New Computing Paradigms: Mark Ritter at TEDxUConn 2013
Articles,  Blog

New Computing Paradigms: Mark Ritter at TEDxUConn 2013


Translator: Nuncio Casanova
Reviewer: James May I was always interested in science. As a kid I was building model rockets. I was performing chemical
experiments in my basement and one time I almost blew up
my parents’ basement trying to build a laser. So when the physics professor who was next door neighbor
to my grandparents invited me to a lecture on lasers, I was eager to attend. I was a six feet two inches
lanky thirteen year old. So when I sat down, right away someone asked me to move, because I was blocking his view. But little did I know who that person was or the effect the work that he did
will have on my career. After that lecture I asked
my professor friend and he said, “That was Bob Noyce,
chairman of Intel”. It turns out he has been a student
of this professor some years before. And the professor had showed him some of the first
transistors from Bell Labs, and told him these were important. And just less than ten years after this, Bob Noyce co-founded
Intel with Gordon Moore The transistor is like a switch and it is in all of our computers. So you think of it as a switch
that you switch with your finger, except the transistor is switched is switched by an electrical voltage
applied to a central contact. In 1965 Gordon Moore observed that the chips they were making
had doubled the number of transistors every two years. This became known as Moore’s Law, and that trend has continued until today. So now, the biggest computer chips
have billions of transistors. Some people have speculated that these supercomputers
will continue this trend and a supercomputer will shrink
to the size of a grain of sand. But that is not physically possible. Parts of the transistor are already
approaching atomic dimensions where quantum mechanical effects
limit its performance. So we think we can only go
about a factor of sixteen denser in transistors on a chip, before we reach the end of the game. So Moore’s Law is ending, and with it, the wave of technology
that I’ve ridden most of my career. And so, what comes next? Is this really the end
of more powerful computing? These questions dodged me so that five years ago I said
I need to shift the research focus more toward new computational paradigms. And some of those, for example,
that people have considered, there are optical computing, quantum, neuromorphic, and even DNA computing. But today I will speak
of quantum computing and neuromorphic or cognitive computing, which I think are the most promising. In today’s computers, again, the transistor acts like a switch, it is either on, in state one,
or off, in state zero. It can’t be in both states, it is only in one of those states
at a time. But physicists have studied atoms
and find they behave quite differently. An atom has discrete
quantized energy levels, that is where the idea
of quantum mechanics comes from. We can label the two lowest levels
a zero state and a one state. And the surprising thing is that scientist have found that atoms
can exist in both states simultaneously. This is known as
quantum superposition state. And that bothered Richard Feynman, who is a Nobel prize-winning physicist. Because Feynman
looked at that and he said, that is a problem. Because our classical computers
have to grow exponentially in size in order to simulate real reality,
which is atoms. To explain why that is so,
consider simulating as a number of two-level atoms. If we have two two-level atoms, to simulate them requires
four two-bit states. Both in the zero state,
both in the one state and then each one in the one state, so that is four states. Four atoms require sixteen states. Ten atoms require
one thousand twenty four states. And you say to me,
what is this guy talking about? Thousand twenty four? We have big computers that can do that
in a billionth of a second. That is true, but I’d like you to consider
just ten times more, one hundred atoms. Now, how long would the simulation take? And the answer is three million years using the largest computer
that we now have, executing ten thousand trillion
executions per second. This is the fact that Feynman realized, that our current computational paradigm can not simulate even a modest
number of atoms in all their quantum mechanical behavior. And that is what he saw is the problem, but being a clever guy,
he found a solution. He said, why don’t we make
things called quantum bits, or qubits? And if they can exist
in superposition states, then I can simulate one hundred atoms with one hundred qubits,
and in a short time, because it will be
exponentially accelerated, because they exist in those
superposition states. So, how will we build such a technology? Scientists around the world, including in our group at IBM, we are looking at different ways
