What ChatGPT really tells us about the future economy.
News of ChatGPT’s capabilities has captured the public imagination. Our algorithmically determined newsfeeds are full of articles about the help or hindrance that Artificial Intelligence (AI) poses to society. But while journalistic bandwidth is largely occupied with questions such as ‘how do we stop students from using AI to cheat?’, ‘how do we use it to make more money?’, and the ever-popular ‘what is going to happen to our jobs?’, the versatility and performance of foundation models like ChatGPT have far deeper implications for the future of the economy.
The first thing to understand about platforms like ChatGPT is how they differ from other types of AI. The term ‘foundation model’ comes from the widely read Stanford University Institute for Human-Centered Artificial Intelligence (HAI) 2021 report titled ‘On the Opportunities and Risks of Foundation Models’ and refers to models that are trained on broad data, usually at immense scale in a self-supervised way that can be quickly adapted to a wide range of downstream tasks.
But although the term is new(ish), foundation models are not actually a new concept. Self-supervised AI models have been released before, such as OpenAI’s GPT3 in 2020 and Google’s BERT in 2019. Instead, the term is significant because of the specific characteristics of these large-scale AIs. While terms like ‘unsupervised models’ or ‘general artificial intelligence’ emphasise the way these models process data or their versatility, a ‘foundation model’ highlights the incomplete nature of these systems and their role as the bedrock (i.e., foundation) of other AI solutions. The term recognises that the widespread belief that adding parameters to models would reach a point of diminishing returns was wrong.1 Instead, developments suggest that these models improve exponentially as the amount of data increases, with AI products built on pre-trained foundation models likely to perform better and be more versatile than supervised alternatives.
The second thing to know about foundation models is the significant amounts of time, money, IP, computing power, and energy it costs to build them. They require tens of millions (or even billions) of dollars and represent a significant investment – even to the biggest of big tech. On the other side of the equation, start-ups and smaller companies rushing to put a viable AI solution on the market will find they can do so much cheaper, much faster, and with much less data if they use a pre-trained foundation model, rather than starting from scratch. If using the pre-trained model also results in a product that performs better and is more versatile than the competition, it becomes clear what these AI companies will do. This gives us an important glimpse into what the future of the industry might look like.
The future of AI is going to affect all other industries because digital transformation is only going to increase as a key driver of productivity across the board. As Microsoft CEO Satya Nadella said in his recent Wall Street Journal interview,2 ‘…tech supply in what is considered the tech industry is now basically getting more distributed because every auto company, every pharma company, (and) every retail company needs to hire tech people’. And if data is the new oil that drives the digital economy, artificial intelligence represents the oil refineries that are needed to generate new insight and value from data to support emerging and innovative applications.
Given the critical place of AI in our shared economic future, APAC policymakers need to carefully develop their national strategies. In the words of Alexandr Wang, the youngest self-made billionaire and founder of Scale AI, ‘One thing that is very obvious to me is that… which countries have access to AI technologies and how they use them are going to define how the world plays out over the course of the next few decades’.3 And here we come to the title of this article. Given the concentration of the best ‘oil refineries’ in a few places, will Kishore Mahbubani’s prophecy of Asian economic leadership come to pass?
Policymakers in APAC know the importance of AI, and it is hard to find an Asian country that does not have its own National AI Strategy. But just like the various articles on ChatGPT, many of these strategies focus on the deployment of AI applications, the commercialisation of AI solutions, and the creation of jobs. With the rise of the foundation model paradigm, it is worth reviewing these strategies. Setting aside concerns about foundation models manifesting the inherent biases of the data sets used, policymakers need to consider the relative advantages and disadvantages of using foundation models to enhance and accelerate AI deployment. As Professor Yutaka Matsuo from Tokyo University told the Nihon Keizai newspaper last year,4 if only a handful of big companies have a monopoly on this kind of AI, the social impact would probably be enormous. There are moves to make general-purpose AI open so that anyone can use it, but the future is uncertain. In Matsuo’s words: ‘We need to think hard about what kind of strategy (Japan) will come up with to achieve the final form of AI’.
As the world’s leading tech policy consultancy, Access Partnership operates at the heart of debates around AI and ChatGPT, rigorously assessing their various implications across sectors and markets. We work with governments, regulators, and industry to ensure the widest, most equitable access to new technologies.
Our latest report, ‘Impact Assessments: Supporting AI Accountability & Trust’, helps to guide this conversation by providing an overview of existing regulatory processes and recommendations on how to build cross-regional harmonisation. Developed in partnership with Workday, the report outlines three main accountability frameworks – algorithmic impact assessments (AIAs), third-party auditing, and conformity assessments – to explain why AIAs should become the global standard.