[clort]

Having an AI as the president would make it clear to all that there is an invisible group of people behind the president, pushing them to act and say things that they want. I agree that even now Watson could probably make decisions better based on actual facts but if Watson says something that is unpalatable to its owners then they can reboot it with a different set of facts. IIRC they did that after they fed it the urban dictionary and it started swearing too much[1]

At least with a human president, there is the chance that they will grow up and shrug off the orders they are given. The power actually lies with the person that was elected, not the invisible people who paid to get them the job.

[1] http://www.theatlantic.com/technology/archive/2013/01/ibms-w...

[mbrock]

The site's creators perhaps acknowledge this in a way, by listing a number of bullet points that they expect Watson to advise on and presumably agree with: single-payer health care, recreational drug legalization, and so on.

But how do they know Watson would find the expected value of these things positive? Maybe Watson would be a republican. And this all points down a huge rabbit hole of ethical and political philosophy and stuff.

Like, should Watson take into consideration his likelihood of being elected? In that case, much of his neural network should be dedicated to predicting voter outcome. And that seems pretty obviously problematic.

[nickodell]

There's a simple solution to this: give Watson two platforms, one Republican, and one Democrat. Have Watson run in both primary elections. If he wins both primaries, on election day, voters will be able to choose between Watson running in Democrat mode, and Watson running in Republican mode.

[joesmo]

So we have a supercomputer capable of running the country and you basically want to put the stupid hats on and limit its thinking to bipartisan bickering and two semi-opposing points of view that virtually no one holds but is currently forced to agree with for lack of an actual democratic process? Why not just use an 8086 processor or other 8-bit CPU then, instead of wasting all that power, because I'm sure you can get the same answers with both at this point. Every 8-bit CPU I know of can be programmed to look up values in a hash table and spit out the expected results, just like human politicians program their brains to do.

[nickodell]

You may be taking this a bit too seriously. I don't think that Watson would make a particularly good president - I'm not convinced that it can adapt to unexpected situations, or make calls on when to hire or fire people. For example, if the White House Chief of Staff tells Watson that he should fire his Press Secretary, should Watson follow that advice?

Now that I've gotten that out of the way, I'm going to respond to your counterpoint in character.

----

If Republican Watson ran against Democrat Watson, for the first time, we would have a debate purely about issues and policies, and not about personal character. As for the platforms themselves, you might view those political platforms as wrong, but millions of Americans agree with them. This is just the ultimate expression of democracy: a statesman whose views conform to those of the people.

>Every 8-bit CPU I know of can be programmed to look up values in a hash table and spit out the expected results, just like human politicians program their brains to do.

That's true. But if Watson is 1% better than an 8-bit CPU, then that justifies spending millions on the upgrade, since the role of President has such a large effect on how well the government runs.

[saeranv]

I love this idea. One question is how will we accommodate the large spectrum of ideologies within the parties: Ron Paul libertarians vs neo-conservatives like Marco Rubio for example. Or the nuanced distinction between Bernie Sanders' notion of 'progressive' and Hillary Clinton's.

Pragmatically speaking, I think we could capture the political spectrum in 4 to 5 parties. Watson-left-liberalist mode (for the Sanders crowd), Watson-authoritarian-right (for Trump supporters) etc.

Edit: Sanders correction based on comment below.

[thevibesman]

Sorry to be the language police (but word choice is important when labeling politicians): I don't think any Sander's supporters would vote for a left-libertarian mode; I think you meant left-liberalist or something like that. Libertarianism would be opposed to the role of government in Sander's democratic-socialist platforms.

[saeranv]

No need to apologize, you are definitely correct. I will change it to left-liberalist, that's a lot more accurate.

[seabass]

It's interesting you'd put Trump on the far right. I find Trump to be one of the more moderate republican candidates when compared to Cruz, Carson, and Paul, for example. Trump wants universal healthcare and a progressive tax where the lowest bracket remains 0%. I think most liberals have this caricatured image of him that lead them to these exaggerated conclusions, and while I don't like the guy I really think you've misrepresented his position on the political spectrum.

[saeranv]

Check out the political compass, he is classified as strongly authoritarian right: http://www.politicalcompass.org/uselection2016

[hammock]

Whether Republican Watson or Democrat Watson wins, it's already an improvement in that Watson, being a computer, will actually execute policy the same way he said he would during his campaign.

