Elon Musk Thinks We Will Have To Use AI This Way To Avoid a Catastrophic Future (cnbc.com) 110
Elon Musk has long said that artificial intelligence will have to augment human abilities, rather than compete with them, in order to avoid a portentous future. He has been active in trying to find ways to evaluate and reduce potential risks posed by AI. From a report: On Monday, Musk tweeted out a set of principles for AI research and development created by a group of scientists at a recent conference for the Future of Life Institute (of which Musk is a board member). Musk said in response to a comment that ensuring AI augments human abilities is "critical to the future of humanity." Musk recently told a Twitter user that there may be an announcement "next month" regarding such as device, which Musk has called, in the past, a neural lace.
Title is worded like clickbait (Score:5, Insightful)
Re:Title is worded like clickbait (Score:5, Funny)
You won't believe how the Slashdot community responded!
Re:Title is worded like clickbait (Score:4, Funny)
Re: (Score:3, Insightful)
Re: (Score:2)
Seriously, we know when we're being clickbaited. Knock it the fuck off,
Kind of naieve statement for him to make (Score:2)
Since its only down to interpretation where augmenting human abilities ends and replacing them begins.
I mean if AI really wanted to take over, they can just augment us sufficiently until there really isn't a clear line between humans and AI, then take it from there.
In fact pretty similar to the "embrace, extend, extinguish" strategy that Microsoft use(d) to make themselves into a megacorp, and now to just keep themselves there.
Re: (Score:2)
Re: (Score:2)
I'll take Transcendence over Watson any day of the week
Re: (Score:3)
>> not anywhere near human-level ...yet...
>> is not conscious,
being conscious is not a prerequisite to being a real threat to humans.
>> You can't sit down and have a conversation with the damned things
Actually you can: Siri, Google Home, Amazon Echo to name but a few already available as products. Admittedly all are currently very rudimentary and not about to do a SkyNet anytime soon, but you can bet they will only get more and more powerful over time. There's also already much better/str
Re: (Score:2)
Even relatively simple AI can be very useful. For example, a cat can provide companionship and emotional support, despite having quite a simple mind. Cats don't understand cause and effect, they only learn by positive and negative experience. They don't even have object permanence, so when they see a toy they don't think "that's the same toy I had yesterday", they just know it's a toy and it smells like them so it must be theirs.
I saw on TV that Japanese scientists are experimenting with robot pets that can
Re: (Score:2)
Okay, now try to program a cat, if you think they're examples of simple intelligence. Rodney Brooks thought that the cat part of AI was horrifyingly difficult and the human part on top would be relatively easy, and I think he was correct there. (His approach of using machines with discrete behaviors modeling housefly-like behavior turned out to be a dead end, though.) I really doubt a machine with current AI would work nearly as well for companionship and emotional support, which (at least to me) seem t
Re: (Score:2)
>> I really doubt a machine with current AI would work nearly as well for companionship and emotional support,
Apparently they're having very positive results with robots for companionship in old peoples homes. I think it was Japan, I'll see if I can find the link.
Re: (Score:2)
To be precise, I think I'd need signs that I'm dealing with a being with an independent mind that seems to like me. I find the idea of a robot companion creepy at this stage of AI, and still don't like being told "Thank you" by a machine, but I'll happily use automated assistance for doing things.
Re: (Score:2)
Neural lace, eh? (Score:1)
Re: (Score:1)
Re: Neural lace, eh? (Score:1)
Re: (Score:2)
~ That is a neural lace, it informed her. ~ A more exquisite and economical method of torturing creatures such as yourself has yet to be invented.
What could possibly go wrong?
They could come up with neural lice?
The very idea makes just under my skin crawl...
(in Harry Potter World, a curse called "dermis squirmis")
Re: (Score:2)
I think you might have meant "Couture". But will it be bespoke or prêt-à-porter?
Re: (Score:2)
Being a Ship Mind with an obliquely insulting name is the best part of that future. I'd want to be one too, maybe GSV Try Pushing the Other Button.
Re: (Score:2)
A cautionary tale illustrating some of the associated risks of this sort of augmentation is Lois McMaster Bujold's novel Memory. One of the major characters, decades previously to the story, had had a mnemonic memory chip implanted, which gave him perfect memory and recall. During the course of the story, the memory chip is damaged (which throws the character into disjoint fugues as it begins to kick back memories randomly) and later removed, after which the character discovers that after decades of the chi
How about these? (Score:2)
Re: (Score:2)
Re: (Score:2)
Accidentally moderated "offtopic" instead of "insightful" - posting here to undo moderation.
Re: (Score:2)
Re: (Score:1)
A robot may not injure a human being or, through inaction [etc]
Don't know whether you were intentionally suggesting these at face value or tongue-in-cheek/ironically making a point.
People really seem to forget that the whole *point* of a lot of the "three laws" stories wasn't the laws themselves, but to show that such apparently clearly-defined laws had unforeseen consequences that showed they couldn't be relied upon in the way intended.
*Incredibly* perceptive for a guy writing during the 1940s, but the reality we have today suggests it would be even harder in practice
Re: (Score:2)
Nice, but these are in a form that requires human-type intelligence to comprehend, and Asimov's robots have basically human intelligence with restrictions like this. We can't do human-type intelligence or anything near it. As an example, First Law requires that a robot understand what is a human (in one of the Lucky Starr novels, an enemy attempted to convince robots that Lucky's sidekick was not a human), what harm is, and requires a sophisticated universe model to calculate whether a given action would
Portentous! (Score:2)
You keep using that word... I do not think it means what you think it means.
Really - look it up... Although the use of it does bring to mind its third definition from the American Heritage dictionary: Marked by pompousness; pretentiously weighty.
Re: (Score:2)
I thought the same thing, but look-y here.
Paranoia will kill it (Score:2)
I agree - AI's strength is with details (Score:5, Insightful)
I develop marketing automation software. Lots of people talk about 'set and forget' on their marketing, where the AI takes care of everything. This is a really bad idea. Every time I've tried to do something like this it has backfired with the AI doing something spectacularly stupid.
Instead where I have had success is where the AI's role is to fill in the little details that would be boring for a human. Essentially the role of the AI is to tune rather than create. For example, the human might craft the first couple emails and then leave the AI to start moving sentences around for better effectiveness. It is completely unrealistic for people to craft the best message for every single person on their database, but it is perfectly reasonable for a person to produce the first ten or twenty and then leave a computer to fill in the gaps.
Similarly, feedback from the AI needs to go back to the human so they can provide guidance. For example: "the content was not very effective with this segment", and the human provides more training data on how to communicate with people who fall into that segment. I think about it as giving power to the human - adding richness and fine-tuning to all of their decisions. The AI is never in control. Even if almost all the decisions are made by the AI, it is always within the guidance provided by the human.
Maybe this will change one day; at the moment AI sucks at extrapolating but is awesome at interpolating. This means a human is going to do a far better job of setting strategy, but will quickly lose interest if they have to do every micro-execution.
PS: What's up with the article title. How about 'Elon Musk believes AI needs to augment humans instead of replacing them'?
Re: (Score:2, Funny)
I develop marketing automation software.
Kindly die in a fire, please.
Re: (Score:1)
I second this.
Re: (Score:3, Interesting)
I develop marketing automation software.
Kindly die in a fire, please.
LOL!
You don't like interacting with companies that try to understand you? You'd rather receive the same generic product offers as everyone else?
I just don't believe this. I believe that the more a company tries to make their communication relevant for each customer, the more value the customer gets from the company. Incidentially I don't know if you've thought about the impact of AI personal assistants on marketing. If say Walmart sends you an offer that your personal assistant believes won't interest y
Re:I agree - AI's strength is with details (Score:5, Insightful)
What does drive me to going to a different competitor are advertising / marketing methods I find any of the following: invasive, absurd, immature, over the top, lacking in class, offensive, over budgeted, unethical, and others. I vote my pocketbook against vendors I dislike rather than for a particular or special one.
I don't need product offers in the first place. The idea that I need to be offered products that I am not already researching prior to purchase is intellectually insulting, and part of the the consumerism + marketing driven problem. You remove value from the interdependent system of producers and consumers and increase the cost of products.
Re: (Score:2)
It's not really my field, but what do you think of Hubspot and their inbound marketing? The basic idea is that you help the consumer with their research - trying to point them towards content that you think is what they need to know to make a decision. Obviously you want them to choose you, but the reason I'm bringing it up is that you're selecting information to help inform your decision rather than simply saying 'this product exists'.
I'm not going to comment on your list of invasive, etc as it's just no
Re: (Score:2)
A hundred times this.
Re: (Score:2)
What does drive me to going to a different competitor are advertising / marketing methods I find any of the following: invasive, absurd, immature, over the top, lacking in class, offensive, over budgeted, unethical, and others. I vote my pocketbook against vendors I dislike rather than for a particular or special one.
Damn, take this down! He's just giving out these amazing ideas for free!
Re: (Score:3)
You don't like interacting with companies that try to understand you? You'd rather receive the same generic product offers as everyone else?
I just don't believe this. I believe that the more a company tries to make their communication relevant for each customer, the more value the customer gets from the company.
Yes, I would prefer to receive the same generic product offers as everyone else—the less relevant the better. That way, I am less likely to be subtly influenced against my will and contrary to my own interests. I'd prefer to avoid these ads entirely, but to the extent that is impossible, by all means, fill the ubiquitous ad spots with products I will never buy and services I will never require.
It's a different matter when I'm in "shopping mode" searching for a specific product. In that case relevance
Re: (Score:2)
There's a company called OpenDNA (as oposed to Elon's OpenAI) which is trying to develop an understanding of what everyone likes. Their idea is that they will expose it to you and you'll be free to make changes. I.e. it's your profile, and you get complete control. They'll then sell the ability to match products to customers based on the data they look after.
I have my doubts about whether they'll succeed. I'm not sure consumers are quite ready for being told where they sit on the premium to budget conti
Re: (Score:2)
Re: (Score:2)
Yes, you can always get past once, but it's a dumb thing to do. I have that conversation frustratingly frequently - we build up a good base of highly engaged customers because we've consistently delivered something relevant. Then some stakeholder wants to push a particular product and has enough political clout to overrule me (the consultant).
In some ways their offer will succeed - it will generate more sales than if I had my way and only went to the people who cared about it. The damage it does to your
Re: (Score:1)
Just get rid of the scam ads and the javascript bombs and I'll be happy enough.
Re: (Score:2)
You don't like interacting with companies that try to understand you? You'd rather receive the same generic product offers as everyone else?
Yes, that would be great.
A while back I was looking for an otoscope, a device for looking inside the ear. Unfortunately the cheaper ones come with a variety of attachments so that they can be uses for examining other orifices, including the anus. I made the mistake of being logged in to the site I was searching on, and now every time I go there it offers me a wide selection of products to shove up my arse.
Had to burn that account.
I try hard to protect my private data, which includes my preferences and shopp
Re: (Score:2)
That's a brilliant anecdote, do you mind if I use it?
I understand what you're saying. There's a lot of truth to it and there's a fair number of people who feel the same way as you. I still feel we're better with more personalisation than less, but you are absolutely right that there's an uncanny valley as the computer starts to understand you.
A few years ago I worked for a very large loyalty program where I built customer personas. One of our participating companies was a clothing retailer. Big brand, n
Re: (Score:2)
Thing is, I expend a fair bit of effort trying to avoid the computer getting to know me and in poisoning marketing data. Personalization is creepy, I don't know these computers or these companies, I don't want them tracking my habits and interests. I worry that the data will be misused, and try to block as much advertising as possible because what I do get is often embarrassing, inappropriate or malware.
Re: (Score:2)
Kindly die in a fire, please.
"Did you mean: 'Kindle die in Amazon Fire, please.'?" -- your marketing automation software.
Re: I agree - AI's strength is with details (Score:1)
The computer program. Stop saying AI you fucking moron. That is not AI.
Re: (Score:1)
Re: (Score:1)
With someone like the GP, the definition always boils down to "whatever hasn't been invented yet". This obviously excludes anything he previously counted as AI that was then invented. If it becomes real, it stops being AI.
Clearly the answer is to use cats. (Score:1)
I've had exactly the same problem with humans!
Re: (Score:3)
Re: (Score:2)
I've been trying for a number of years, unfortunately with very little success. Even with clients I still get people expecting the computer to extract the topic of an email, but when I describe an approach of say TF/IDF they claim that is not AI and they don't like it. I suspect I'm better at implementing it than selling it.
Re: (Score:2)
- Minimax type AI has spectacularly good micro, but sucks at macro. E.g. chess AIs, or see minimax used on RTSes [arxiv.org] - quote: " RTMM plays perfect short term micro-scale game, but plays a very bad high-level (long term) strategy
- UCT type AI has somewhere between consistently poor and consistently average play on both macro and micro, see e.g. Go programs prior to AlphaGo.
- Neural net AI has good macro but suck at micro. See AlphaGo: as long as it coul
Re: (Score:2)
AlphaGo is really interesting.
We use deep belief nets for product selection (with thumbs up/down as feedback). So far it has been pretty good at selecting individual products but the overall email is a poor representation of all products on offer. Contrast that with AlphaGo which played stong fuseki across all of its games. Perhaps like AlphaGo we need to bring in another layer and train that on 'any purchase' rather than positive engagement with the email.
The other place we don't do so well is that we h
Slashdot thinks the title (Score:2)
belongs on buzzfeed. You won't believe how the Slashdot users responded!
Laptop batteries! (Score:1)
Yeah (Score:2)
More inspiration from Iain M. Banks (Score:3)
The Neural Lace is yet another concept from the Iain M. Banks Culture novels. So I guess it's clear at this point that Elon's a sci fi fan, and a fan of Banks in particular. If the names of the drone ships USS Just Read the Instructions and USS Of Course I Still Love You weren't evidence enough, this seals the deal.
Interestingly, in his book Excession, it's mentioned that along with being a direct mental interface to computers, the Neural Lace is the most effective means of human torture ever devised...
Competing is augmenting (Score:2)
Elon Musk has long said that artificial intelligence will have to augment human abilities, rather than compete with them, in order to avoid a portentous future
But programming an AI to do things I don't want to do is augmenting my human abilites, and competing with the human who I normally pay to deal with it.
As an example: I can't possible cut my lawn more efficiently than a heavily-invested landscape crew: they charge much less than I could theoretically make in the same time, and paying them frees up multiple hours for me to do other things.
And as soon as a sufficiently talented lawn robot becomes cheaper, they'll be gone and it will also free up my time and li
Peter Watt's Blindsight (Score:2)
Or we could end up like the soldier character Amanda Bates in Peter Watt's Blindsight [rifters.com], a human inserted into a network of AI driven machines, each quicker and more lethal than her fragile human self. Yet she has the final say and utmost authority over the decision to kill.
The unfortunate implication? Her AI team becomes far more dangerous once the slow-thinking and squishy human dies, and they get let off the leash. Meaning that she has as much to fear from her superiors as from her enemies.
A.I.Monitization (Score:2)
Re: (Score:2)
Amusing for Musk to say that... (Score:2)
Given the giant push for fully autonomous cars, it's pretty fun for him to say the only safe use of AI is augmented systems... if that were true you would just use AI to augment driver inputs instead of taking over from them entirely. That's not a bad idea, but that sure is not where Tesla is going.
An AI augmented car would be one that would see you were trying to veer around something and enhance your attempted steering input to succeed if possible - including putting the car up on two tires...
For once, a non-destructive AI use. (Score:2)
Unlike Watson, it helps humans do their work without the unnecessary destruction.
Competing with AI (Score:2)