42 karma · joined July 12, 2020
"An AI is some constructed entity which, starting with a goal, uses its intelligence to make changes to the world in order to bring that goal to fruition. One of the things about AI that is worrying is that its system-of-values might have no relation whatsoever to those possessed by human beings, and the best way for it to accomplish its goals might be to do things that most people would consider quite nasty. AI systems that are trying to maximize something are in constant tension with the laws that we put in place to try to keep them from steamrolling over everything in pursuit of their goals. Corporations fit the bill, under the assumption that the goal is 'maximize shareholder value'."
While a nest of ants or a swarm of bees certainly exhibit interesting emergent behavior, perhaps even collective intelligence, they 1). aren't constructed and 2). don't have clear goals.
You could argue that smaller groups of people fit the bill too, I guess. Maybe the difference is that corporations tend to have distinct goals which outlive their members, and that they have special recognition under the law as independent entities. Not sure.
However, I suspect that there's also a counter-force, a "diminishing returns" effect: as more people get involved in an organization, coordination becomes more difficult, communication becomes more expensive. If this converges to some limit, then that's the smartest a human organization can be.
Maybe better methods of organizing knowledge could help raise that limit, or more effective modes of communication, but my suspicion is that the upper bound for "organizational decision making capability" is within one or two orders of magnitude of a single person, not vastly more. No idea how we'd actually quantify that, of course!
Corporations _are_ acting as misaligned optimizers. Solving that problem is hugely important. However, the "AI" comparison breaks down somewhat when you start thinking about how we might actually fix the problem. With corporations, we (i.e., states) have tools that we can use to regulate bad actors. Software AI, however hypothetical at the moment, seems likely to be a different game all-together.
An R&D lab certainly can do more work than any single person could do in their lifetime, but are the ideas fundamentally better? Can you point to an idea and say: "no person could possibly have thought of that"?
Another way to look at it is with regards to capacity for self-modification. If a corporation can't re-structure itself into being much smarter than the smartest human, its intelligence is fundamentally limited, and therefore so is the risk. Does software have this restriction? We don't really know yet, but it's hard to point to exactly why it would.
To paraphrase: yes, corporations function as agents, but their maximum performance is limited by the capabilities of their employees. A corporate AI may have a much broader set of skills than any single person, and may be able to tackle many concurrent tasks, but as an "intelligent agent" its decision making capabilities probably don't scale exponentially (or even linearly) with its headcount. In the sense that a corporation's maximum intelligence is likely to be in the same ballpark as the smartest humans, it can't be seen as a true superintelligence.