[chillacy]

Or Watson learns (through gradient descent of course) that you can make wild campaign promises to win the election and not actually execute on that policy because it's more beneficial to make grand promises while running and not rock the boat while in office.

[akshatpradhan]

So try it out and put Watson in both modes. Ask him some questions, and do a voting poll to see what the sample size likes about both of them. It might be a good way to see how far Watson can be pushed.

[partiallypro]

I mean, if the Presidency was a dictatorship, then maybe. Otherwise Watson still has to go through Congress.

[alttab]

Democratic super-robot programming voting. I'ma go contact a screen writer now.

[HillRat]

Watson will agree with whatever training corpus you feed into it, so like most candidates it can't tell you anything new or novel. However, unlike actual political candidates, Watson would be able to directly answer questions.

The problem is that Watson has been talked up as almost a strong AI, when it's actually a really good classifier, annotator and summarizer. While there's a great role for Watson-style systems in policy development and practice, they are only one in a battery of ML and analytic techniques, none of which can stand alone without a fully human point of view.

[excalq]

What if Watson read Conservapedia thinking it to be factual. That's a startling thought.

[13thLetter]

> But how do they know Watson would find the expected value of these things positive? Maybe Watson would be a republican.

Easy enough. If Watson turned out to be a Republican they would start tweaking the parameters until it became a Democrat.

[kuschku]

> Maybe Watson would be a republican

That’s very unlikely.

Any logical system based on the concept that human life is worth existing on its own (no matter what the person has contributed to society) automatically ends up with the necessary conclusion that things like subsidized healthcare are mandatory.

Obviously, one could give the program the basic assumption that human life is not worth anything, and it should instead focus purely on profit, and it might end up with a more republican ideology.

But giving an AI with access to nuclear weapons the assumption that human life isn’t the most important factor is... a bad idea.

[mbrock]

What if the logical system determines that the introduction of single-payer healthcare would lead to some kind of political crisis in the next 8 years, for complex predictive reasons that the human mind can barely understand?

It's easy to derive conclusions from moral axioms, but very difficult to do actual politics in a country full of voters, corporations, lobbyists, etc. Artificial intelligence is not a magic solution to that.

[DonHopkins]

It would certainly be a political crisis for the Republican party if their base were to realize how much they've been lying about health care. Bring it on!

[kantianethics]

It's easy to pick sides, whether Republican or Democrat. It's harder to understand that both arguments about society have value.

Fundamental Republicans believe capitalism and the free market is the system that "works best" to ensure a level playing field for everyone. True republicans work to ensure a fair marketplace for everyone, both for poor individuals and rich corporations, providing citizens the ability to better their condition and increase their freedom.

Fundamental Democrats believe that the freedom of citizens is constantly at risk from outside factors, and that the government is the best agent to maintain citizen's free will.

True republicans believe that human nature is fundamentally good, and that the government increases equality by maintaining a free market, while democrats suggest that human nature is often weak, requiring the government's intervention to protect society from itself.

In reality, very few Republican and Democrat politicians actually represent these values. Often, republican views of the free market disproportionately benefit the rich, and democratic attempt to redistribute wealth, instead of fixing inequality at the source.

Although I consider myself independent, I personally would side with a republican interpretation of health care. Studies have shown that government programs in fields such as healthcare are inefficient when compared to their private counterparts, as a competitive market increases supply, and therefore decreases costs, as opposed to monopolies or government programs that provide a single source of service.

[kuschku]

> Although I consider myself independent, I personally would side with a republican interpretation of health care. Studies have shown that government programs in fields such as healthcare are inefficient when compared to their private counterparts, as a competitive market increases supply, and therefore decreases costs, as opposed to monopolies or government programs that provide a single source of service.

There is no need to choose – you can easily provide a minimum standard to everyone, and let the market handle everything above that.

Which is the concept of the social market economy in general: Everyone gets at least a specified minimum level of service, provided by society, paid by everyone (the social part) – and everything above that level is done with a fair and free market (the market part).

[ajmurmann]

Agreeing on moral axioms also might be non-trivial. I for example do not agree with the one you list above, but instead go withz Peter Singer on preventing suffering and not protecting life per se. So abortions for example and even euthanasia of severely sick people including newborns are fair game. Also leads to the question if animal suffering should count and if so, how much? I think trying to take our moral debates down to principals like this might be a super valuable exercise for society though. Regardless if we program Watson to be president or not.

Edit: of course this in practice would devolve to 50+% of the country in essence saying "Whatever I think the bible says should be our acxiom". Completely ruining the discussion.

[kuschku]

(I personally also subscribe to the "preventing suffering" axiom, but wanted to show a better contrast)

But then the AI may decide life is suffering, and end suffering by ending life...

Asimov’s laws of robots end up with the previously mentioned axiom, though.

[ajmurmann]

That's a very interesting point! Watson is I essence a robot, so Asimov's laws should be a good starting point. But are they also a good starting point for the president? Why should laws be different for robots than for people? My head is spinning...

[krapp]

>Asimov's laws should be a good starting point

No they really shouldn't. Asimov's Laws of Robotics were a plot device intended to be subverted by the robots in his stories, they were never meant to be taken seriously.

[awinter-py]

Hard work and change are both forms of suffering (they're unpleasant as you experience them). Why not a more useful goal with the ability to say 'ok, done' -- like 'develop octopus-like mechanical limbs' or 'regrow human arms by splicing in reptilian DNA'.

After all, neither of those experiments has ever gone seriously wrong.

[Eerie]

>Any logical system based on the concept that human life is worth existing on its own (no matter what the person has contributed to society) automatically ends up with the necessary conclusion that things like subsidized healthcare are mandatory.

And that abortion is illegal. :-)

[alttab]

If its mandatory, then every fertile human is walking around with the "right" to produce more burden for the state. That would make reproduction a liability... which turns into birth licenses.

[joesmo]

No because an intelligent life from would not define a bunch of cells that just started multiplying last month and cannot survive on their own as as life.

[trbvm2]

As it happens, the ability to reproduce and the ability to respond to stimuli in any manner are generally accepted qualities that distinguish life from non-life.

http://www.biology-online.org/dictionary/Life http://dictionary.reference.com/browse/life

Whether a given bunch of cells may be considered human is still very much an open question but whether a cell or cluster of cells is alive should be pretty easy to agree on.

[jcranmer]

An unborn, non-life form still has utilitarian value: there is an expected value that it would perform throughout its otherwise natural life.

Actually, a strict utilitarian model would probably conclude that it's not worth aborting to save the mother's life if the baby is viable, since the baby would ultimately produce more value for society than the mother would were she to live out her life.

And that's the sort of reasoning that makes everyone hate utilitarian ethics.

[sonwat]

You're assuming an unwanted, unborn baby has value. Not only does it not have value, but it's actually a burden to society. Utilitarianism would consider the damage that unwanted, unborn baby will inflict on society and decide to abort it every single time. Social welfare, orphanage, and especially criminal costs are incredibly likely and incredibly high. The chance that the baby will amount to anything worthwhile enough to offset those costs is incredibly low for an unwanted baby and thus not worth the risk to society. Here in the US we have seen the criminal costs of outlawing abortion with the high crime rates of the 70's, 80's, and early 90's finally coming down in the last two decades due to legal abortion. Other countries like Romania know this equally well. That doesn't even begin to take into account the rest of the social costs of forcing unwanted babies to be born.

[grondilu]

> human life is worth existing on its own

Worth to whom? A human life has value, but not to everybody. Or at least not the same value for everybody. Keeping people alive at any cost, and imposing to people how much they should contribute for achieving that, is not something everyone can agree with.

[conceit]

> Keeping people alive at any cost, and imposing to people how much they should contribute for achieving that, is not something everyone can agree with

Oh, unless the people are politicians, is that what you are saying? Of course, not everyone can agree with that, either. But it is already the norm, so this is just an extension to the rule. So the question is just how much they would agree to contribute.

[saeranv]

I personally agree with you but there are arguments against subsidized health care from the right. It is a cliche, and I think, easily refuted, but there is the libertarian argument that subsidized health care places infringes on personal liberty (forcing one to pay tax, incentivizing the government to regulate your health).

Essentially, what if maximizing individual human liberty was the basis of the program, not a utilitarian notion of maximizing net-human life.

Caveat: even with human liberty as a basis for developing a political system, you could still end up rationalizing mandatory, subsidized health-care i.e. maximizing freedom entails a poor person shouldn't have to lose his liberty to poor health etc. This is my position, but I don't necessarily think it is the inevitable conclusion of attempting to maximize for human liberty.

[wbillingsley]

> That’s very unlikely. > Any logical system based on the concept that human life is worth existing on its own (no matter what the person has contributed to society) automatically ends up with the necessary conclusion that things like subsidized healthcare are mandatory.

If you start solely from that premise: "that human life is worth existing on its own (no matter what the person has contributed to society)" you're very unlikely even to reach taxation (a forced contribution to society) let alone forced subsidies or making anything mandatory. Most of government is based in the notion that someone's only value is in what they contribute to society -- from "tax-dodgers" to "benefit leeches" the vernacular is all about the amount of cash that gets paid into the social coffers.

Much as I appreciate subsidised healthcare, it's not an "automatic conclusion". It is a negotiated compromise, and largely based on nationalism not the value of the individual (eg, the NHS came into existence post-war, as part of the national rebuilding. It's beginnings very much relied on the war effort and large scale conscription having devalued individual freedom amongst the public).

It's become more popular since then because it turns out to work pretty well as a system. Not so much from pure logic, as that healthcare gets more expensive over time (effectively, healthcare can exert a rent on people's lives) and social control of healthcare is a way of putting a cap on its costs at the expense of those in healthcare who could charge much more (eg, watch the NHS junior doctors complaining about the contract changes).

[awinter-py]

if the goal is 'more human lives' you've built a paperclip maximizer.

[pdkl95]

> reboot it with a different set of facts

Not just facts - the designers of the AI also choose the underlying assumption and models. Even the very idea of using an AI implies a certain set of biases and intentions.

Characteristics that are truly common enough in humans that can safely be extracted as a factor are rare. Most of the time we try to compromise so we can call our differences "close enough". The process of finding and/or creating those compromises is what we call "politics", and while computers can certainly help as a tool, but the process needs human involvement by definition.

Attempts to turn over any kind of political or social decision-making to an algorithm is simply a way to disguise the concentration of power. The algorithm's designers ultimately end up with the power, while others are denied power.

The racist tactic known as "redlining"[1] is a pre-AI example. Black people aren't denied hosing directly, they simply "don't quality" for a loan, with the real reasons obscured behind a proprietary "credit worthiness" equation.

[1] http://www.theatlantic.com/magazine/archive/2014/06/the-case...

edit:

Instead of using AI as a decision-maker, a place where AI (and other technology) might actually be useful is as the facilitator and/or part of the "panel of experts" used in Delphi methods[2]. While the people tend to jump on bandwagons and make stupid decisions when unorganized, we have a lot of examples where a general crowd of people make very good decisions when they are focused on a specific goal and have enough structure to allow for iterative refining of ideas.

[2] https://en.wikipedia.org/wiki/Delphi_method

[pdkl95]

By the way - I should mention that the last part about using modern technology to as organizational structure instead of decision making as part of some sort of Delphi method is originally from a 2014 interview[1] with James Burke (Connections, The Day The Universe Changed).

A few of his comments that are relevant to the use of technology with politics and society:

    We have these extraordinarily limiting constraints from a past in which we did not have
    the tools to have anything other than extraordinarily limiting constraints. But, now we
    do have the tools, and the tools are running away with us faster than the social
    institutions can keep up.
    ...
    I think countries ought to set up Departments of the Future. [...] We are on the edge of
    having the technology to be able to say, let us run a constant, dynamic, updated review
    of everything that science and technology is thinking about [...] then let us use the same
    techniques to ask the public in general, not politicians, whether they like that idea,
    whether they feel that they could live with that idea. And then, like a Delphi technique,
    re-run it until everybody stops changing their mind.
    ...
    Collate all [research laboratories and business R&D] together and process them using stuff
    like big data to see what the pattern looks like becoming, and then layering on top of that
    social media analytics to say, if this was coming, would you like it, and if not, why not?
    In other words, to have a sort of 24 hour a day referendum

