Scientists Create Artificial Brain With 2.3 Million Simulated Neurons

22,563 18 Loading

Another computer is setting its wits to perform human tasks. But this computer is different. Instead of the tour de force processing of Deep Blue or Watson’s four terabytes of facts of questionable utility, Spaun attempts to play by the same rules as the human brain to figure things out. Instead of the logical elegance of a CPU, Spaun’s computations are performed by 2.3 million simulated neurons configured in networks that resemble some of the brain’s own networks. It was given a series of tasks and performed pretty well, taking a significant step toward the creation of a simulated brain.

Spaunstands for Semantic Pointer Architecture: Unified Network. It was given 6 different tasks that tested its ability to recognize digits, recall from memory, add numbers and complete patterns. Its cognitive network simulated the prefrontal cortex to handle working memory and the basal ganglia and thalamus to control movements. Like a human, Spaun can view an image and then give a motor response; that is, it is presented images that it sees through a camera and then gives a response by drawing with a robotic arm.

Will AI brains of the future look more like Watson or Spaun?

And its performance was similar to that of a human brain. For example, the simplest task, image recognition, Spaun was shown various numbers and asked to draw what it sees. It got 94 percent of the numbers correct. In a working memory task, however, it didn’t do as well. It was shown a series of random numbers and then asked to draw them in order. Like us with human brains, Spaun found the pattern recognition task easy, the working memory task not quite as easy.

The important thing here is not how well Spaun performed on the tasks – your average computer could find ways to perform much better than Spaun. But what’s important is that, in Spaun’s case, the task computations were carried out solely by the 2.3 million artificial neurons spiking in the way real neurons spike to carry information from one neuron to another. The visual image, for example, was processed hierarchically, with multiple levels of neurons successively extracting more complex information, just as the brain’s visual system does. Similarly, the motor response mimicked the brain’s strategy of combining many simple movements to produce an optimal, single movement while drawing.

Chris Eliasmith, from the University of Waterlook in Ontario, Canada and lead author of the study is happy with his cognitive creation. “It’s not as smart as monkeys when it comes to categorization,” he told CNN, “but it’s actually smarter than monkeys when it comes to recognizing syntactic patterns, structured patterns in the input, that monkeys won’t recognize.”

Watch Spaun work through its tasks in the following video.

One thing Spaun can’t do is perform tasks in realtime. Every second you saw Spaun performing tasks in the video actually requires 2.5 hours of numbers crunching by its artificial brain. The researchers hope to one day have it perform in realtime.
It’s important to note that Spaun isn’t actually learning anything by performing these tasks. Its neural nets are hardwired and are incapable of the modifications that real neurons undergo when we learn. But producing complex behavior from a simulated neuronal network still represents an important initial step toward building an artificial brain. Christian Machens, a neuroscientist at the Champalimaud Neuroscience Programme in Lisbon and was not involved in the study, writes in Science that the strategy for building a simulated brain is “to not simply incorporate the largest number of neurons or the greatest amount of detail, but to reproduce the largest amount of functionality and behavior.”

We’re still a long way from artificial intelligence that is sentient and self-aware. And there’s no telling if the robots of the future will have brains that look like ours or if entirely different solutions will be used to produce complex behavior. Whatever it looks like, Spaun is a noble step in the right direction.

Peter Murray

Peter Murray was born in Boston in 1973. He earned a PhD in neuroscience at the University of Maryland, Baltimore studying gene expression in the neocortex. Following his dissertation work he spent three years as a post-doctoral fellow at the same university studying brain mechanisms of pain and motor control. He completed a collection of short stories in 2010 and has been writing for Singularity Hub since March 2011.

Discussion — 18 Responses

  • matt December 10, 2012 on 4:40 pm

    While they aren’t lying when they say they have created the largest functional spike network, what they have done in essence is reduced the capabilities of spike neurons to something more simple. It was shown years ago that one can construct any regular computation using spike networks, and this is what they have done. It is easy to take a function, and work backwards so that a spike neural network can perform the said function. What is difficult is creating networks with complex behaviours where the underlying function is unknown and has to be learnt. It’s great to see people researching this field, but unfortunately this project adds no real innovation to building or understanding brain like structures.

