Artificial Intelligence (AI) has been around since the 1950s. Developments around it, till recently, were mostly limited to an informed few. However, the launch of ChatGpt and emergence of deep fakes (audio and video) made it to the headlines and AI became a household topic of discussion. A matter of more concern has been the possibility of the development of super-intelligent machines which could potentiality harm human beings.

In order to curb this, in December 2023, the European parliament agreed to an AI Act which is yet to be adopted. This classifies AI-related risks in tiers of unacceptable, high and limited risks. The USA has issued an executive order to combat any racial bias and also to protect people right and safety. In India too, after the recent emergence of deep fakes, there has been a debate on ways and means to control such rogue deployment of AI.

Some of the areas which are planned to be controlled are racial bias in data banks used to train AI models and also the introduction of transparency in how content was generated by AI. Areas of concern have been in racial profiling using facial recognition, copyright violations, surveillance and law enforcement. These concerns are serious, but there is a genuine question of how to ensure these are taken care of.

Ensuring that the databanks used for training are free of racial bias would require detailed audit of the content used to train the AI models prior to deployment. This in itself is a huge task. The first issue is to ensure the availability of content to be audited. Once the content is available, which may run into several terabytes, the effort to audit would be tremendous.

In line with this is the issue of transparency. Here the need is to reproduce how the AI model reached a particular decision. This would require disclosing each parameter that the model considered while reaching a decision. As models involve learning algorithms, it may be near impossible to retrace each step that the model took to arrive at a decision.

Besides the complexity of auditing and issues involved in ensuring transparency, there is an issue of cross-border cooperation. While the European Union may have initiated the process to regulate and the USA has laid down guidelines, if other nations do not follow suit, in an interconnected world, these regulations in themselves may not be fruitful.  

China, among other nations, has been at the forefront of AI development. It was in 2017, that China formulated ‘New Generation Artificial Intelligence Development Plan’ which disclosed the country’s vision to be at the forefront of AI by the year 2030 [https://www.newamerica.org/cybersecurity-initiative/digichina/blog/full-translation-chinas-new-generation-artificial-intelligence-development-plan-2017/]. It was against this backdrop that Elon Musk, while interacting with UK prime minister Rishi Sunak on the eve of the AI Safety summit in UK in November 2023, said that it was important to get China on board when considering the impact of AI.

This is even more important as China wishes to be setting the standards for AI rather than take any standards from the western world. In view of this, it is imperative that any regulation, if it were to succeed, needs to include China and also other nations who may not be too visible as of now.

Having said this, let us consider how these regulations, if made, will be enforced. To begin with, it requires international consensus to enforce. Even if this is achieved, technicalities involved in evaluating or auditing content and also to reproduce steps taken to arrive at a decision may be humanly and technically impossible. And due to this, probability of a rogue system making it to the internet may be quite high. In the recent past, Microsoft’s Tay, an AI chatbot trained on Twitter, went rogue. Luckily, it could be brought down before it could spew unwanted venom. The same may not be possible with other future rogue systems. And the worst-case scenario is a machine which is super-intelligent goes rogue and does not have a kill switch.

This is a possibility as the internet is available across continents. To consider that AI-related systems are only being developed in countries which are debating or drafting AI regulations is being naïve. Therefore, any successful control to ensure the world does not have to face rogue machines has to bring in international coordination. Once this is achieved, a set of regulations or legislations have to be uniform across continents to ensure uniformity of enforcement and development. Alongside controls, success of any such regulation will depend on liabilities for developing rogue systems. These need to be clearly spelt out and imposed uniformly across borders.

Daunting as this may appear, there will possibly be no short cuts. Risks associated with unbridled development may be too high for humanity to sustain.

Pandey, a former Indian Police Service officer, has a BTech in Computer Science from IIT Kanpur and is a Certified Information System Security Professional (CISSP), apart from holding LLB from Bombay University and MPA from Harvard University.