For
an AI to be a country’s prime minister or president, that country will
first need to grant “natural person” status to AIs so they are
equivalent to humans. Assuming this is done, we then need to consider
what the prime minister/president does as head of the executive. The
exact tasks vary between countries, but they generally include things
like:
- Appoint and manage heads of government departments (treasury, taxation, social services, infrastructure, military, etc.)
- Manage the elected members of the ruling political party or coalition. The management may be hands-off if the executive branch is separate from the legislative branch, or more direct if the two branches are co-located.
- Possibly appoint members of the judiciary, or approve/veto appointments proposed by another part of government.
- Overall military command: decide on military deployments and the parameters of those deployments.
- Communication with the press and with the general public in many ways.
All
of these tasks are quite fuzzy and ill-defined, at least from a logical
perspective. At the moment, it’s not at all clear how one could create
an AI of doing such a job at least as well as your average politician
(notwithstanding the many opinions of the human-based ruling class).
Are there other government jobs where AI could do well? Yes, including the following:
- Judiciary: Basically, judges combine codified laws with presented facts into decisions, and explain how they arrived at those decisions. All of this is written down in excruciating detail. It is ideal for data mining and machine-based reasoning. Technologies such as IBM’s Watson are already being tested for this type of application (I think).
- Taxation and Infrastructure: If you have played the computer game SimCity, you might know that there is a complex societal model underlying the game’s mechanics. Such models are becoming more applicable in the real world as measurement of real-world inputs (e.g. traffic, pollution, energy demand, people’s locations) becomes more reliable and real-time.
- War: AI in war has been investigated for a long time - even a large chunk of the computer gaming industry simulates warfare using AI (e.g. Halo, Command and Conquer). The 1983 movie WarGames looked at the possible consequences of putting an AI in charge of nuclear weapons. While high-level well-defined battlefield engagements and war logistics are probably well-understood by now, wartime human psychological models are not. Summarized in a nutshell, current AI would be good at winning battles, but it would probably lose modern wars which feature fuzzy motivations and shifting identities.
In
summary, I expect that AI will infiltrate government from the bottom
up, via the government departments and their specific tasks. The
president/prime minister is probably the last job that will go. More
likely, the AIs assigned to different tasks will start to coordinate
with each other to work out solutions to problems - first withn a
department, and then between departments. This process will be
facilitated by the human masters looking for more efficient solutions by
sharing knowledge between AIs.
As time goes
on, the humans will become more and more redundant in the solution
process, centralizing decision power among fewer and fewer remaining
humans running the machines. Power struggles will emerge between the AI
system architects and the politicians who are supposed to control them.
At a certain point, the power will lie entirely with the system
architects. Perhaps this was why the system architect was such a
prominent character in the 1999 movie The Matrix.
No comments:
Post a Comment