  • Roaidz December 11, 2012 on 5:17 am

    Excuse me, but I don’t believe that scientists can create a robot with
    the same inteligence as humans. Even thousands years will past but
    still, they can’t fall in love, they will not have pity, and will not recognize
    our Almighty God.

    • Roy Lindsay Roaidz December 11, 2012 on 7:29 pm

      Why are you here?

      • Roaidz Roy Lindsay December 12, 2012 on 6:58 am

        OK Roy, remove me from your list.

      • seemsArtless Roy Lindsay December 19, 2012 on 9:24 am

        If I can answer for @Roaidz, perhaps he is here trying to understand how his religious world view will ever map to current and possible future scientific developments. A noble cause with a long, long history! Of course our Almighty Helios drives the sun across the sky on a chariot! …no?… Well, of course our Almighty God placed us at the center of the universe with the sun, planets, and stars rotating about us. …no?… Well, of course we humans are special, nothing else can love…. All beliefs that we’ve used to make sense of the world, then gave up thanks to science.

        And of course he is also serving to remind us what most of the world fears about these sorts of developments, so I think it is important to have his views considered and clarified with respect, not just rejected.

    • Alonso Sharman Roaidz December 17, 2012 on 12:39 pm

      Love, pity, and deity-recognition are the exact opposite of “intelligence.”

    • Drazil Roaidz January 3, 2013 on 9:44 am

      but what if they where to make robots with a real brain in it?
      but then again i wouldn’t want to be robot (something a Little more advanced then that)
      if they look around a little bit they might find some people interested in a new body

      complete with brain computer interface and all the things to keep the brain healthy
      I would be such a person that would be willing to give up my human life

  • Frank Whittemore December 12, 2012 on 5:31 am

    Click below and learn that the Washington, District of Columbia, United States, Executive Office Of The President is interested in brain building –

  • Roaidz December 12, 2012 on 6:57 am

    Robots can not reproduce by itself. They need humans to do so.

    • drhandy Roaidz January 16, 2013 on 1:14 pm

      You clearly have not heard of the reprap, the 3d printer that can print a fully functioning copy of itself

  • Frank Whittemore December 13, 2012 on 12:32 pm

    Looks like the Washington, District of Columbia, United States, Executive Office Of The President is researching the progress in the area of artificial intelligence.

    Visit #2 to my blog. This time from the Hub.

  • Ray Silva December 14, 2012 on 6:59 pm

    Let’s see, a baboon brain has about 14 billion neurons, and a human brain upwards of 85 billion. While numbers alone may not relate to functionality, given that for the simplest task one second performance requires 2.5 hours of number crunching, real-time performance seems quite a distance away.

    • phoenixxl Ray Silva December 14, 2012 on 10:31 pm

      2.5 million , If morre’s law stands and not accounting for more complexity of infrastructure : 20 years.

      85 billion 20 years. after that. Assuming we still use the same binary systems , and parallelism is still done at the same level.

      Since a few months 10 gigabit networks have gone mainstream and is affordable. 100 gigabit is available as well. Once we reach Terabit networking projects like these will easily become distributed and go forward in leaps and bounds.

      Also , making a very slow very complex distributed intelligence has a romantic quality to it imo. I wouldn’t mind something to be created that uses the world’s spare computing power and memory.

      • drhandy phoenixxl January 16, 2013 on 1:15 pm

        network speed is not the bottleneck of distributed computing systems, the issue is management overhead

  • eldras December 18, 2012 on 8:14 am

    NB the synapse has NOT been mapped. It is the most complex machine on earth and we dont know fully how it works yet.

    Any brain simulation is deficient without this.


  • Socrates February 12, 2013 on 8:07 pm

    I know that Prof. Eliasmith would disagree with the ways that some people categorize Spaun and dismiss it as “not adding really innovation or understanding.” In fact, during my interview with him he went into some detail in comparing his brain simulation to all the other major projects in the field and discussing some of the pros and cons of each project.

    You can see his own explanation and judge for yourself if he is convincing enough for you or not by checking out my full interview with him:

  • R.L.Gilkes November 11, 2016 on 5:26 pm

    Is it possible as a way to fix the issue of the hard wiring to instead make the artificial brain out of a large mesh like wire frame type interconnecting computer chips and then write a program set to remember synapse previously learned? Or is my question completely unrealistic?