3
$\begingroup$

Imagine a space ship cruising in the interstellar void (this context is not necessary, but helpful, I think). Now, imagine the ship is filled with normal human beings. Immediately, things like people going crazy, people hating each other, people conspiring for power, the leader doing stupid things or people feeling nihilistic and suicidal become possibilities for this setting.

Now imagine another ship where there is only one AI on board. It has the calculating, reasoning and creative powers equal to the sum of the previous ship's crew complement, (creativity perhaps can be achieved with random-number-generators) and suddenly, all of those terrible possibilities seem to vanish.

In general, could it be said that a single intelligent agent managing a system is objectively better than a group of less intelligent beings? Seeing that no mutual conflict, difference of interest, bureaucratic red-tapes, law or democratic decision making is required for the former.

You may assume that there is a purpose in the management of this system (e.g. finding a planet in space, producing paperclips, surviving a nuclear winter) that doesn't involve letting society devolve to death and destruction

$\endgroup$
5
  • 6
    $\begingroup$ Your question is missing the purpose these beings are trying to fulfill. People are just as good as super AI at simply floating in space. Major Tom might be better. $\endgroup$
    – df86
    Nov 30, 2016 at 5:39
  • $\begingroup$ This is akin to govt of any country, the objective isn't to please everyone but rather how to win the majority over. So now I'll throw the question back to u: do you still think everyone will watch flashy piece of circuit breaker or an intense battle royale with occasional furniture storm... $\endgroup$
    – user6760
    Nov 30, 2016 at 6:55
  • $\begingroup$ Does the AI's intellect accounts for the crowd intelligence? There was a documentary about it by BBC. If have not seen it, you are in for a surprise. $\endgroup$ Nov 30, 2016 at 11:15
  • 1
    $\begingroup$ A single minds loses the benefit of multiple perspectives. You seem to beg that, though. Even if you have a mind that's better than normal, it's a single point of failure. Things that are bad, but recoverable, are catastrophic for the one. $\endgroup$
    – The Nate
    Nov 30, 2016 at 14:33
  • 2
    $\begingroup$ Lets take a vote on that... Oh, looks like the single super-intelligence is in the minority! :-P $\endgroup$
    – Abulafia
    Nov 30, 2016 at 16:31

5 Answers 5

9
$\begingroup$

Experience has shown over the ages that concentrating decision-making power into one individual is better than distributing it among many when there is a specific objective to be achieved. That's why armies need unity of command and rigidly enforce chains of command.

And even in armies at war soldiers and officers are taught to use their initiative and not overload their commanding officer with each and every detail; information overload is a thing. You want multiple intelligences so that tasks can be divided among them and not fall all on one individual who may be overwhelmed.

When the task to be achieved is not crystal clear or takes a long time to achieve then the disadvantages of absolute monarchy become dominant: one single intelligent being (an AI, a tyrant, a dictator, an absolute monarch) imposes their failings on the entire system. They can go mad; they can focus on one thing and lose track of others; they may lack the imagination required to adapt to changing circumstances; they can let themselves be enslaved by the initial goal and not observe that it has become unachievable or undesirable; they can fall into the trap of micromanagement and stop doing their job to provide leadership. History is full of bad emperors but has a much smaller number of good emperors.

As Churchill said, "democracy is the worst form of Government except for all those other forms that have been tried from time to time".

Conclusion: Even of a ship, where there is a clear and enforced chain of command, it is preferable to have multiple intelligent individuals, so that they can provide alternative opinions, exert a moderating and correcting influence on each other, and provide a stimulating medium for problem solving.

$\endgroup$
2
  • 2
    $\begingroup$ The other disadvantages of a single, centralized directing intelligence can be summed up by F.A Hayek's "Local Knowledge Problem", which explains why Socialism in all its forms (from the Fascist Corporate State to Communism) fail in every implementation. econlib.org/library/Essays/hykKnw1.html $\endgroup$
    – Thucydides
    Nov 30, 2016 at 15:31
  • $\begingroup$ I disagree. Investing power in one person is the best way to get only one decision (useful for Armies, say) but it is not the best way to get the best decisions for accomplishing goals. You should read this book for many example of the power of distributed thinking. $\endgroup$
    – kingledion
    Nov 30, 2016 at 18:45
5
$\begingroup$

Probably not. While human beings working together in groups are riven with petty hostilities, empire building, backstabbing, and colluding against one another this is a normal state of affairs and human groups can and do work very successfully together. So a ship's complement boldly going where no ship's complement has boldly gone before can be expected to carry out their mission successfully.

Human organizations have learned an entire repertoire of techniques to keep people working together without killing one another or ruining their main task. Otherwise civilization would be a complete failure, and it isn't, despite the wheels occasionally dropping off, the show does stay on the road.

Sending a solitary super-intelligence sounds plausible, but is it? Alone there will be no checks and balances on any decisions the singleton super-mind might make. Multiple individuals can bring a wide range of different sets of experience together on any problem. One super-intelligent machine might be able to deal with known problems, but could it cope with the unexpected and the unknown? true that multiple smart humans could have similar problems, but their collective and comparative ways of thinking may be better at finding resolutions than any single intelligent agent.

