Embedded Files
We’ve all heard that AIs might soon be smarter than humans at any activity. So how much smarter could an advanced AI possibly get, and what does that mean in terms of comparative behavior or job performance?
If we define intelligence to be, in part, the ability to do mathematical computations quickly, then it is clear that nothing can be “infinitely intelligent”. There are limits set by the speed of light (latency), the Second Law of Thermodynamics (Landauer’s Principle), quantum mechanical uncertainty (Bremermann’s limit), Computational Complexity, Gödel's Incompleteness Theorems, Turing's Halting Problem, Chaos theory, and so forth. These limits apply even if we consider massive parallelism or quantum computing. Nonetheless, the amount faster that a theoretical computation-based AI could be than the human brain is cosmologically high, like 38 to 50 orders of magnitude (assuming that all atoms in a mass contribute to the computation.)
These considerations imply that some form of AI could be incredibly faster, more efficient, and more accurate than a human at any analytical activity; in other words, any task that could be broken into logical steps.
The big questions revolve around the “soft skills” of humans. Is it possible for an advanced AI to deal with human emotional consideration, judge the character of its human collaborators or customers, express true empathy, make ethical judgements, or have “skin in the game”? Of particular importance is the question of ethics. Is it possible for the AI to make sound ethical decisions without the capacity for human vulnerability? Many opinions that I read say “never!”
I challenge those statements. First of all, “never” implies a very long or infinite time. The Law of Truly Large Numbers and the Poincaré Recurrence Theorem show that the probability that anything can happen is 100% given enough time. Secondly, I submit that programming a code of ethics into an advanced AI is quite achievable, including the multitude of mitigating circumstances to be considered, assuming that we humans can agree or reach a consensus on the components of such a code. And finally, does an intelligence need to have “vulnerability” to make the best ethical decisions? On the one hand, I’m not so sure that being vulnerable always contributes to the fairest decision, and on the other, what makes us think that a super-AI might not exhibit vulnerability? As it becomes more complex, it might worry about its own “mental” stability, for example.
Finally, I challenge the opinion that humans are superior to some form of advanced AI at making ethical judgments. For example, one only has to read a current newspaper to see the fallacy of the ethical judgments of our top global leaders.
In conclusion, I submit that it is quite possible, and I believe even probable, that some form of advanced AI will someday be superior to humans in all aspects of any job.
Note: I used AI to help me with the research for this blog, but the prose and opinions are entirely my own.
Page updated
Google Sites
Report abuse