Ever heard of the "AI Day"? Me neither, but

[spinnerlys]

it promises to surpass Christmas, and if so, that is more than alright we me as I quite dislike (hate is such a strong word and does put too much emphasis on the energy involved feeding that emotion) that now consumerist holiday (not as bad as that Black Friday stuff and better than that Bloody Sunday).

Anyway, via stumbling I found this interesting, but short article, and while I am only an observer and not that very well versed in all things "Singularity", I am still learning, as those ideas fascinate me, but I am also often unable to communicate those via spoken words around gatherings of people. I hate small talk though. Hmm.

Here it is: AI Day Will Replace Christmas as the Most Important Holiday in Less Than 25 Years

 

[CorporalCaptain]

No, it won't.

 

[Robert Maxwell]

The link is little more than ad for the author's book. How shameless.

 

[spinnerlys]

The link is little more than ad for the author's book. How shameless.

I did not perceive it that way, as I was more thinking about the AI Day replacing Christmas.

 

[FPAlpha]

2 things

The article or the author of that book is gravely misjudging religion as a factor. Religion has been present since the day humanity was able to think and it is so firmly entrenched in society as to be near unremovable. For the common, religious person they see no reason to elevate technology over faith. Technology is there to support our life and make it easier but faith fills a place in human existence no machine can ever attempt to fill or even replace except for the few who already have rejected faith and replaced it with science or something else.

The challenge of the human species is to not let this kind of AI get beyond our controls; to have adequate safety measures and kill-switches built in. Such measures would not be dissimilar from how civilization delicately handles nuclear weaponry, which some political experts believe have staved off world wars in the last half-century.

A nuclear bomb is not sentient, it doesn't need to be because all it needs to do is go boom when its controller wants it to. How would one "shackle" such a thing as a true AI? Wouldn't it be able to analyze its code and discover the control mechanisms?
Would it be ok to be vastly more intelligent and maybe more powerful and still be a slave to less intelligent beings? Would you?

Would that even be the right signal to the AI that one wants to use its possibilities but will destroy it as soon as it does something we might perceive as a threat?

I don't know how the "thought" process of an AI will look like.. is it just so much faster than our fastest computers today so it can accurately simulate human responses and fool us into thinking it is actually intelligent? Or can it become self aware and start thoughts of it own with needs and desires it was never programmed with? Would our handling of such an entity create something like Skynet just because we behave full of fear towards it and constantly hold a gun to its head or would it be better to build trust and hope for the best?

How can we hope to understand something like an AI when even our smartest people combined couldn't hope to match it?

There are just so many questions and i seriously doubt we can answer them before an AI emerges.. it might happen by "accident", i.e. not intentional and the time to implement control measures may have already passed but i seriously doubt we could effectively control an AI even if we wanted to.

 

[rhubarbodendron]

I am not sure if our smartest people combined would be excelled by an AI. If an AI would be something computer-like it would be hampered by its ability to only think in 0 and 1. Admittedly, quite a lot of humans tend to think in black and white, too, but as a general rule we are able to think in greys as well. As long as an AI would be based on computer technology it couldn't follow us there.
There are, however, interesting experiments with semi-organic structures that combine single nerve cells with computer technology. So far they can't do much but the first step is made.
This would however in the long run raise serious questions: if an AI one day was really able to think like a human and in addition contain organic tissue, would it count as a machine or an organism?
As a human with a ton of artificial joints, an artificial heart etc. won't stop being considered a human, the amount of human tissue must not be the critical factor for our definition of a person. This will make a few very interesting and controversial discussions in the future (if this semi-organic stuff is really better than a computer which remains to be seen in the next decades).

As for Christmas - I'm afraid it already is a totally commercial holiday. Yet, getting presents in the middle of the cold and dark season is such a nice custom that I think it'll survive.

 

[Robert Maxwell]

No offense, but it sounds like your background in biology has left you with a bit of an ignorant prejudice in terms of what computers can do by comparison. You seem to be implying a computer can only make simple yes/no, on/off kinds of decisions and judgments. That's not true at all. A computer can make very complex decisions based on multitudes of factors. For instance, search engines rank results based on complex calculations that determine their relative relevance.

Yes, a digital computer is going to have a very hard time emulating how humans think because it's built on hardware that works in a fundamentally different way. That doesn't mean it's impossible, but obviously we're nowhere near it. But saying computers "only think in 0 and 1" is like saying humans "only think in action potentials." It gets the physics (basically) right while missing the point.

