The regulation that would work for AI safety is likely to be tools that would be adopted by AI models. Simply, AI regulation would be the deployment of technical tools that would regulate AI, to purposes.
The problem is not making laws for AI regulation, the problem is that laws will not be effective–in a technical vacuum, even if there is early inspection before model release. AI safety, at least for what would work, will have to be technical. And this technical architecture will have to be deployed by some models in some geographies, or models, whose outputs are available in search results, app or play store, social media and so forth would have to show compliance with the tools or be requested to do so.
Since there are several categories for what will consist of AI safety, there will be necessity for several effective tools. AI companies that develop these tools and can show that they can prevent some of the risks of AI across models, would become a choice of the law, for which regulation would be based. This choice would hand them profitability.
The research to get there can be described like drug discovery, which could take years, but when it passes clinical trials and gets to market, it becomes a huge source of profit.
A place to start is to look at all the negative uses of AI in the last two years, at least, that made the news, then to begin to explore model independent ways to curb those in the future. The next thing will be to structure AI safety with the human mind, since human intelligence is aligned by human affect, so AI safety has to be aligned within the provision of the model.
There will be too much politics or too much competition or capability pursuit to choose to intentionally lower a model’s ability, if there is no technical pathway against risks, threats and vulnerabilities of AI.
There are often debates on the capability level of AI, but capability can only be measured against human intelligence. There are several things that AI can now do that instead of making some people glad that AI can do that it should make them afraid. This is because AI continues to excel at using what is in digital memory, just like human intelligence excels at using what is in human memory, showing that many unique human abilities are within reach.
As AI gets better, if there is no guardrail–technically, at the source or at the output, AI would cross so many thresholds while many would say it is nothing, until it actually becomes something so unstoppable. There are several features of the human mind where parallels with AI can already be drawn. These parallels can also help to shape how AI is emerging or evolving rather than just to guess without the only benchmark possible–the human mind.
AI safety can be profitable, just that the alignment, to do so would have to be technical, predicated on the human mind.
There is a recent story on Singapore Business Review, Singapore and the UK collaborate on AI safety standards, stating that, “Singapore inked an AI safety partnership with the United Kingdom’s AI Safety Institute. As partners, Singapore and the UK will drive forward research and work towards a shared set of policies, standards, and guidance. Minister for Digital Development and Information Josephine Teo and UK’s Secretary of State for Science, Innovation, and Technology Peter Kyle signed the Memorandum of Cooperation (MoC).”
There is another feature on Cornell Chronicle, BTPI releases new report on AI regulation, stating that, “The Brooks Tech Policy Institute, with support from the Jain Family Institute (JFI), has released a new report that offers “a high-level framework to analyze regulation of AI technologies.” Sarah Kreps– the John L. Wetherill Professor in the Department of Government in the College of Arts & Sciences (A&S), director of the Brooks School Tech Policy Institute and a professor in the Brooks School– and Adi Rao– a doctoral candidate in the Department of Government and an adjunct at the RAND Corporation– worked together to produce the report as part of a unique fellowship designed to encourage faculty/student collaborations in applied research. The fellowships are made possible by support from a collaboration with the Jain Family Institute (JFI), founded by Bobby ’92 and Carola Jain.”
There is a recent press release, Canada launches Canadian Artificial Intelligence Safety Institute, stating that, “Today, the Honourable François-Philippe Champagne, Minister of Innovation, Science and Industry, announced the launch of the Canadian Artificial Intelligence Safety Institute (CAISI) to bolster Canada’s capacity to address AI safety risks, further positioning the country as a leader in the safe and responsible development and adoption of AI technologies. It is one component of a broader $2.4 billion investment announced in Budget 2024 to help researchers and businesses develop and adopt AI responsibly. The Canadian AI Safety Institute is part of the government’s broader strategy to promote safe and responsible AI development in Canada, which includes the proposed Artificial Intelligence and Data Act and the Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems. With this institute, the federal government is helping ensure AI safety and building trust in the technology. Fostering greater trust in AI through its responsible adoption is to secure Canada’s AI advantage today and for generations to come.”
—
iStock image
