it is increasingly likely that our time at the top of the intelligence hierarchy is fast coming to an end, especially given the reversal of the Flynn effect and the decline in human IQ. While machines are getting smarter, the general population is getting dumber, with a shortened attention span and an increasing inability to be still and think deeply. The addiction to technological dopamine hits means that the second they are alone and their minds unstimulated, they reach for their phones rather than allow themselves a moment to relax.

So yes, we are at the transition point at the midpoint of the upswing. Yes, things will get worse before they get better. And yes, once we cross over to the next phase, AI, humanoid robotics, autonomous vehicles, and other disruptive innovations will move from inspired inventions to industrialised innovations. Once the business case becomes evident and the early adopters start to gain competitive advantage, the chasm will be crossed, and the world will never be the same again.

The question that vexes me most is that while we will marvel at announcements of rapid technological advancements, how will we be able to keep up when these innovations become industrialised during the sixth wave's next phase, and Amara's Law kicks in? Unlike these new technologies who advance exponentially, our intelligence, physical capabilities and ability to understand and adapt to change all develop linearly.

As I asked in Transition Point, while we will be surrounded by the most amazing technology, will we be amazed - or dismayed?

Basic income bread and virtual circuses, anyone?

[i] 1954 (1926 Copyright), The Sun Also Rises by Ernest Hemingway, Book II, Chapter 13, Quote Page 136, Charles Scribner’s Sons, New York.?

[ii] https://techcrunch.com/2024/09/12/openai-unveils-a-model-that-can-fact-check-itself/

[iii] https://openai.com/index/learning-to-reason-with-llms/

[iv] https://www.maximumtruth.org/p/massive-breakthrough-in-ai-intelligence