to make qubits. Some are spins on an electron or an atom. It Is like a magnetic moment or more like a little magnet
that points up or down. Others are like the atom
that I have already described, and ions or atoms in crystals. And still others,
like we are developing at IBM, are superconducting circuits that behave like atoms,
that have energy levels. In all of these, you can get
quantum superposition states. But these quantum states
are very easy to disturb. So many of these new technologies are in a vacuum and cooled
to a very low temperature in fact to within almost
below a tenth of a degree within an absolute zero,
in order for them to work. We’ve already demonstrated
with our technology to the right there, three coupled qubits that show this
quantum superposition with high fidelity. We’ve also have been able to factor a number like fifteen into the primes
five and three. We’ve executed small
database-search programs. And you’ll say, well,
that is pretty trivial. But the point is that,
unlike transistor computing, every qubit we add
doubles our computational capacity. So we just have to add enough qubits and control them, and we’ll have exponentially
faster computation than we have today, and be able to calculate
intractable things, like designing new molecules for new drugs
that you really can’t do today. Just like Feynman envisioned, we would be able to simulate
physical reality as it really is, in a quantum sense. And the other aspect
is neuromorphic or cognitive computing. Imagine the usefulness of a computer
that can do what a small child does. Understand language, be able to learn and create new things, recognize objects and people accurately while consuming only a few watts, where today the largest computers,
consuming megawatts, can’t do as well as a child at these task. That’s why we feel
that a different approach is required for learning recognition and inference
in machines, as people have tried it for years. But in fact one
of the differences is that, in a computer every transistor
that I described is connected to five or ten
other transistors. However, in the brain,
the neurons, or basic brain cells, have inputs from ten thousand
other neurons through structures called synapses, which are shown
in that red outlined insert there. And this synapses are like connections
between all the neurons, and they change when we learn things, so we believe a large part
of the memory in the brain is stored in the synapses. And so, memory and computation
are distributed through the brain, rather than contained in separate regions, like in computer chips today. That makes the brain
very hard to understand. But people are working to study this, and this is a video from Misha Ahrens and his group
at Howard Hughes Medical Institute. This is the brain
of a transparent zebra fish that has been genetically engineered so the neurons fluoresce
when they are active. This shows how complex the brain is. We are only beginning
to understand the brain and we need similar data on mammals that have real recognition
more like humans in order to understand the algorithms or the recognition
that is being used in the brain. So we already have, inspired by biology, algorithms that work pretty well, like speech recognition
in your cellphones. That works well, but it is no quite as accurate as a human and it is not as efficient, and it doesn’t understand speech,
as you must’ve noticed. IBM has created this Watson computer that was able to defeat
Jeopardy champions, but still is ten thousand times
less efficient than the human brain. So we need new cognitive technologies, that my group and others are working on, that have memory that behaves
more like the synapses in the brain. New technologies coupled
with new algorithms will make highly efficient
cognitive computers. Finally, I would say that we are not
trying to create a brain. I’ll be very clear about that. We want to take inspiration from the brain to create cognitive technologies
that are useful to us, just like the Wright brothers took inspiration from birds
to create airplanes, not giant. mechanical wing-flapping birds. And that is the difference. We need cognitive technologies
to sort through the vast amounts of information we are flooded with,
to find useful knowledge. It’s almost like an answer
to the rhetorical question T. S. Elliot posed. “Where is the wisdom
we’ve lost in knowledge?” “Where is the knowledge
we’ve lost in information?” Yes, Moore’s Law, although it is ending, there is a new wave of computation that is even greater
in efficiency and promise that is just beginning with cognitive
and quantum computing, and I’m glad to be a part of that. Thank you. (Applause)

One Comment

  • judaspreistvlct

    Sounds like they are going to positively be able to dig up dirt on absolutely anybody! Also, CCTV is going to be something you cannot run from! Now, who's your God?

Leave a Reply

Your email address will not be published. Required fields are marked *