Deeply sought

As we have established in the past, Murphy’s Law — that which can go wrong, will go wrong — is a canon of life. You may dismiss it as a mock law, but wherever there is human involvement, the worst seems to have the most substantial chance of occurring. As they say, too many cooks spoil the broth.

If you are a person with a good imagination and an optimistic outlook, what kind of future do you imagine, especially in relation to AI? Some science fiction writers of moderate disposition view it as an important upgrade to our lives. For instance, Jarvis from Iron Man is always a loyal servant, eager to assist his master. Similarly, the seamless integration of new technologies (AI and robotics) into human life in Asimov’s Bicentennial Man can make one eager for the future. In sci-fi series like The Orville and Star Trek, devices like food synthesisers or replicators were instrumental in ending world hunger. We could have long discussions about the potential of these imagined devices alone, and numerous papers could be written. However, while flights of fancy can take us to remarkable places, it is human nature, political constructs, and international identitarian fault lines that are bound to define the pace and direction of material progress. Consider this: when Star Trek and The Orville envision these advancements, they portray humanity united under one leadership, not divided into small violent entities in the name of nationhood.

For a long time, mankind looked forward to the advent of AI and robotics merely as force multipliers for human capacity. Then Hollywood took over. Since then, with a few notable exceptions, Hollywood has tried to frighten us with the possible rise of an authoritarian and dark superintelligence. The Terminator franchise and the Matrix trilogy did what they could to further strengthen innate human paranoia about the rise of machines. It hasn’t stopped yet.

But now that the AI world is coming to life, machines have opened their eyes (ours, too) and are learning to take baby steps. We know that Hollywood-style supervillain intelligence isn’t an imminent threat. Recently, when there was significant hubbub about AI autonomous agents, OpenAI’s co-founder and former chief scientist, Ilya Sutsveker, gave a talk on the future of agentic AI. The crux of the talk was a warning about not getting too excited about such developments because it will take time and better resources for truly autonomous AI to become a reality. Artificial intelligence is such a humongous and transformative fork in the road that we do not have any historical parallel to compare. Let me qualify.

In the past, from the domestication of plants and animals in the agricultural revolution to industrialisation, the advent of computers, the internet, and the IT revolution, nothing has sought to supplant human intelligence. They served in aid of our intelligence, not in its place. Now, you already have machines that churn out texts longer than and occasionally (when they try very hard) nearly as insightful as this one within seconds. They can potentially do everyone’s job better. Also, thanks to advances in robotics, they may also have more customisable, hence more useful bodies than ours, with fewer limitations. So, as I have told you a hundred times, tech displacement, where machines invariably take everyone’s job over a short period, is a real and imminent danger. Yet, if the Covid crisis has taught us anything, it’s that employers will not easily let go of human labour – that is, if humanity can work in tandem and closely to avoid any unforeseen eventuality. But then, remember the start of this piece where we discussed Murphy’s Law. What was about to happen was not entirely unpredictable, though.

Kai-Fu Lee’s brilliant book AI Superpowers foresaw this coming. It is a book, among many other things, about AI competition between the US and China. The author, an Asian AI pioneer, believes both China and the US have advantages in data training. The US has a significant advantage in business and industrial data thanks to its long-established industries and multinational corporations that operate at immense scales. US companies like Google, Amazon, and Microsoft have vast consumer data. Still, their strength is more pronounced in high-quality, structured business data integral to developing AI applications in finance, logistics, and management.

On the other hand, China has a considerable edge in consumer data. Due to its vast population and high rate of digital engagement, Chinese companies collect immense volumes of data. This is why successive governments in America (Trump I and Biden administrations) have imposed export bans on China to block its access to high-tech chips like Nvidia’s GPUs.

Right when President Trump took over and happily announced the launch of ‘Project Stargate’, a 500 billion dollar investment in building AI infrastructure like data centres and power hubs, China gave a wake-up call. So far, the visible competition was between American AI companies, which had no financial or technological supply shortage. Then, they stuck to legacy architecture and high-investment models. But then, a Chinese startup named DeepSeek has rattled the market.

Initially, DeepSeek aimed to collect 10,000 NVIDIA A100 GPUs. However, due to US export controls limiting access to the latest NVIDIA chips, DeepSeek has adapted by using NVIDIA H800 GPUs for recent developments. These GPUs are specifically designed with reduced capabilities for the Chinese market but have proven effective in supporting DeepSeek’s advanced models under restrictive conditions. Its logic model was recently made available in the Apple Store and other such markets, and in no time, it beat American chatbots to the number one position. This news caused Nvidia and other AI-related tech companies to crash. Nvidia lost nearly 600 billion dollars worth of value. The cumulative loss to the tech industry was to the tune of one trillion dollars. The markets have recovered since then, but the industry’s confidence has not.

The DeepSeek model is programmed to ration high-end resources to offset the limitations. It is a 27 million dollar startup and has reportedly used only 6 million dollars to train its AI, which is negligible compared to what is available to US companies.

What does it teach us? The cost of innovation is collapsing. But so is the argument in favour of regulating this cutting-edge technology. What happens now? Competition will make this technology smarter and cheaper for companies. This, in turn, will make the temptation to replace human labour with AI more irresistible. Sanity today is deeply sought. May the heavens have mercy on us!