Reflections on the State of AI: 2018 | Hacker News Reader

Reflections on the State of AI: 2018

evolutionone.ai

57 pointsAlexDStern18 comments

Some points : duplex... the verge article linked (https://www.theverge.com/2018/12/5/18123785/google-duplex-ho...) is very interesting about the constraints on the roll out and also the work that is required to get duplex to do what you want it to do; including covering off contingencies (like unavailability) that may not arise. This means it's actually more work to get duplex to do an average reservation than it is to do it yourself.

Also it's evil (disinter-mediating people preventing human contact, gathering more data, closing options, preventing plurality of opportunity), but never mind ehh?

Second point; they aren't "self driving cars" they are not driving driving is a human thing that requires agency that no AI has at the moment. Autonomous vehicles are environmentally responsive guided robots or something like that. This confusion is important because many people see something that is "self driving" and impute many characteristics and capabilities to "it" that "it" just doesn't have; and this colours their expectations about what it can and can't do and what autonomy and responsibilities to ceed and delegate to that thing.

Most perniciously both Duplex and Autonomous Vehicles are parts of systems, they are not "your assistant" they are an interface to a collection of machines and people who's function is not to help you, but instead is there to enrich someone else and constrain your ability to choose and act as you would if you didn't interact with them.

Stalin would have loved this - the citizens go where they are taken, they can't even think of an alternative. They choose the food we say they can. The pictures that they take a manipulated and managed by us.

At any time your photos of your friend who's no longer in favour might morph over night to be landscapes only, or to contain just the friends who remained loyal to the party/government, you might wonder what's happened, but how will you know?

Hacker News should require anyone who downvotes to post a rebuttal.

I don’t think things are quite as distopian as envisioned above but would love to see the counter argument.

Yeah - it's more that I think that the potential for misuse exists, and the harm could be very great.

environmentally responsive guided objects, ERGO

It's always possible to imagine bad and worse scenarios. Fear is always available, and can be applied to any situation. Believing that technology is the tool of an "evil" system, or that it facilitates "evil" is not healthy for the believer or anyone else.

Naturally, we are all aware of the power of tools and especially modern "smart" systems. How can we trust anything? We can imagine how things could be as bad as we can imagine! I've decided, even at the risk of being naive, that I must not fear the machine, fear the network, fear anything. We living beings all basically want the same outcome- happiness. Having convenient tools help us actualize that outcome is a good thing. Just because we can imagine some deep dark conspiracy of "evil", behind the scenes, deceiving and manipulating us, doesn't mean that there is any such thing. Nor, of course, that there isn't. Since we can't know, let's just enjoy all of our cool gadgets while being happy...

Although, I'm also beginning to feel as though the best decision will be exiting the information super-highway :)

If y'all haven't read it, The Joy Makers is a pretty fun sci-fi exploration of technological routes to happiness, and possible downsides to "plugging in".

I'm of the view that Google, Facebook (et-al), many governments and assorted other groups are not evil, not deliberately in any organized way at any rate. But, the mechanisms that are being developed have huge potential for very insidious application, and it's important that we get ahead of this as soon as we can.

It's really depressing that a lot of machine learning tasks can now be performed without much programming knowledge and even less knowledge about machine learning...

Why?

Why is that depressing?

I use heaps of tools I probably couldn't create myself, or at least don't have the gumption to try.

It's interesting though, you can do most (all?) statistics using R with very little (or almost no) programming, or knowledge of statistics.

I can use power tools, but you sure as hell don't want me building your extension.

Yup, this is what I think, trouble ahead I think!

abstracting the complexities away and making the technology more accessible is a good and inevitable thing. Do you think academic gatekeeping and elitism have more value? Doesn't mean that there is no room for experts with deep understanding though.

Replace "depressing" by "dangerous in the case of certain applications" and we can agree: https://en.wikipedia.org/wiki/Weapons_of_Math_Destruction

Did you mean "exciting"? A few steps of matrix multiplication shouldn't require extensive programming knowledge. Backpropagation is intuitive to someone who already understands algebra and calculus. It doesn't take much knowledge. Creating useful models still requires real practical skills, especially for high confidence levels in complex datasets.

It's exciting unless you have invested considerable time in learning data-science and machine learning from the ground up and want to make money from it.

Yes, it still needs that expertise to run models well. But those without the expertise sound more confident and then ruin the market with the expectations they elicit.

If the problem is standard to the point where you won't encounter any weird edge cases (e.g., runs analogously as it does in a tutorial on the internet) , then you don't need programming or machine learning to perform the tasks.

But once you want to extend a problem into an unfamiliar domain or debug why an algorithm is not working, then I don't see how you can accomplish this without domain-specific and machine learning knowledge. Or have I been missing out on some amazing auto-debugging AI that I am unaware of?

Depressing why? Oh I know. Because you are an elitist who believes that what can come from AI should only belong to those who can code it?

Or because your own lively hood is dependent on machine learning?

Whatever the case may be and my response is that I don't care. I firmly believe that this is a GOOD thing.

As developers we should be striving for a) open sourcing machine learning so that everyone can contribute to the good it can do and b) that we democratise the result of those efforts so that everyone can benefit from it for their own projects.

When time goes forward these efforts will translate into an NPM of machine learning. What would have taken a small team and a year to do, now just takes installing a plugin and having an API at the developers disposal. This is definitely a future that I cannot wait to see and it's within reach.

Out of the 3, AN (Narrow) I, AG (General) I and AS (Super) I. Artificial Narrow Intelligence is going to grow in prevalence in the next 10 years and the effects will be very good. I don't think it will kill jobs, merely make people much more productive.

> As developers we should be striving for a) open sourcing machine learning so that everyone can contribute to the good it can do and b) that we democratise the result of those efforts so that everyone can benefit from it for their own projects.

Why? Oh I know. You are a wild-eyed optimist who would have cheered unfettered globalization in late 90's because "the whole can benefit from it".

You know what - as developers, we should care about maintaining scarcity of developers around the world so that we can continue being in demand and continue having great salaries. I would rather be a part of AMA (cartel which limits supply of Doctors in the USA and erects legal/regulatory barriers for competitors, thereby ensuring high standard of life for Doctors) than being a manufacturing engineer whose job was lost (or threatened) first by globalization and now by automation.

You know who benefits from "democratization of results of our efforts" or when "everyone can contribute to the good machine learning" (or other software engineering in general)? VCs and mega-rich folks, whose capital now doesn't have to share productivity gains with labor because labor is abundantly available.

Reply on news.ycombinator.com