Is human-equivalent AI closer than we think?
AI takes the Spotlight 🌟
During our keynote at the Artificial Intelligence Summit of Nov 2023, we identified early on that AI has hit the tipping point and entered the mainstream market, a prediction recently confirmed by NVIDIA's CEO, Jensen Huang.
AI has taken the spotlight following ChatGPT's successful debut in late 2022! This has sparked a frenzy of development, with millions using the new technology and thousands of new AI tools being launched every year.
This is a major shift from previous years when more advanced AI tools were mainly used by an “elite” of AI experts, enthusiasts, and visionaries but were unavailable to most of the market. Today, everyone is talking about AI, even more than other pressing issues like the economy or global conflicts. With every new application, AI becomes more intertwined with the fabric of our lives, and its popularity and importance increase.
The elephant in the room 🐘
But let’s address the elephant in the room. The goal of AI has changed a lot since its beginning in the middle of the 20th century, when the aim was to match human intelligence, or, in other words, to achieve Artificial General Intelligence.
Artificial General Intelligence (AGI) is AI with reasoning that can adequately deal with all sorts of tasks and problems that a typical person would do.
Initial optimism soon gave way to skepticism, as the ambitious expectations for AGI were not realized. AI quickly lost popularity and entered a limbo state, awaiting a new direction. This came in the 1990s with the rise of Machine Learning. Professionals and scientists started using AI models and statistical methods in specific areas, like neural networks in computer vision. The concept of AI shifted from creating AGI to a more practical approach known as Narrow AI.
Narrow AI involves using AI for more down-to-earth and specific tasks, such as Conversational AI for chatting or Predictive AI for price optimization and forecasting.
This focus allowed for the development of highly specialized AI systems, which turned out to be very successful. Everything out there is still Narrow AI of different levels of sophistication. Some, like Large Language Models (LLMs), have glimpses of AGI, and Google DeepMind characterizes them as emerging AGI.
The bottom line is that AGI is currently out of reach, Narrow AI dominates the market, but expectations are rising quickly, along with one more thing.
Giving AI flesh and blood 🤖
As Sam Altman, OpenAI’s CEO, recently said, the new oil is not data anymore but compute! AI is instrumental here, but you must blend software with hardware and devices to succeed. This blending can take numerous forms, such as:
➡️Processing
Software running on hardware performing complex data processing. NVIDIA is leading the hardware part with their GPUs, we all know about the 7 trillion chips initiative from OpenAI, and others like Google and Intel are weighing into the race.
➡️User interface
Using advanced ER/AR/VR devices for intelligent human-machine interaction. Apple Vision Pro with Spatial Computing is the best example so far, but others like Meta and Microsoft have their initiatives.
➡️IoT
Giving AI “eyes & ears” and the means to sense our world. This has been an emerging market for some years, but the key here is the interaction with more sophisticated AI systems like LLMs.
➡️Robotics
Giving AI "flesh & blood" and the power to execute in the real world, not using electronic pathways. Tesla, NVIDIA, and OpenAI are already involved in such initiatives.
In simple terms, AI will gradually become more human-like, interactive, and naturally integrated into our world.
Is AGI just around the corner? 🤔
So, apart from the numerous applications of Narrow AI today, should we also expect AGI to appear any time soon?
Well, if you ask Jensen Huang, it’s five years out. On the other hand, Elon Musk argues that this could be even closer. Sam Altman agrees that AGI is not far away based on the current rate of AI development. We hear these claims from well-respected leaders of AI-related companies, and that's exciting! But we need to question if we all have the same definition of AGI, which is probably not the case. Just watch the linked videos, and you'll see what I mean! Also, let's not forget that AI leaders are developing, selling, and promoting AI technology, so it's to their benefit to inflate expectations. And I am not saying they do, but we must keep an open eye here.
For the time being, evidence supports the opposite. As said, all AI tools are Narrow AI, not AGI. Ask ChatGPT, and it will tell you so!
Even the most advanced LLMs, like ChatGPT 4, are mere probabilistic models with no reasoning or real understanding of what they are talking about.
So, is AGI just around the corner? Well, if you ask ChatGPT, in all its “wisdom”, it disagrees and thinks these claims are somewhat optimistic😜.
However, one could argue (Elon Musk does) that given the rapid rate of AI development, all existing tools will soon become obsolete, and other more advanced tools will appear closer to or on par with AGI. Admittedly, AI has achieved wonders in our lifetime and will surely evolve in the following years. But technology adoption usually comes in waves, and its growth follows a bell curve that evolves exponentially initially but then declines. We are still at the tipping point, and although we have a long way to go, the growth frenzy will end at some point.
But the most important thing is something else. Which elements of Human Intelligence will we be able to capture with AGI?
The Essence of Human Intelligence💡
If we refer to straightforward tasks, even complex ones, based on existing knowledge, AI could eventually match or exceed Human Intelligence.
For instance, LLMs are continuously getting better at solving examination tests, even the most difficult ones. However, these tests include questions that have been answered before and well-known concepts, perhaps not to the person taking the test but to someone else. LLMs leveraging tons of data can find clever pathways to retrieve the right information. But this is not reasoning; it is more like a highly sophisticated search. The more complex the problem, the more “intelligent” AI looks when retrieving the information. Although this could qualify as partial intelligence (it's definitely better than animals and trees😊), it is still not reasoning and definitely not creative.
Reasoning, creativity, and our ability to navigate with minimal or no data truly define our species and intelligence. Against all odds, humans can find their way into uncharted territories.
This is the essence of our civilization, as most discoveries and achievements that propelled humanity to where it is today were based on unconventional actions with limited or no information.
Coming back to AI, LLMs, the most advanced type of AI today, overfit as they rely too much on available information. For those not exposed to data science concepts, overfitting means that the AI model performs pretty well in what it knows (the data used to train the model) and poorly in what it doesn't know.
LLMs, the closest to AGI today, fail to extrapolate and anticipate the unknown. They will need to master this human skill to become true AGI, and it's unlikely this will happen anytime soon.
So, what’s next? ⏭️
So, is AGI coming or not? No one can be 100% certain, and of course, we don’t know what’s happening in every AI company’s lab. However, at this point, achieving true AGI seems like a bold claim, even coming from top-notch AI leaders or experts.
One thing is certain: AI is here to stay, and it will continue to evolve rapidly in the near future. Even if it does not reach the level of AGI, it will become increasingly human-like and seamlessly integrated into our daily lives. The more advanced AI becomes, the greater the benefits but also the challenges, whether practical, ethical, or even existential, especially as we approach AGI.
Bottom line? While we should be mindful of potential threats and challenges, it's vital we embrace now the transformational power of AI technology and its incredible benefits!
Interested in learning more about AI-Powered Price Optimization and Strategic Forecasting?
Comments