 

[rhubarbodendron]

I don't think my background clouds my perception in this case.
Based on the very way they are built, computers work only with 2 options (current vs no current = 1 vs 0). A bit can only be either 0 or 1 but nothing between them. That principle is as old as Zuse's first computer and it's still used today, only the relays have gotten much tinier.
All programming must very carefully avoid contrary commands or unclear commands as they will inevitably lead to either the programme going into a loop or breaking off.
This is why computers (as long as they work as the ones we are currently using) will always be a little limited, compared to our brains. The latter actually can have bit-like states between 0 and 1, depending on how much Ca++ gets released into the synaptic cleft (this does, however, not go for the motorical nerves. Those have indeed only an on/off option like a Bit).

If you like, I can invite Santaman and Spaceball to this thread. They make a living of building and programming computers and might make excellent referees. Also, they are sure to have interesting views to add to the topic.

 

[Robert Maxwell]

Uh, I am a software engineer and I build computers. I know how they work, thanks.

I'm saying that claiming they think only in 1 and 0 is too reductive. It implies they cannot handle nuance, ambiguity, complexity, or fuzzy situations. This is clearly not the case.

 

[JarodRussell]

You can break ANY process down to a composition of 0 and 1. It's not impossible to recreate the processes in the human brain.

A state of yes, no, maybe is represented with 0, 1 and 2 (00, 01 and 10) then. No issue. A neuron's action potential can be recreated digitally.

 

[rhubarbodendron]

I didn't mean to offend you, Robert. My appologies.

I beg to differ, Jarod. Certain nerve cells can fire fractions of an impuls instead of a full one. That's not a 00/01/10 switch but a 0/0.1/0.2/0.3 ... /1 one (actually it works even in finer grades, I just was too lazy to introduce more digits)

In my experience you can't tell a computer "maybe". It will always require more information of the if/then kind.

Robert, could you perhaps give a short example for a maybe programming?

(completely off-topic but not unimportant: is it ok if I abbreviate your names like that, guys? Feel free to call me rhubarb)

 

[Robert Maxwell]

Abbreviate me however you like!

A good example in real life is the self-balancing washing machine. Washing machines don't have "balanced" and "unbalanced" states. It's relative. But with a combination of sensors and programming, the hardware understands when it is "more balanced" and "less balanced" and can take steps to promote a "more balanced" state, dynamically as the load progresses through its cycle. Again, this is not an either/or situation, but one where the goal is to obtain the most optimal state possible when suboptimal conditions are occurring. Well-programmed computers and devices are much better at this than one might think--the algorithms have advanced considerably in recent years.

 

[rhubarbodendron]

I would have assumed that to be a 0/1 decision, too: in balance (meeting a given standard) and out of balance (deviation from given standard) with the countermeasures being taken according to the degree of the deviation from the goal as measured by the sensors.
like (this is no coding, of course, just a sumup of the general idea) :

goal = 0° deviation from the vertical
if sensorinput = not 0°, then correct by (-1) x (sensorinput)
if sensorinput = 0° = goal, then don't correct

That would make a 0/1 decision (meeting goal or not meeting goal). And the correction to make would be the same force the sensors indicate, but in the opposite direction.


As far as I can see, we have different opinions on whether a washing machine's state of balance is relative or absolute. I would have thought it to be absolute, it only changes really quickly which is a challenge for every sensor. If I understand you correctly, you consider the machine's movements rather like the position of electrons in Heisenberg's Uncertainty Principle?

(That's something that even as a student I felt sceptical about. If you were fast enough to watch an electron's course and anything that might influence its speed (impuls), shouldn't you be able to predict its movement and position at any given time? Doesn't it just appear to be unpredictable because we are too slow to watch it properly? I'm looking forward to future developments. Maybe one day our sensors will be fast enough after all.)

 

[Robert Maxwell]

I wouldn't say the Uncertainty Principle applies, but rather that there is a substantial degree of unpredictable behavior in the action of a washing machine. Items can shift around inside the drum quickly, changing the weight distribution and requiring rapid correction. The washer has no way to know the size, shape, weight, and location of every article of clothing, so it cannot anticipate what will shift and when. It can only respond to circumstances as they develop. It would appear "random" to both us and the washer (if the washer had the ability to think), but it's just governed by the usual physical laws. There are just too many factors involved to predict in real-time.

Your scenario is far too rigid. A washer is always at least a little bit out of balance because it's impossible to maintain a perfect weight distribution within the drum. So, it works to keep things as optimally balanced as possible, and can respond more aggressively when it senses the load is drifting out of balance quickly.

Essentially, your "balanced" state never happens, it's always "unbalanced" and trying to correct. In that context, making it a binary distinction is neither informative nor useful for addressing the problem, and so it must be programmed accordingly.

Facial recognition is another area where computers tend not to deal in absolutes. While the end result is either a positive match or no match, the actual process involves a lot of ambiguity and fuzziness. In the end, the system has some level of confidence that the face in question matches a known image, and a threshold above which it can express confidence that the match is accurate.

It sounds like what you may be going for is that computers are deterministic, which they are by design. Living brains may well not be.

 

[JarodRussell]

I'd say since all humans and animals work the very same way with some minor variations, the brain seems pretty much deterministic. All input is processed the very same way in every normally functioning human brain.

 

[Robert Maxwell]

I'd say since all humans and animals work the very same way with some minor variations, the brain seems pretty much deterministic.

Yes and no.

The brownian motion involved in chemical signaling is random, which affects neuronal behavior. However, this system of unreliable parts gives rise to an overall reliable machine. So, brains aren't strictly deterministic in terms of their basic functioning, but they are deterministic enough for us to actually understand what it's doing.

Incidentally, computers function in a similar way at the physical level. A transistor doesn't know what 0 or 1 are. It's simply a switch that flips to a different position when it has a high enough voltage. High voltage is 1, low voltage is 0. Since voltages are variable thanks to noisy power supplies and other factors, the threshold between 0 and 1 has to be carefully set, otherwise the CPU won't work. Once again: a reliable machine generated from unreliable parts (and energy inputs.)

That's not to draw too direct a comparison, as human brains are essentially sophisticated analog signal processors and computers are digital number crunchers, but it's interesting that they both emerge from similar principles.

All input is processed the very same way in every normally functioning human brain.

Not exactly. The devil is in the details. Ten people, all with "normal" brains, will not respond the same way to having their fingertips poked with a needle at the same pressure. Some will react sooner than others. If you introduce external noise (such as vibrating the person's hand), that may improve or reduce their ability to sense the needle. (I believe some research says it actually improves threshold detection.)

If Rhubarbodendron knows anything about stochastic resonance, maybe she can comment more on this.

 

[JarodRussell]

But isn't the same thing true for computers? 10 CPUs running the same calculations will show variations in time, heat distribution, etc...

 

[CorporalCaptain]

But isn't the same thing true for computers? 10 CPUs running the same calculations will show variations in time, heat distribution, etc...

But the differences are all masked by the fact that the ten CPU's only produce different results if there is a bug or malfunction. When there is no bug or malfunction, the calculations generated as output should all be identical. And, I guess I should explicitly state certain qualifications (which I assume are implicit in your premise), such as that presumably the executed programs don't sample hardware registers for the purpose of random number generation, generally speaking that the state of external devices to which they attach are functionally identical (e.g., network drives appear the same when they are accessed by the different computers), etc. And we're talking about digital computers, as opposed to analog, right?

On the other hand, in the case of people, the idea that different people do different things when confronted with the same situation (in what they they choose to do) isn't a bug; it's a feature and it's expected as normal variation.

 

[JarodRussell]

But isn't the same thing true for computers? 10 CPUs running the same calculations will show variations in time, heat distribution, etc...

But the differences are all masked by the fact that the ten CPU's only produce different results if there is a bug or malfunction. When there is no bug or malfunction, the calculations generated as output should all be identical. And, I guess I should explicitly state certain qualifications (which I assume are implicit in your premise), such as that presumably the executed programs don't sample hardware registers for the purpose of random number generation, generally speaking that the state of external devices to which they attach are functionally identical (e.g., network drives appear the same when they are accessed by the different computers), etc. And we're talking about digital computers, as opposed to analog, right?

On the other hand, in the case of people, the idea that different people do different things when confronted with the same situation (in what they they choose to do) isn't a bug; it's a feature and it's expected as normal variation.

First of all, I realized that the fingertip needle example is lacking a bit, because it depends on so many variables like the pathways from fingers to the brain, the density of the receptors in the finger tip, etc...

But the result is always the same: the brain recognizes that a needle touched the fingertip.

The difference is speed of recognition, intensity, etc...


And that's what I meant with the CPU example as well. The RESULT of the calculation will always be the same. But there will be minor variations in each CPU regarding the calculation time, the generated heat, the amount of power needed, etc... that's just natural, because no CPU is exactly identical to the other, even when they come from the same assembly line.

And in human brains, this difference is even greater, because every human brain develops differently because of different outside factors. If you compared the brains of twins, you would already see variations in the processing. BUT the results to a given input would always be the same.

 

[Robert Maxwell]

I don't know about you, but a different reaction time is certainly a different result to me. Reaction times are the difference between life and death, after all.