M.I.T. Plans College for Artificial Intelligence, Backed by $1 Billion

From The New York Times

"Mr. Schwarzman said he hoped that the M.I.T. move might trigger others to invest in America’s A.I. future, not just commercially. He points to the major push the Chinese government is making, and notes the fruits of United States government-funded research in the past — technologies that helped America take the global lead in industries from the personal computer to the internet.

“I think we’ve been lagging, for whatever reason,” Mr. Schwarzman said."

Artificial Intelligence is the new Space Race, it seems. This one is more dangerous though since a lot of hard or good can be done using AI. It can be used to cure and to manipulate people.

Now that is a college I would have tried very hard to get into had it been around when I was young.

minamicruiser said:

From The New York Times

"Mr. Schwarzman said he hoped that the M.I.T. move might trigger others to invest in America’s A.I. future, not just commercially. He points to the major push the Chinese government is making, and notes the fruits of United States government-funded research in the past — technologies that helped America take the global lead in industries from the personal computer to the internet.

“I think we’ve been lagging, for whatever reason,” Mr. Schwarzman said."

Artificial Intelligence is the new Space Race, it seems. This one is more dangerous though since a lot of hard or good can be done using AI. It can be used to cure and to manipulate people.

Click to expand...

I would have been more inspired if Mr Schwarzman had become educated in the problem before he and MIT launched this attempt at a solution.

Crazy wild hair idea I know....

Hawkeye10 said:

I would have been more inspired if Mr Schwarzman had become educated in the problem before he and MIT launched this attempt at a solution.

Crazy wild hair idea I know....

Click to expand...

Ready, fire, aim.

I think we need to start here,:mrgreen:
A robot may not injure a human being or, through inaction, allow a human being to come to harm.
A robot must obey orders given it by human beings
except where such orders would conflict with the First Law.
A robot must protect its own existence as long as such protection
does not conflict with the First or Second Law.
Seriously, AI can be broken down into several basic levels, the first is in use and has been
effectively in use for several centuries, the Expert system.
The Expert system has a collection of examples from experts of how they responded to a given situation.
it is a spinoff of a General Staff, who can function as if the General were still alive, (even when he is not).
In Modern times NASA used interviews with hundreds of Apollo Scientist, Astronauts, and Engineers, to create
the launch control expert system. The system that said NOT to launch Challenger, but was overridden.
The next level is self learning, and we are still working out the bugs.
Most of the Science fiction says that at some point self learning systems will become sentient,
some become benevolent servants, some become toddlers who throw tantrums.

longview said:

I think we need to start here,:mrgreen:
A robot may not injure a human being or, through inaction, allow a human being to come to harm.
A robot must obey orders given it by human beings
except where such orders would conflict with the First Law.
A robot must protect its own existence as long as such protection
does not conflict with the First or Second Law.
Seriously, AI can be broken down into several basic levels, the first is in use and has been
effectively in use for several centuries, the Expert system.
The Expert system has a collection of examples from experts of how they responded to a given situation.
it is a spinoff of a General Staff, who can function as if the General were still alive, (even when he is not).
In Modern times NASA used interviews with hundreds of Apollo Scientist, Astronauts, and Engineers, to create
the launch control expert system. The system that said NOT to launch Challenger, but was overridden.
The next level is self learning, and we are still working out the bugs.
Most of the Science fiction says that at some point self learning systems will become sentient,
some become benevolent servants, some become toddlers who throw tantrums.

Click to expand...

My favorite versions of that fiction is when the smart-bot realizes his masters are idiots. My guess is that in r/l this phase does not take very long to develop.

minamicruiser said:

From The New York Times

"Mr. Schwarzman said he hoped that the M.I.T. move might trigger others to invest in America’s A.I. future, not just commercially. He points to the major push the Chinese government is making, and notes the fruits of United States government-funded research in the past — technologies that helped America take the global lead in industries from the personal computer to the internet.

“I think we’ve been lagging, for whatever reason,” Mr. Schwarzman said."

Artificial Intelligence is the new Space Race, it seems. This one is more dangerous though since a lot of hard or good can be done using AI. It can be used to cure and to manipulate people.

Click to expand...

IMO all of this is a vastly overblown fad which has a bigger impact on the wallets of VC's than it does long term technological breakthroughs. There's this pipe dream that somehow we'll be able to write a single smart algorithm that will be able to solve all sorts of problems without the need for lots of smart people. Maybe I'm way off base here, but I just don't see it happening, at least not soon. The human brain has more computing power than the fastest supercomputers (except for maybe China's Sunway TaihuLight) and it takes 18 years and an army of teachers to train one to be useful. (and even after 18 years the reliability isn't very good)

Not to get too mathy, but all optimization algorithms are governed by the No Free Lunch (NFL) theorem. This was work done in the late 90's by Wolpert and Macready in the height of the evolutionary programming fad. They also expanded it to machine learning. In short, NFL means that all algorithms perform the same averaged across all possible objective functions. That implies that techniques which target specific problems will work better than more general approaches. Even if an algorithm seems to perform wonders, it has to pay for that performance in other ways like a reduction in provability of correctness.

ML is the fad of the day because it outperforms anything else in vision. But just because it is good in one application doesn't mean it has universal applicability. ML is just making decisions based on many statistical inferences. And for the most part it's doing these in complicated networks which are mostly a black box. That means it can learn things that you really don't want it to learn, like superstitions or racism. And the learning is also psuedo non-deterministic. So you can train on a set of data with great results, add something innocuous and suddenly your entire system becomes unreliable. It's also extremely hard to prove reliability or improved performance.

eg Deep neural networks are easily fooled: High confidence predictions for unrecognizable images | Evolving AI Lab

In short, it's pretty easy with ML techniques to quickly generate a compelling demo for almost any application. Just download Google's Tensor, grab a bunch of application data, and after a few days you should be able to show a video of something compelling. The problem comes when you have to do it for real. ML is a powerful collection of techniques with a wide variety of applications, but ML comes with some significant limitations that aren't understood even by many experts.