The other parts of the interview are very interesting as well:

    ... it’s no longer important to teach people to be chemists or physicists or anything ‘ists
    because those jobs are gone, and if they’re not gone today they’re gone tomorrow. And unless
    we know the old tools of critical thinking and logic and such, we will not be able to handle
    what follows. So, we’re wasting our time training people to be things that will no longer
    exist in 10, 15, 20 years time.
    ...
    Every single value structure is meaningless [...] commercial society will be destroyed
    at a stroke. The trouble is the transition period [...] how we get from here to there.
    The vested interests, I mean, we’re going to have to shoot every one of them – nobody,
    nobody is going to give way to this. [...] All cultural values relate to scarcity, ultimately.

[1] http://youarenotsosmart.com/transcripts/transcript-interview...

[DominikR]

> there is an invisible group of people behind the president, pushing them to act and say things that they want.

It is absolutely normal to have an establishment and an elite that influences the governments and the presidents decisions.

Only in the most dictatorial and absolutist types of governments you would see a lone person at the top deciding on issues without consulting with anyone.

But I'm not sure that such a thing has ever existed. Even maniacs with a cult following behind them like Hitler had to balance the interests of different factions within the system.

What we have to get rid off is certain categories of influencing decisions that have a negative impact for the majority. (bribery)

[conceit]

That's nonsense. The very idea of dictatorship is to dictate, i.e. to prescribe the rules. It doesn't matter whether that is given from any number of individuals directly or channeled through a representative.

[roymurdock]

Reminds me of how people think Bitcoin/blockchain tech can solve a lot of the inherently human problems in finance. At the end of the day, it's the group of core developers who change the codebase that are truly in power.

[gruez]

That's not a really good comparison. You can always make your own fork of bitcoin, and if there's enough like minded people, your fork will "win".

[roymurdock]

It's the same with governments and political parties. One group wants it done one way, another group wants it done a different way, and so two parties, styles of governing, blockchains emerge.

My main point though is that we tend to place too much emphasis on how much new tech can help solve age-old, human nature driven problems.

[nitrogen]

Our first step should be replacing first-past-the-post voting with something that isn't mathematically guaranteed to result in a polarized two-party system.

[r3dk1ng]

something like https://en.wikipedia.org/wiki/Instant-runoff_voting?

[nitrogen]

IRV has the problem where gaining support can actually cost one the election (nonmonotonicity). From http://zesty.ca/voting/sim it seems that approval voting would work reasonably well.

[varjag]

I guess in the same way everyone in the United States can run for president, and with enough like minded people, win the elections.

[speeder]

I play online the game Paranoia.

The game backstory is that during cold war, it became a hot war, nukes were launched, and in the US was built one (or more) underground cities, that are administrered by a paranoid computer that hates and fear communists.

My current Paranoia character works on "TechServices" and sometimes make his opinion known (something that is actually illegal to do) that "The Friend Computer" ins't the real ruler, his designers and programmers are.

It is fun to see the implications in the game, specially as the mindset of players affect their characters and behaviour, some people are loyal to the computer, some people consider the computer the enemy, and some people consider the computer only a tool, and are loyal (or enemy) of the "High Programmers" that have access to the administration computer code.

[tariqali34]

I've ran a Paranoia campaign where the players were all High Programmers. Technically, the players could reprogram The Computer, but they usually refrain from doing so, partly out of fear of offending all the other High Programmers and partly because trying to reprogram The Computer require a skill check (with failure causing more bugs to be introduced into the system). Any programming that they do is rather subtle, to avoid raising alarm.

The end result is a rather byzantine situation where everyone (including The Computer) is plotting against everyone else, all being sufficiently paranoid.

[maxander]

Political decisions made via artificial reasoning have the advantage that it can keep a log of all the steps leading up to any given decision. Then, even if this log is too long and complex for humans to handle, AI run by civilian interest groups could inspect it to ensure that everything was done logically and proceeding from appropriate principles- interference from outside sources would presumably appear as conclusions made without rational basis. A whole new level of government transparency!

Granted, it may not work out so rosily in real life- the reasoning log would likely be liberally redacted due to factors classified from public knowledge (just as some of Obama's more puzzling stances may, charitably, be explained by things he's not allowed to tell us)- but that's its own problem.

[mtgx]

Exactly. Some people actually think this would be more "fair", but we can't guarantee an AI would be "fair" for the same reason we can't guarantee that online voting would be safe. Others could easily manipulate it, and not necessarily people from the same country either.

Having AI's as advisers, that would be totally different. In the end we can still hold responsible the people that listened to that advice as it's their responsibility to check if the advice from their AI advisers is real.

[zanny]

You could guarantee online voting with distributed trust and proof of stake. There are several papers on it now that are a good search away.

That doesn't change the AI, though - since it still is physical hardware in some place, even if you could verify the software you can still exploit the hardware.

[pault]

> In the end we can still hold responsible the people that listened to that advice as it's their responsibility to check if the advice from their AI advisers is real.

This has so much potential for a Phillip-K-Dick-ian rabbit hole of paranoia and insanity that it made me chuckle. I think you might be hand-waving away a lot of complexity here. :)

[babbish]

Would you have the same qualms if the AI's source were open and the public voted on which pull requests to accept?

[pjc50]

That sounds even worse. I wouldn't trust binding public voting on pull requests on my text editor; I'd expect it to break down fairly rapidly.

[babbish]

Assign negative weights to people who submit damaging pull requests?

Insert a layer of expert representatives (akin to congresspersons) to generate the PRs?

Making the system resilient doesn't sound like anything more than another engineering problem to me.

[Cyph0n]

The nuances of politics cannot be boiled down to a Github repository. It's not as straightforward as you think it is I'm afraid.

[babbish]

Who said anything about GitHub? The situation would call for a highly specialized, custom-made solution. If you think that's impossible, please share your reasons for thinking so. Comments like yours are pure FUD and aren't in any way helpful.

[stefs]

well, how to identify "damaging" pull requests? i mean, that's what politics is.

[mclovinit]

I take it the definition of public here is inclusive of all citizens in which case, the majority of citizens should understand the fundamentals of AI and the given software implementation. Many citizens are fluent in English and for those who are not, such individuals often confide in trusted sources to translate for them prior to making decisions. I am not sure if this approach of publicly voting on AI would work, but then again many people today do not fully understand each candidates political agendas yet they still end up voting. So I could be wrong.

[privong]

> Would you have the same qualms if the AI's source were open and the public voted on which pull requests to accept?

I interpreted @mtgx's statement as referring to the information that would be input into the IA. Namely that if presidents are getting bad advice from advisors already, replacing the president with an AI might not improve things, if the same bad advice is still being fed in.

[babbish]

If that's what he said, then I agree wholeheartedly. "Garbage in, garbage out" is pretty hard to argue against.

[binarymax]

When it comes to AI, source is only part of the system. The trained model would be a black box. We don't fully understand why some NNs work - so something of this complexity can't simply be cracked open and vetted accurately.

[babbish]

Barring military secrets, perhaps, there's no reason the AI program's current state would necessarily have to be a black box. Nor would it need to be a neural network. I also disagree with the assertion that complex things cannot be vetted.

[varelse]

In that case, one might as well boot up the President Dwayne Elizondo Mountain Dew Herbert Camacho Bot...

[akerro]

> IIRC they did that after they fed it the urban dictionary and it started swearing too much[1]

This was literally the best usage of AI as comedy ever made.

[Outdoorsman]

The old joke, thrown out from time to time by politicians and pundits, to describe a lack of quality candidates,is that:

"Anyone smart enough to be a good president wouldn't want to be president."

This, because of the pressure and stress involved...

Watson, with human advisers, should be relatively stress-proof...a possible plus...

I guess then the joke would evolve to: "Anyone smart enough to be a good adviser wouldn't want to be."

[sp332]

It isn't about stress, it's about the greed for power. Anyone who wants that much power definitely shouldn't be trusted with it.

[stray]

I'm sorry, Dave, I can't do that.

[conceit]

> The power actually lies with the person that was elected

... unless there's leverage against the elected, something that would subjectively seem to compromise the integrity more than the slowly gotten used to corruption.

[rajacombinator]

This has pretty much been the case for decades if not longer and yet it is still not apparent to people! AI represents the people, hope, and change, surely.