Walking the Tightrope: Navigating the risks and rewards of AI

When technologies and markets are still in their infancy, it is difficult to predict how they will evolve and what the implications might be.

The recent fundamental shift in AI is associated with scale - AI models now have access to massive data sets and tremendous computing power to train on. That’s why the latest AI-powered tools and language models seem like something out of science fiction. The question now is this: is AI really a hidden threat capable of eventually dominating and controlling society? 

Well, it is hard to imagine that kind of threat materialising. Still, AI has already demonstrated its potential to manipulate and influence human opinion. Thanks to the surfacing ethical and moral dilemmas and the growing concerns regarding AI getting out of hand, controlled and responsible use of AI is becoming a central challenge for the AI community. Several tech visionaries have called for a moratorium on further AI development, at least until we have controls in place or until we understand its true nature and potential for harm.

AI Optimists vs. AI Pessimists

The exponential growth of AI in the past few months means that experts have not been able to fully analyse AI risks against its benefits. This uncertainty has led to two opposing camps in the AI industry. On one side, we have AI optimists, led by figures like Bill Gates, who advocate for the careful development of AI with appropriate guardrails to ensure responsible growth. On the other hand, Elon Musk and Steve Wozniak believe AI poses significant threats to humanity and could lead to catastrophic consequences.

If anything, this debate underscores the importance of responsible AI use and development. While AI has the potential to revolutionise the world, it is important to ensure that it is developed ethically and mindful of humanity. The excitement about this technology shouldn’t blind everyone to the potential risks, such as malign actors exploiting it for malware coding, creating deepfakes, launching sophisticated phishing campaigns, and spreading misinformation or state propaganda. As AI advances, there is a need for investing in new tools and harnessing the existing ones to handle the new technology safely and responsibly.

The Challenge of Regulating AI

One of the biggest challenges arising from the exponential growth of AI is figuring out how to regulate it. The idea of a trusted authority regulating the AI space instead of leaving critical decisions up to individual technologists and hoping for the best sounds reassuring. But in practice, regulations always lag behind the market, and a major challenge concerning disruptive technologies is that there is no well-defined roadmap. When technologies and markets are still in their infancy, it is difficult to predict how they will evolve and what the implications might be.

The required combination of skills to craft policies around AI development and use is also almost non-existent right now. People with backgrounds in computer science typically lack any understanding of policy, legal, and other related aspects, whereas people with policy experience lack the intuition required to keep up with the fast-paced evolution of machine learning and AI. Unfortunately, this isn’t a gap that can be filled overnight. It will require technologists or policymakers to equip themselves with the right skills and knowledge needed for effective regulations in this area years before.

The Implications of Early Release and Mass Deployment

Releasing a new technology into the public domain is always a difficult decision for companies, especially when the technology is still in its nascent stages. Experts believe that releasing such powerful technology too early can pose security risks because much of its use cases remain unknown. On the flipside, early release can bring certain benefits, as in the case of ChatGPT, where its public release and rapid adoption (100-million users in two months) sparked meaningful conversations about the ethical, moral, and responsible use of AI. It is precisely the technology’s hasty mass deployment that potentially raises significant security concerns.

Many companies are hopping on the AI bandwagon, aiming to replace people with technology. For instance, IBM just recently announced possible plans to halt recruitment for around 30% of non-customer-facing positions that can be replaced with AI. The problem with this approach is that so far, companies lack the expertise required to accurately calculate the risks associated with deploying these models and granting them access to internal data at scale. Before integrating any new technology into a business process, organisations must thoroughly understand its implications and ensure that it’s mature enough to minimise potential risks.

Another challenge is the need to ensure the accuracy and reliability of the data being used to train AI models. When companies make decisions based on data-driven outputs from AI, mistakes, errors, even inherent human biases reflected in the training data, simply magnify. One way to overcome that is by analysing AI-enabled decisions over a period of time to discover systematic issues. However, the technology is too new and still evolving to be able to establish its reliability and accuracy through such statistical analysis.

Charting the Course for Responsible AI

The immaturity of AI, uncertainties around its future, and lack of theoretical understanding are some major challenges surrounding effective policymaking for responsible AI. Ultimately, it requires collaboration between researchers, policymakers, and stakeholders from a range of industries and sectors. It is essential that businesses collectively slow down on AI adoption for now and carefully consider its long-term and immediate implications. They must work toward developing a consensus on how to address privacy, security, and ethical concerns of AI, before making it a dominant part of critical business processes.

Want to find out more?

Listen to Steve Durbin in conversation with Ellie Pavlick, professor at Brown and research scientist at Google: