Home / Blog / Faster than a human brain. Smarter? That’s not a given

Faster than a human brain. Smarter? That’s not a given

The impressive numbers of Cerebras CS-2: all together it can support 120 trillions parameters processing, beating the 100 trillion synapses that process data in our brain. Image credit: Cerebras

A significant number of scientists, but not all, are convinced that if we were to replicate the neurones functionality and connectivity of a human brain in a machine we will get… a human brain. This means we will get the same characteristics generated by our brain, including intelligence in its various forms (creativity, social empathy, math skills, language, awareness of the context, sense of self, …).

Hence the race to create such a machine. The challenge is staggering. Our brain has some 100 billion neurones and 100 trillion synapses. Those are BIG numbers but the real challenge is the connectivity.  Its connectivity is not just huge, it is complex (cannot be reduced) and keeps changing over time in any given person, plus it is different from person to person (it is unique to each person even in case of twins), so it is a nonsense to say that we can copy the “human” brain connectivity in a machine. At the very best (still impossible with today’s technology) you would be able to create a mirroring image of a given brain at a given time.

Assuming this will ever be possible would that machine have the same thoughts of that person at that time? The answer is NO. The fact is that you would have to “copy” also the status of each neurones (its excitation level) and this seems beyond any foreseeable technology. However, one might ask if that machine would react in the same way as that mirrored brain when exposed to the same stimuli, like seeing a face. Again, the answer is NO but we could expect a quite similar reaction because the same kind of processing will be going on.

Anyhow, scientists are not searching for a way to duplicate thoughts (or emotions, feelings) in a machine. Their quest is for ways to endow a machine with the same “capabilities” of a human brain in terms of tackling problems, evaluating and finding solutions. The mechanistic approach is betting on creating structures mirroring the ones we find in the human brain. It should be noted that even those that do not agree with the mechanistic approach agree that in order for a machine to display “intelligence” it needs plenty of processing power and the capability to concurrently evaluate alternatives.

Of course, it should be no surprise that everybody agrees with this requirement since it has been proven by results that as we increase processing power and parallel processing we get “smarter and smarter” results.  The progress in AI in these last decade has been staggering, probably exceeding the most optimistic expectations we had 20 years ago, and that is thanks to processing and big data. More recently it is thanks to data and much more processing!

Everyone is also recognising that the wiring of a brain (doesn’t matter which brain species) is supporting smart processing, hence again it is no surprise that researchers have been striving to create processing structures mimicking neuronal circuits, the so called neuromorphic computing, and creating chips that support massive parallel processing.

Cerebras Systems is on the leading edge of this trend and has now released a second version of its gigantic chip, the CS-2, with a surface of 46,225 mm², that with its 2.6 trillion transistors can support 120 trillions parameters, vs the 100 trillion parameters that are believed to be supported by our brain. In the figure you can see other impressive numbers of this new chip, providing a 100 times increase in processing parameters capability (see the previous version in the clip).

This chip will allow the processing of “brain-size” models in terms of variables, hence comparing in terms of complexity management to a real human brain. However, this does not imply that it will be as smart as a brain NOR that it cannot be smarter! The overall brain processing is different from the one of the machines we have today, also when we are introducing software processing to make the hardware structures more flexible, and to a point mimic in software what cannot be mimicked through hardware. The truth is we are yet in the dark to understand our brain and what makes “intelligence” arise and tick.

What we have been able to create are systems that seen from the outside can take smart decisions, sometimes as smart and even smarter than the ones we can take as humans. As a matter of fact it is nowhere written that the only intelligence possible is the one emerging from our carbon based brain, silicon may eventually prove to be as good, may be better, although it can use a different approach to intelligence.

About Roberto Saracco

Roberto Saracco fell in love with technology and its implications long time ago. His background is in math and computer science. Until April 2017 he led the EIT Digital Italian Node and then was head of the Industrial Doctoral School of EIT Digital up to September 2018. Previously, up to December 2011 he was the Director of the Telecom Italia Future Centre in Venice, looking at the interplay of technology evolution, economics and society. At the turn of the century he led a World Bank-Infodev project to stimulate entrepreneurship in Latin America. He is a senior member of IEEE where he leads the Industry Advisory Board within the Future Directions Committee and co-chairs the Digital Reality Initiative. He teaches a Master course on Technology Forecasting and Market impact at the University of Trento. He has published over 100 papers in journals and magazines and 14 books.