AI - Artifical Intelligence

[Sojourner]

Anyway, the way I see it: biological immortality, brain emulation, at will gene programming and AI.

Any of those mean the end of the human race as we know it. And all of those are quite close to reality.

Hmmm... what do you mean by "quite close"? 25 years? 40 years? I also follow this subject with some interest, and I'd say the consensus view is that "the singularity" is certainly not going to occur within the next 25 years, and probably not even the next 50. If you think differently, please share some links that support that point of view.*

The problem, as I understand it, is primarily that true AI -- that is, machine intelligence that is deeply analytical, can generalize and model problems, and that self-adapts and improves as it learns -- is nowhere near being a reality. Of course, there are many AI research projects ongoing around the world that are attempting to further the field in fundamental ways, but a tremendous amount of progress will need to be made before we're close to creating something akin to an AI human-like brain (or even a lizard-like brain).

* According to a poll of attendees at a "Future of Humanity" conference in 2011, there is about a 50% chance of developing an artificial human-level intelligence by the year 2050, and a 90% chance of developing it by 2150. Only 10% thought it would happen by 2028. I imagine the people at this conference would tend to be "fast AI" enthusiasts, so IMHO these poll results could be a bit optimistic... but I could be wrong. The results were posted on the Kurzweil AI website, after all.
Machines will achieve human-level intelligence in the 2028 to 2150 range: poll | KurzweilAI

 

[Sunset]

This issue has come up for self driving cars, and they are not really dealing with AI, just decision tree algorithms.
What to do when car is about to collide with person who stepped off curb?

• Hit and possibly kill them.
• Immediate stop injuring multiple passengers in car, possibly still hitting pedestrian
• Swerve into oncoming traffic

A deliberate decision has to be made, and sure they will modify it based on circumstances like speed, distance, road conditions etc.



Right now we call such event an accident, in the future it will simply be an action.

 

[Tadpole]

Computing speed advances have always been more important than algorithms.

 

[braumeister]

Another breakthrough may be imminent.
As we all know, Susan Calvin designed the robot brains using positronic circuitry, not the clumsy electronics of today. As soon as someone figures that detail out, it's smooth sailing.

 

[M Paquette]

Consider an artificial general intelligence dedicated to running a paper clip factory. The AGI is given an overall mandate to optimize production of paper clips. It commences it's goal-seeking behavior, exploring all pathways available to produce the most possible paper clips at the lowest per-unit cost.

That's all it wants to do. Make as many paper clips as cheaply as possible.

Delivery trucks bring wire into the plant as feedstock. Optimize by moving wire fabrication in-house.

Now any metal can be used as feedstock. Optimize by using delivery vehicles as well as their cargo. Adds facilities for removal of organic contaminants from feedstock.

Optimize away the need for specific delivery vehicles, by using the stream of metal conveniently passing by on the highway. Adds wireless controls to allow overriding of autonomous vehicles to divert them to the plant as feedstock, and additional capacity for stripping organic contaminants...

And so on... Very quickly.

 

[Totoro]

Hmmm... what do you mean by "quite close"? 25 years? 40 years? I also follow this subject with some interest, and I'd say the consensus view is that "the singularity" is certainly not going to occur within the next 25 years, and probably not even the next 50. If you think differently, please share some links that support that point of view.*

"The singularity" is a confusing concept to me, so apologies if I miss the mark here.

With quite close I meant practical reality in some form for any of the things I mentioned, roughly anywhere from 5 to 40 years.

To go down the list, where it seems we are today, you most likely know all of these already. I've added a sort of random link for flair

• At will gene (re)programming can be done today. We can alter or eradicate entire species including humans within a few generations. Jennifer Kahn: Gene editing can now change an entire species -- forever | TED Talk | TED.com
• Biological immortality with regards to organs: simple organic structures can be built and have been implanted. Whole organ (re)generation is probably a 5 to 10 years out (e.g. kidney regeneration from one cell from your own body), it's in experimental stage in the lab. Scientists growing livers, kidneys, ears in labs amidst organ shortage - CBS News
• Biological immortality with regards to the rest of our body: probably several decades out. Most research groups are still laying the groundwork, including Google's Calico.
• Brain emulation: just one approach, one research group is simulation at molecular level parts of a brain ( https://en.wikipedia.org/wiki/Blue_Brain_Project ). They've done rats (partly, the neo-cortex). Looking at their timeline it's about 10 to 20 years out before we get at near-human level. That's only one approach. Functional simulation is another: Consider the human brain has ~70 billion neurons, and a modern GPU has ~5 billion [-]neurons[/-] transistors. How fast that'll go is anyone's guess. I don't know of a research group targeting that, which surprises me.
• AI: It's a matter of perception here. I'd argue that the whole class of systems like AlphaGo, Google self-driving car, and DeepMind are already deeply analytical with capabilities to generalize and model problems. Basically what humans are good at are pattern recognition and vision. The last few years a few techniques have enabled us to jump leaps and bounds there, beyond what humans can do in many cases.

That's why the way things are going I won't be surprised I'll live to see the end of humanity as we know it. If any of the above pans out, we're "done for".

I don't think we'll ever really build a human look-a-like though. We'll just jump straight passed it.

 

[aja8888]

It did
"Lee Sedol then lost Game Three, and AlphaGo claimed the million-dollar prize in the best-of-five series."

I didn't read the article, just read the quoted part where it said:

The machine claimed victory in the best-of-five series, winning four games and losing only one.

The way this read to me was that the machine won four games, then claimed the victory.

 

[photoguy]

I'm not really worried. Futurists have a pretty poor track record.

+1 and AI in particular has been dismal with projections.

The problem, as I understand it, is primarily that true AI -- that is, machine intelligence that is deeply analytical, can generalize and model problems, and that self-adapts and improves as it learns -- is nowhere near being a reality.

Most of what I see that's dubbed as AI today is just impressively good application of computational power to statistics (in often very clever ways). That can take you an incredibly long way in specific applications. But the "learning" that occurs can be thought of as similar to the improvement you get when estimating a population mean from 10 samples versus 100 samples.