Is the AI doomsday approaching us? A quick read of daily media and social media articles on the subject reveals a bewildering variety of views.
During our career as data scientists, we have been building machine learning models (a subset of AI) for production applications for over a decade. However, with the recent public availability of generative AI technology (using large language models (LLM) to generate text such as his ChatGPT, or images such as DALL-E), the use of AI by ordinary people is becoming a precedent. No scale.
The questions we would like to consider are:
- How much should we fear AI technology and its advancements?
- If we are so afraid, what should we (government, society, individuals) do about it?
Federal Minister of Industry and Science Ed Husik made a justifiable comment: “We want to avoid destroying or spreading rapidly evolving technology.”
But rather than the technology itself at heart, AI is the mechanism by which AI delivers benefits, combined with the social structures and incentives of the individuals, companies and governments that use it, raises concerns.
Since the beginning of the Industrial Revolution in 1760, technological advances have brought about major social changes. Today we call those who insist on printing documents in order to edit them “Ludites”, but the term originally referred to British textile workers who opposed the editing of documents. . Their livelihood has been lost to the introduction of mechanized looms. Historically, technological advances have created winners and losers.
In retrospect, the rise of social media was a huge social experiment. Initial results are now available, and it is clear that the experiment has caused serious harm.
Social media takes and scales normal but devalued human traits, such as gossiping or ostracizing individuals or groups considered “others.” As a result, it has adversely affected the mental health of users (especially young people), co-created radical worldviews, and undermined democracy. As Professor Toby Walsh points out, we saw only harm. rear They have had a great impact on society.
One of the key characteristics of AI technology is that it leverages and scales activities and processes normally performed by humans. So the comparison to social media is valid. The positive impact of scale (such as efficiency) is usually the underlying mechanism for the positive impact of AI technology.
However, you should also be aware of the negative effects of this scale. The problem is compounded by the rapid pace of technological advancement, as it is difficult to predict the specific risks that these new technologies may pose.
Generative AI requires supercomputing-like power, most of which is in the US, China, and Europe. In the United States, the capital and intensive infrastructure, computational and human resources required to develop and continue research in generative AI are concentrated in a few of the largest technology companies.
Some of China’s tech giants likely also have the scale to advance generative AI. Search giant Baidu will release his LLM product in March, with many other Chinese tech giants planning to follow suit.
We sometimes see hilarious claims in the media that generative AI will democratize AI by enabling even smaller organizations to take advantage of it. This overlooks an important fact that organizations such as: create Models have great power.
Big tech companies already have more wealth and probably more power than many countries. In 2021, Apple’s market capitalization will exceed $2.1 trillion, but he is only seven countries with a higher GDP.
In discussions with many opposing opinions, keep in mind the incentives of both individual commentators and the major players in AI development, and how they can influence opinions and actions (consciously or unconsciously). It helps to recognize that there is of these groups. What motivates organizations and countries to invest in AI research and innovation?
Profit is an important driving force for organizations, and documented examples of unethical behavior by profit-motivated organizations are long and shameful. Examples range from the loss of life caused by the Radium Dial Company (look it up!) and Big Tobacco, to the unethical use of personal data by Cambridge Analytica, and the widespread issues uncovered in the Uber files. over.
Nation-states are concerned with protecting their sovereignty and security, ensuring access to resources, and growing their economies (at least at the naive level of how data scientists understand political science). The technology required to ensure the security of a nation-state is usually the same technology that supports aggression.
So yes, I think it’s appropriate to be afraid. Fear enough to take action now. We believe regulation is necessary.
This massive social experiment in which we are all participating is too dangerous to be left alone without proper governance.
Different countries are taking different approaches when it comes to AI regulation. Self-regulation is the path the US is currently on, and Canada and the EU are examples of countries taking governmental regulatory action. European governments tend to adopt the precautionary principle and actively develop technology governance.
Australia has historically used a “soft law” approach to developing specific governance frameworks for the development and use of AI systems.
Organizational profit incentives make self-regulation seem unrealistic. To understand the importance of regulation, one need only consider a parallel sector like finance, which is similarly driven by profit motives and affects people’s lives on a large scale. Regulation also creates a level playing field so that organizations are not penalized for acting ethically.
The incentives to unite many nation-states on the Nuclear Non-Proliferation Treaty may be a useful analogy. Similarly, Australia needs to harmonize its regulatory approach with its major allies and, if possible, its major trading partners.
So where are we going next?
The development of good regulation requires a multi-stakeholder effort, with contributions from various sectors as well as the public and private sectors.
One proposal by Congressman Julian Hill is the establishment of an Australian AI Commission. Hill believes this approach is “preferred to traditional government sector processes” in tackling such complex and rapidly evolving challenges. The committee will consist of a cross-disciplinary team of industry, APS, academia and civil society representatives.
Such a committee, in principle, seems like a good approach. However, designing sensible vehicles and structures is an easier task than designing the right requirements for the purpose to be achieved.
From an APS perspective, Hill said: “Government managers need the confidence to deploy technology for the benefit of all with adequate safeguards.” increase.
Now is the time for Australia to act. As Mark Twain once said, “Progressive improvement triumphs over retarded perfection.” And it is important that governments consider diverse views when formulating policies.
The government recently released a consultation document seeking views on possible regulatory responses to curb the risks of AI.
Now is your chance to shape the conversation.
