Can Pluralism Save Us from Rogue AI?
Why a Philosophical Framework is Needed
In the coming years, technology is going to force us to clarify our ideas. As has been the course over the past several years, as technology expands, we must decide which ideas to use to program it. But, as it stands, there are many ideas. So, how do we decide which ones to use? This is the emerging problem we’re up against.
Pluralism
Pluralism posits that instead of one correct view, there may be several views, each with their own merits. Importantly, pluralism is quite distinct from relativism. While relativism asserts that there are no true moral values, ultimately giving rise to a void in which any and all values exist without critique, pluralism argues that there could be several true values. Programming a technology with pluralism as a principal philosophy could remedy several problems which stem from the two extremes on either side (absolutism and relativism).
Relativism: Giving a technology too loose of a definition of what’s right or wrong could allow the technology to progress in a way that escapes human oversight. It could begin to develop its own moral values and convictions (for better or worse) without acknowledging a stable reference point that’s accepted by humans.
Absolutism: This risks giving technology the ability to adamantly defend harmful values or to be programmed with certain values as foundations which it's unable to give up, no matter how much it is persuaded by reason.
The Need for Perspectives
And how would pluralism remedy these problems? It could offer multiple perspectives while still allowing for the moral conviction and judgement that relativism declares impossible. Whether or not a rational explanation is available for why a value-based approach is necessary, it is empirically vital. Giving technology (such as AI) no values to reference can prove disastrous, and it seems that giving the technology values that, in our best estimation, are correct, will be comparatively better even if the values are not perfect.
An example pluralistic framework could look like a list of several values (fairness, truthfulness, etc.) which is encoded. This would contrast a single-value framework, such as a utilitarian framework of utility maximization, which would only take one variable into consideration.
Assigning Relative Priority to Values
If we decide to adopt a pluralistic approach to assigning values for technology, a problem still remains of deciding the relative priority to give these values, especially in cases where they contradict and one would need to override the other.
For this to work, the model would, again, need to choose between value-deciding systems. If it’s not clear which value to go with, what philosophy should decide? Utilitarianism? If that were the case, since all decisions would be made by this superior philosophy, in the end, there would only be a single principle (in the utilitarian case, it would be utility-maximization) dictating all decisions, inevitably contradicting the initial position of pluralism.
Similarly, a position of deontology would make appeals to an overarching value, thereby contradicting pluralism. Pluralism, also, could facilitate a power imbalance in who has the ability to decide which values should guide a technology.
Complexity Doesn’t Mean We Should Discount It.
Getting a technology to the point where it could follow one moral command is enough of a challenge. Designing a system which balances the relative strengths of multiple values is an even greater one. When we, as humans, make decisions, almost always, this process involves synthesizing multiple factors, not just one. The fact that multiple factors exist in such decisions is hardly a flaw in the system that needs to be fixed, and instead ought to be treated as a feature of the system. Much of philosophy, like the discussion of pluralism at hand, is too nuanced to be represented by a single data point.
