Addressing Safe and Ethical Development of AI

By Ethan Bingemann.

The best way of preventing misinformation, or “fake new,” from being spread by AI is holding news sources accountable and promoting fact checking in media. AI deepfakes are not incredibly realistic at their current stage. Going forward, it is imperative that consumers be more critical of the news sources they consume. This, ideally, is to be compounded with a movement by governments to institute ways of holding news sources accountable through fines or penal actions depending on severity.

In the world of AI, data is incredibly valuable. The more data that is available, the more accurate the system is. Put simply: an AI system is only as good as its dataset. This is the most important thing to understand when it comes to bias in AI systems. AI developers must look internally for discrimination to prevent further discrimination. If a country has a history of discrimination, that must be considered in development of AI. AI has potential to be less biased than humans because it does not have unconscious biases, but it only knows what it is given. As such, when a government allows for the proliferation of AI in society, it should be done with an emphasis on the demographics of the country. Unilateral legislative action from the government cannot prevent biases when AI is used by private businesses. What governments can do, however, is develop and offer unbiased data sets for domestic companies to access. This makes it so that companies have no excuse for discriminatory systems.

Governments need to be transparent about how AI is being used in surveillance. This is especially important in countries that establish “innocent until proven guilty” as a legal right. Using AI in policing is a growing trend that should not be condemned because it has serious potential to reduce crime, but there needs to be highly tested AI systems otherwise there is a very big possibility for mistrust in law enforcement. In doing so, we must realize that AI has shown the tendency to make many mistakes and is prone to inaccuracy a fair amount of the time, even if AI has the theoretical potential to be more accurate than humans, especially when doing “mindless” tasks. The amount of testing and vetting for AI systems is dependent on the situation in which the AI will be used. It is important that we emphasize that AI is in the developing stages where the majority of issues are found with any piece of new technology.

AI will both create jobs and replace jobs. In response, the solution should not be to push away innovation by making laws that prevent job loss entirely. Innovation naturally means that jobs move around, and AI is no different. Our delegation advises other countries to invest in education for irreplaceable jobs in STEM fields. Job loss needs to continue to be monitored by world governments and will look different in every country.

With China and the United States housing the majority of AI firms it is important to prevent an arms race of sorts. It will prove increasingly necessary to establish more international coherence in the implementation of AI regulation. This makes it so that no country is disincentivized to create regulation by another.

Limits on type of AI are important. The AI development market should not rush into creating AGI (artificial general intelligence) without first considering the impacts it will have and consulting multiple ethics boards. AI regulation, however, needs to be specific to the task and not broad in application. The number of things that AI can do in one second is more than a human can process or understand quickly, much less the actions of all the AI in the world. These regulations on AI systems need to consider biases holistically and systematically in order interpret and share what a given AI system is doing and access its impacts.