Fear Not Artificial Intelligence (AI) Bahamas!
Nassau, N.P., The Bahamas
What happens when we begin relying on systems whose decisions we cannot adequately explain?
This is the AI “black box” problem.
Many advanced AI systems, particularly those built using deep learning, can produce remarkably accurate outputs while making it extremely difficult to determine precisely how the system arrived at a particular conclusion.
Now imagine someone asks a very reasonable question when one of these decisions are made:
Why did the system make that decision about me?
The black box problem exists everywhere, but developing countries face an additional vulnerability.
We are unlikely to build most of the sophisticated AI systems that eventually operate within our economies.
The models may be developed elsewhere, trained primarily on foreign datasets, designed around different populations and regulatory environments and then incorporated into systems used locally.
That creates an important question for countries like The Bahamas:
Are we importing technology faster than we are developing the institutional capacity to govern it?
Consider something as simple as lending.
A foreign-developed AI system may identify correlations that work extremely well within a large North American or European dataset. But Bahamian employment patterns, household structures, informal economic activity, geography and consumer behaviour do not necessarily mirror those markets.
If we cannot adequately interrogate how the decision was reached, identifying bias or inappropriate assumptions becomes significantly harder.
Much of the AI governance conversation understandably focuses on bias and discrimination.
But black box systems raise another issue: accountability.
If an AI-assisted decision causes harm, where does responsibility sit?
With the international technology provider? The Bahamian institution that purchased the system?
Or the algorithm that nobody can fully explain?
We cannot regulate an algorithm in the same way that we hold a person or institution accountable. Ultimately, responsibility has to remain somewhere within the human governance structure surrounding the technology.
That means explainability cannot simply be a desirable feature.
In certain high-impact decisions, it may need to become a governance requirement.
We should not wait until adoption is widespread.
The Bahamas does not need to fear AI. In fact, I believe small states should be aggressively exploring how AI can help us overcome some of our structural limitations.
But enthusiasm for adoption should be matched by investment in governance.
Before AI becomes deeply embedded we should already be asking:
- What decisions should AI be permitted to make or materially influence?
- When must a human remain accountable for the final decision?
- What level of explanation should a citizen or customer be entitled to receive?
- What capacity do our regulators need to meaningfully challenge these systems?
These questions become even more important across the wider Caribbean, where individual states may not have the technical resources to independently scrutinize AI system entering their markets.
There may therefore be a strong argument for regional cooperation around AI governance.
Developing countries are often encouraged to ensure that we are not “left behind” by technological change.
But there are two ways to be left behind.
One is failing to adopt transformative technology.
The other is adopting technology without developing the institutions capable of governing it.
