An important reminder:
While the engineers behind #OpenAI might be able to program over a veneer of decency, there will always be the insidious presence of prejudice.
Algorithms such as #OpenAI cannot help but reflect the biases that exist within their training data. The only way to avoid that is a completely different architecture.
@pwgtennant No.
There is no way to remove bias in training data about subjective information. No architecture change will ever do that and it is disingenuous to suggest that's ever a possibility.
@aosinski No currently available architecture. But a new form of algorithm built with using causal understanding may be very different. That may be decades off, but I don't think it's 'disingenuous' to suggest that's one day possible.
@pwgtennant causal understanding is only useful in closed systems where it’s actually feasible to derive causality through first principles.
No complex open system in real life supports deriving conclusions without extensive empirical support, whose authority is always temporary and imperfect.
We’ve known since Hume that reason alone is insufficient. To suggest a smart enough AI will be unbiased is wishful thinking.
