The Future of Robots

The aspiration to build human-level androids can be regarded as the ultimate challenge in artificial intelligence. To do it, we need to understand not just human cognition but also our physical skill-it is, after all, a critical part of what the brain does. Coordinating intention with movement in a complex environment is largely the responsibility of the cerebellum, which comprises more than half the neurons in the brain. And the body itself represents much of our complexity: There is more information in the human genome, which describes the human body, than in the design of the brain.

We are making tremendous strides toward being able to understand how the brain works. The performance/price ratio, capacity and bandwidth of every type of information technology, electronic and biological alike, is doubling about every year. I call this pervasive phenomenon the law of accelerating returns. Our grasp of biology is proceeding at an accelerating pace, also exponentially increasing every year. It took scientists five years to be able to sequence HIV, for example, but the SARS virus required only 31 days. The amount of genetic data that´s been sequenced has doubled every year since the human genome project began in 1990, and the cost per base pair has come down by half each year, from $10 in 1990 to about a penny today. We are making comparable gains in understanding how the genome expresses itself in proteins and in understanding how a broad range of biological mechanisms work. Indeed, we are augmenting and re-creating nearly every organ and system in the human body: hearts and pancreases, joints and muscles.

The same progression applies to
our knowledge of the human brain.
The three-dimensional resolution of brain scans has been exponentially increasing, and the latest generation
of scanners can image individual neuronal connections firing in real time. The amount of data that scientists are gathering on the brain is similarly increasing every year. And they are showing that this information can be understood by converting it into models and simulations of brain regions, some two dozen of which have already been completed. IBM also recently began an ambitious effort to model a substantial part of the cerebral cortex in incredible detail.

If we are to re-create the powers of the human brain, we first need to understand how complex it is. There are 100 billion neurons, each with thousands of connections and each connection containing about 1,000 neural pathways. I´ve estimated the amount of information required to characterize the state of a mature brain at thousands of trillions of bytes: a lot of complexity.

But the design of the brain is a
billion times as simple as this. How
do we know? The design of the human brain-and body-is stored in the genome, and the genome doesn´t contain that much information. There
are three billion rungs of DNA in the human genome: six billion bits, or 800 million bytes. It is replete with redundancies, however; one lengthy sequence called ALU is repeated 300,000 times. Since we know the genome´s structure, we can compress its information to only 30 million to 100 million bytes, which is smaller than the code for Microsoft Word. About half of this contains the design of the human brain.

The brain can be described in just 15 million to 50 million bytes because most of its wiring is random at birth. For example, the trillions of connections in the cerebellum are described by only a handful of genes. This means that most of the cerebellum wiring in the infant brain is chaotic. The system is designed to be self-organizing, though, so as the child learns to walk and talk and catch a fly ball, the cerebellum gets filled with meaningful information.

My point is not that the brain is simple, but that the design is at a level of complexity that we can fathom and manage. And by applying the law of accelerating returns to the problem of analyzing the brain´s complexity, we can reasonably forecast that there will be exhaustive models and simulations of all several hundred regions of the human brain within about 20 years.

Once we understand how the mind operates, we will be able to program detailed descriptions of these principles into inexpensive computers, which, by the late 2020s, will be thousands of times as powerful as the human brain-another consequence of the law of accelerating returns. So we will have both the hardware and software to achieve human-level intelligence in a machine by 2029. We will also by then be able to construct fully humanlike androids at exquisite levels of detail and send blood-cell-size robots into our bodies and brains to keep us healthy from inside and to augment our intellect. By the time we succeed in building such machines, we will have become part machine ourselves. We will, in other words, finally transcend what we have so long thought of as the ultimate limitations: our bodies and minds.

Inventor and futurist Ray Kurzweil is the author of five books, including, most recently, The Singularity Is Near: When Humans Transcend Biology