To paraphrase: Democracy is the worst possible form of governance until you compare it to the rest.

Creativity cannot be replicated by mere random number generators. Creativity needs to be balanced against reality. Most creators are blind to how good their own ideas are, and, in fact, they're usually convinced all their ideas are brilliant. Genuine creativity requires people to try idea after idea and subjecting them to the impartial scrutiny and scepticism of others.

Left alone for long enough, it might become deranged. This wouldn't be a good advertisement for humanity to send a mad mechanical mind to the stars as our emissary.

$\endgroup$
3
$\begingroup$

I will generalize my answer to the game-theoretic problem we're trying to solve.

The answer, unsurprisingly, is "it depends."

The fundamental problem is one of comparing:

  • a strongly unified leadership chain (in our extreme case, the enterprise consists of a single superhuman decision-maker/worker/evaluator/enforcer)
  • a loose leadership chain (separate decision makers, workers, evaluators, enforcers with some degree of principal-agent problems)

Human organizations generally can achieve more than a single human individual, despite partial or complete non-overlap in incentive structures and strategic goals among the members of the organization. This is due to several factors. First, humans have limited attention span and limited capabilities to manipulate the environment per unit of time. When a task to be accomplished can be broken up into parallelizable chunks, organizations that have evaluators and enforcers can often produce results broadly aligned with the goals of the decision makers in a fraction of the time and with significantly better error correction than a single individual could.

For example, it takes the combined efforts of thousands (tens millions if you work through the supply chains) to build a Space Launch System, and no single human can hold all that information in their mind.

However, large human organizations come with significant downsides, sometimes due to incentive differences between the leader of the organization's goals and that of portions of their team, sometimes due to the limited capability of the leaders themselves who are after all human and cannot foresee the future, or may even be playing political games if they are more interested in maintaining leadership than in accomplishing the stated goals of the organization. This is why for instance large organizations have a very hard time innovating -- the upside for a potential innovator for undertaking a risky project is overwhelmed by the perceived downside if the project fails in the eyes of their manager. Most people are content to receive a salary and are risk averse, since most of the benefits from a major innovation would not accrue to them. This is also why creative people rarely thrive in large organizations, unless special measures are taken to shield them from the bureaucratic incentive structure (special innovation labs etc).

Now imagine that you can build a human-like mind (a goal-oriented mind), but can expand the attention and reality manipulation capabilities without limit. Imagine an entire research or industrial complex run by a single vast mind with thousands of versatile manipulators. With proper protocols and subroutines, there would be no miscommunication, and a properly scaled decision maker could be aware of all the relevant factors available to the entity. Such an entity could be vastly more proficient at a particular task than any human organization.

Seems fabulous, right? So what are some potential downsides? Well, if the organization does not have a clearly definable goal (i.e. it doesn't exactly know what it wants to do at the outset) the peculiarities of its problem-solving and pattern-detection algorithms might prevent it from seeing solutions that other differently designed minds mind find obvious.

So, ask yourself. Does the solution to whatever the organization is trying to accomplish require combining multiple perspective and thinking outside the box? If so, having multiple minds will bring an advantage. Does the solution lend itself to the kind of scaling a super-human (but human-like) mind might be able to accomplish. Would a human-like mind somehow capable of controlling a thousand hands do it about a thousand times faster? If so, having a single unified entity might be a superior solution.

$\endgroup$
0
$\begingroup$

In a task as focused and specific as navigating and operating a starship, there's little question that a single overarching AI would be more effective than a crew of humans. Not only is it a task that computers are much better at handling than humans, it eliminates problems of imperfect communication between biological beings.

We see exactly this in biology. Compare the capacity of an animal, with a central brain that directs the rest of the body, to that of a tree. The tree may live for longer and grow larger, but it's also totally unable to react or adapt to changing circumstances, unable to move, to effectively sense the world - it's essentially passive. An animal, which has a central brain, is able to react to stimuli in the world and respond with a much broader array of behaviours than a tree can.

BUT...!

You need to be careful how you design your AI. If you try to model it on a human brain, you're going to have some serious problems. Human brains are molded and corrected by interaction with other human brains. We're not good at solitude, we need an active flow of stimulus to keep our brains functioning. If you want an AI that can handle operating a starship, do NOT, under any circumstances, give it a human personality. Such a being would be massively unpredictable.

$\endgroup$
0
$\begingroup$

One AI is a bad idea.

It may become faulty and needs to be shut down.

I would recommend to scatter the system. One AI for navigation, one AI for Life Support and so on. They may (or must) report to a superior instance but the systemwide concept should be able to handle the failure of one component.

$\endgroup$
2
  • 1
    $\begingroup$ That makes the superior instance into the single AI, just with subunits. A better option would be to have three equally capable AIs making decisions based on the same data. In the event of a conflict, whatever decision at least two of them can agree on is implemented. $\endgroup$
    – Werrf
    Nov 30, 2016 at 18:18
  • $\begingroup$ That was my second idea :) $\endgroup$ Nov 30, 2016 at 18:24

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .