[Technocrat]

You guys are all wrong, Skynet is real and it has started spreading its roots

Jokes apart, I too believe what Elon Musk has been advocating about AI i.e. regulating AI before its too late.

As for the FB AI Bot shutdown, here is an interesting take by Gizmodo which at least sounds less worrisome than the other articles, not sure if it is is factual or just written with an intent to ease the alarm that it has caused.

In recent weeks, a story about experimental Facebook machine learning research has been circulating with increasingly panicky, Skynet-esque headlines...

The reality is somewhat more prosaic. A few weeks ago, FastCo Design did report on a Facebook effort to develop a "generative adversarial network" for the purpose of developing negotiation software.

The two bots quoted in the above passage were designed, as explained in a Facebook Artificial Intelligence Research unit blog post in June, for the purpose of showing it is "possible for dialog agents with differing goals (implemented as end-to-end-trained neural networks) to engage in start-to-finish negotiations with other bots or people while arriving at common decisions or outcomes".

The bots were never doing anything more nefarious than discussing with each other how to split an array of given items (represented in the user interface as innocuous objects such as books, hats and balls) into a mutually agreeable split.

The intent was to develop a chatbot which could learn from human interaction to negotiate deals with an end user so fluently said user would not realise they are talking with a robot, which FAIR said was a success.

When Facebook directed two of these semi-intelligent bots to talk to each other, FastCo reported, the programmers realised they had made an error by not incentivising the chatbots to communicate according to human-comprehensible rules of the English language. In their attempts to learn from each other, the bots thus began chatting back and forth in a derived shorthand — but while it might look creepy, that's all it was.

Facebook did indeed shut down the conversation, but not because they were panicked they had untethered a potential Skynet. FAIR researcher Mike Lewis told FastCo they had simply decided "our interest was having bots who could talk to people", not efficiently to each other, and thus opted to require them to write to each other legibly.

Read the full article - Source

[navin]

Originally Posted by Samurai

Can your agent bots handle voice, or do they just deal with text based queries?

If they can handle text why not voice? If Siri can why not more sophisticated applications?

I am looking forward to the day that AI enabled applications can make computers what I want to do instead of what I tell them to do (there is often a difference).

mazda4life, why are "bot frameworks" only useful for "routine, repetitive tasks"? AI would mean that even tasks that require some level of "learning and deduction" should be within their area of ability.

Ref: "In their attempts to learn from each other, the bots thus began chatting back and forth in a derived shorthand"

So these AI bots did nothing that our kids do not do with their SMS lingo and abbreviated spellings. Cool.

I think FaceBook should explore "bots that can efficiently talk to each other". Then CEOs can say "I'll have my bot talk to your bot" instead of "I'll have my secretary talk to your secretary". And you don't have to give your bot a raise or even coffee breaks.

[mac187]

Originally Posted by navin

If they can handle text why not voice?

They definitely can, we have integrated Alexa with our in-house chat bot. They work easily. There is no point in investing in developing voice recognition systems anymore. Anyone can easily piggyback on Apple, Google or Amazon engines.

[im_srini]

Originally Posted by navin

So these AI bots did nothing that our kids do not do with their SMS lingo & abbreviated spellings. Cool.

That's the coolest way I've seen anybody put this.
I absolutely detest SMS lingo.

Originally Posted by navin

If they can handle text why not voice ? If Siri can why not more sophisticated applications ?

Text is cleaner, with voice you get pronunciation & accents, besides Siri is proprietary.
That said, there are now voice-recognition libraries that can be used to do most of the heavy lifting.

Originally Posted by navin

Why are "bot frameworks" only useful for "routine, repetitive tasks" ?
AI would mean that even tasks that require some level of "learning & deduction" should be within their area of ability.

That's the state of open AI libraries, or frameworks, currently, they're equivalent to a bundle of a few hundred thousand neurons.
They can, literally, be trained quite easily to do a specific pattern recognition task extremely well.
Sort of like this...

https://www.youtube.com/watch?v=b5xpXecR3LY&t

Originally Posted by navin

I am looking forward to the day that AI enabled applications can make computers what I want to do instead of what I tell them to do, there is often a difference.

Yes, there is, & a time may come when AI do what they want to do instead of we want them to do

Originally Posted by navin

Then CEOs can say "I'll have my bot talk to your bot" instead of "I'll have my secretary talk to your secretary".

Secretaries ? Imagine a day when we do not even have to go to work !

Last year, I had the opportunity to work with an AI startup that's working on some cool tech.
The basic idea was to use their proprietary AI to learn to play an RTS game ( like StarCraft ) & then use the strategies they learnt to manage other adversarial resource-management tasks ( for example, the electricity grid ).
We got 2 AI bots to play StarCraft against each other & within a week & change, each was a formidable opponent ( to a regular human player ).
When we realised that we were no match to the bots, we brought in a "Twitcher" ( the gaming term ).
But once again, it was simply no contest, the bots had played a life-time of games in 10 days & the smelly, geeky, 21-year old BoW guru was simply no match to the machines.
After losing many, many, many sessions the poor despondent soul told us in exasperation that the bots were using weird strategies he'd never seen before.
The bots even learned sacrificial strategies on their own, it was a sobering moment for everyone in the team.

Once you start layering these simple AI tasks on top of each other, the end results can be quite spectacular, not to mention scary.
This is just the beginning...
.

[rubicon]

i have been following this subject (AI) for close to 2 years now and have had a chance to work directly on implementing a pilot project in my workplace involving Chat Bots. I had to provide a headcount reduction plan as and when the chat bot's got intelligent over time via learning.

Interestingly, the 2 brightest minds (with the deepest pockets who have also invested heavily in AI) have a completely different views on the safety aspect...

Elon Musk : AI needed to be regulated because it poses a “fundamental risk to the existence of human civilization”. I have exposure to the very cutting edge AI, and I think people should be really concerned about it. I keep sounding the alarm bell, but until people see robots going down the street killing people, they don’t know how to react, because it seems so ethereal.

Mark Zuckerberg : “I have pretty strong opinions on this. I am optimistic…. And I think people who are naysayers and try to drum up these doomsday scenarios — I just, I don't understand it. It's really negative and in some ways I actually think it is pretty irresponsible.”

[alpha1]

Originally Posted by im_srini

Once you start layering these simple AI tasks on top of each other, the end results can be quite spectacular, not to mention scary.
This is just the beginning...
.

To think of it, human beings (the programmers) cannot think of each and every check and balance and incentives to program the AI. Which means we cannot really predict how the end result being will "behave".

[srgntpepper]

Originally Posted by alpha1

To think of it, human beings (the programmers) cannot think of each and every check and balance and incentives to program the AI. Which means we cannot really predict how the end result being will "behave".

We Human beings like to think of ourselves as special. That we call other intelligence "artificial" is symbolic, implication that our intelligence is natural or innate.

On a very basic level, we are complex chemicals that get inputs from 5 senses, and then those chemicals interact to give us better understanding of our surroundings than other living creatures. This creates the notion of us being intelligent. Fact is, our inherent tendency is to go with instinct or herd mentality because thought is costly. We operate largely on biases, habits and often defy logic.

Given that machines have endless computing power and thought is no cost, They will become capable of taking in more inputs, communicate better, and compute faster. It is incredibly naive of us to not be consider that the machines will be able to match our capabilities, and infact, go far beyond. It only depends on what objective function or intent they are programmed to follow. If self-preservation becomes a learned or programmable intent, it'll be difficult to shutdown or control.

[msdivy]

Inspired by this thread, I watched 'Colossus: The Forbin Project'. Rating: 4/5. Highly recommended

Link: http://www.imdb.com/title/tt0064177/

[ramprakashr]

As some one who is into AI/ML, I thought I would share my perspective here. It's going to be a bit lengthy though...

So typically, a programmer writes algorithms to solve real life challenges. Every algorithm, no matter how complex can be reduced to just three operations : AND, OR and NOT. But building fortresses of layers around these operations tickles the complexity monster and he comes waging war against the programmer. Enter AI, We can think of AI as the inverse of programming, in the same way that the square root is the inverse of the square, or integration is the inverse of differentiation. Just as we can ask “What number squared gives 16?” or “What is the function whose derivative is x + 1?” we can ask, “What is the algorithm that produces this output?”. Basically, given a problem, a machine learning would spit out an algorithm/program to solve it. So does that mean a Machine can replace all of us? Well relax, they're not there yet...

To put it in simpler words, We know how to drive cars and decipher handwriting, but these skills are subconscious; we’re not able to explain to a computer how to do these things. So instead, you give multiple examples of the above operations and the computer tries to learn patterns from it. Some learners learn knowledge, and some learn skills. “All humans are mortal” is a piece of knowledge. Driving a car is a skill. In machine learning, knowledge is often in the form of statistical models. Skills are often in the form of procedures: if the road curves left, turn the wheel left; if a deer jumps in front of you, slam on the brakes. (Unfortunately, as of this writing Google’s self-driving cars still confuse windblown plastic bags with deer.) Often, the procedures are quite simple, and it’s the knowledge at their core that’s complex.

As businesses grow, they go through three phases. First they do things manually, think of owners of a mom-and-pop store. They personally know their customers, and order, display and recommend items based on their tastes. But this won't scale. In the second and the least happy phase, the business would need computers. In comes programmers, consultants, database architects, sales and support engineers and millions of complex lines of code get written. After a point the programs written would fail to match the versatility of human whims and fancies. The third phase would be the Machine Learning phase, where the companies would let learning algorithms loose on the realms of data they've accumulated and let them divine what customers would prefer. Amazon can’t neatly encode the tastes of all its customers in a computer program, and Facebook doesn’t know how to write a program that will choose the best updates to show to each of its users. Yes, there's a lot left to be desired in both, but this would be the first step to it. If they try to program each and every transaction, then they would never be done.

Also, if AI would be sold in a departmental store, it's carton would read : "Just add data". Well it's data that's driving the world crazy today. Bing's algorithm may be better than Google, but you can't switch to Bing, thanks to Google's head start and the bucketful of data it has about you. Traditionally product management was all about code re-usability, feature roadmap, extensibility and so on... The latest addition to product management would be to care about what data points the product would generate, so that the AI algorithms can do their duty of making the life of the user easier. The biggest challenge, however, is assembling all this information into a coherent whole.

As one Pedro Domingos puts it in his Machine Learning book, Machine learning is just like farming. In an industrial society, goods are made in factories, which means that engineers have to figure out exactly how to assemble them from their parts, how to make those parts, and so on—all the way to raw materials. It’s a lot of work. But there’s another, much older way in which we can get some of the things we need: by letting nature make them. In farming, we plant the seeds, make sure they have enough water and nutrients, and reap the grown crops. Why can’t technology be more like this? It can, and that’s the promise of machine learning. Learning algorithms are the seeds, data is the soil, and the learned programs are the grown plants. The machine-learning expert is like a farmer, sowing the seeds, irrigating and fertilizing the soil, and keeping an eye on the health of the crop but otherwise staying out of the way. Imagine if farmers had to engineer each cornstalk in turn, instead of sowing the seeds and letting them grow: we would all starve forever...

The crux is, let's not worry about AI taking over the world and all that, it's too early. An AI system is not curious enough and it can only do tasks that it has been trained to do. There is no possibility that a self driving car is hatching a plan to kill you

[coolvenk]

The fact-checking website Snopes rated this story to be false.

Read for yourself:
http://www.snopes.com/facebook-ai-de...-own-language/

Mods: I would respectfully suggest that this story be moved out of the main page. Having factually incorrect material on the main page is not to our credit.

[Samurai]

Originally Posted by coolvenk

The fact-checking website Snopes rated this story to be false.

It does not matter. The story was merely a trigger for an interesting discussion. Members have contributed so many interesting posts, which makes the valuable by itself.

[navin]

Originally Posted by im_srini

The bots even learned sacrificial strategies on their own, it was a sobering moment for everyone in the team..

I am getting old and one of my biggest fears is losing my memory and ability to think (I have witnessed many around me with Alzheimer's for example). So in hope of preventing/forestalling mental decay I am drinking my coffee and playing interactive games. One of the games I play is called Clash of Clans. And the programmers who developed this game have somehow managed to deduce which players are using bots and ban them.

So while bots are alarming capable it is obvious that bots operate in a manner that is dissimilar to humans.

Originally Posted by rubicon

Elon Musk : AI needed to be regulated because it poses a “fundamental risk to the existence of human civilization”.

Mark Zuckerberg : “I have pretty strong opinions on this. I am optimistic….”

Maybe it is my age and/or the influence movies like 2001: A Space Odyssey, Tron, Ex-Machina and the cult classic Blade Runner have on me but I tend to share Musk's apprehensions.

Originally Posted by ramprakashr

An AI system is not curious enough and it can only do tasks that it has been trained to do.

Why can't a "learning machine" develop a "virtual curiosity"? If it can learn to deduce and predict, elementary curiosity can't be that far off.

Originally Posted by ramprakashr

There is no possibility that a self driving car is hatching a plan to kill you

You, obviously, have not met HAL (see 2001: A Space Odyssey)

I, on the other hand, prefer to hope for bots like Pris Stratton (see Blade Runner).

[Nav-i-gator]

Let me try to summarize what I understood from the brilliant discussions going on here:

1. AI can create algorithms given the specific outcome goal

2. AI can communicate with each other, in a language that may not be decipherable by humans

3. (from 1 & 2) AI can collaborate and work as one (scalability, like neural networks inside our brain,as there is no "distinct personality")

4. AI can learn on their own, objectively (no subjectivity, no head & heart dilemma)

5. Machine learning is a kind of curiosity, machines learn by experiential inputs, just like humans.

6. Since learning is experiential as well as objectively reinforced, AI is incapable of committing same mistake twice. But, it would learn to arrive at the same conclusion through different algorithms (thus capable of progressive increase in efficiency, at a much faster pace than human evolution)

7. (most important) AI is like a brain, and hence have same limitations. It can generate ideas, algorithms, sentience and intent too, but still need "input and output devices" to put the intent to use. At most, AI (after reaching sentience) can harm the humans by using our own creations ("idiot" machines like an electric saw, or dumb ones like a car with ECU) against us. But such machines can be controlled by humans easily in, by simply pulling the plug. As long as such idiot and dumb machines do the bulk of work and form the majority of machine world, AI can not overcome humans in superiority (e.g. AI can not create a new machine, let's say an electric saw, starting from mining iron ore from one place, producing copper wires at another place, plastic moulds etc)...While humans can roll back to producing only dumb machines where we can control them fully. You know what I mean?

[SilentEngine]

Originally Posted by Nav-i-gator

7. (most important) AI is like a brain, and hence have same limitations. It can generate ideas, algorithms, sentience and intent too, but still need "input and output devices" to put the intent to use. At most, AI (after reaching sentience) can harm the humans by using our own creations ("idiot" machines like an electric saw, or dumb ones like a car with ECU) against us. But such machines can be controlled by humans easily in, by simply pulling the plug. As long as such idiot and dumb machines do the bulk of work and form the majority of machine world, AI can not overcome humans in superiority (e.g. AI can not create a new machine, let's say an electric saw, starting from mining iron ore from one place, producing copper wires at another place, plastic moulds etc)...While humans can roll back to producing only dumb machines where we can control them fully. You know what I mean?

This is a very important point. Certainly, doomsday scenarios of Robotized AI walking around seem far fetched.

However, to cause some real havoc in society, an evolved AI may not need physical/mechanical manifestation. It's quite possible that there can be a virtual action (AI deciding on some outcome) which causes effects which have real/physical ramifications on humans - examples: AI crippling banking networks or rogue AI elements gaining access to launch missiles.

[V.Narayan]

Where AI is concerned we need to be careful of two things - (a) the intentions of the human creator. (b) we should not underestimate how rapidly technology will change. Jules Verne's novel 'From Earth to the Moon' was written in 1865 and just 104 years later came Apollo 11.

Originally Posted by coolvenk

The fact-checking website Snopes rated this story to be false.

Read for yourself:
http://www.snopes.com/facebook-ai-de...-own-language/

Snopes - for the sake of the next generation I sincerely hope they are right. At the same time lets accept that the PR machinery of the likes of FB is fully capable and presenting things in their own light.