Machine Intelligence: Ray Kurzweil
Era of smart people is dawning

In 1990, Raymond Kurzweil described in his book The Age of Intelligent Machines how computers would come to be able to recognize human speech, spinning predictions that came true in less than a decade. This year, the pioneering inventor has released a bold new volume that projects his beliefs about machine intelligence deep into the next century, to envision a time when computers evolve far beyond the abilities of humans.

In his book The Age of Spiritual Machines (Viking, 1999), Kurzweil posits processors that will out-perform humans in every field of endeavor by 2050. "By the middle of the next century, machine intelligence will be so superior to human intelligence as we know it today, that in order to be intellectually relevant we will have to enhance our brains with neural implants," said Kurzweil, whose accomplishments span the development of an opti-cal character-recognition system, a speech-recognition system, scanners and innovative music keyboards.

"Of course, ultimately we won't bother with neural implants, because by the middle of the next century we will be able to scan a human brain and instantiate it into a neural computer. Our old carbon-based bodies are going to be obsolete.

"Even the concept of a body will become unclear, since you could run your virtual personality on a network of computers, or you could have multiple neural personalities all running on a single neural computer," said Kurzweil. "There are just not going to be the kinds of boundaries we have today. It's going to be difficult to determine where one person ends and another person starts."

Eventually we will be sharing virtual-reality experiences with each other that are as convincing as real experiences, Kurzweil believes. We will be able to stroll down virtual beaches in our virtual bodies with tactile feedback so convincing that in his view, we will prefer sharing experiences with other virtual personalities to the real thing. We will do things with each other in VR that will have consequences in the real world, according to Kurzweil, but they won't be the same consequences that would normally occur in the real world. Eventually we won't even need any kind of external equipment to share virtual experiences with others, at least not any equipment that is not already resident within our enhanced brains.

"By 2019 we will have hardware for under $1,000 that equals the processing power of the human brain-about 20 million billion calculations per second-but it will take another 10 years to organize these resources and provide the software to match human intelligence," he said. "Once machines become as intelligent as people, they will necessarily soar past us. Because computers evolve much faster, they can remember things much more accurately, and they can share information much more efficiently than people can. But I want to stress that this is not an alien invasion of intelligent machines-it is a phenomenon arising out of our own society. Neural implants will be how people keep up with the machines, at first anyway. Eventually, there won't be a clear distinction between where a person stops and where his implants begin, but without neural implants human learning will just be too slow. DNA-based evolution is extremely slow too, and the protein synthesis of our bodies suffers from too many limitations, so eventually we will even shed our carbon-based bodies in favor of electronic circuitry, which is much faster, smaller and more flexible," Kurzweil said.

These bizarre predictions would be all too easy too shrug off as sheer science fiction, were it not for Kurzweil's credentials. A pioneer in music, speech recognition and text-to-speech educational software, Kurzweil was named Inventor of the Year a decade ago, in 1988, by the Massachusetts Institute of Technology. In 1994 Kurzweil was awarded Carnegie Mellon's top science award, the Dickson Prize, and he won the Association of the American Publishers' award for the most outstanding computer-science book, The Age of Intelligent Machines, in 1990.

Kurzweil has been a busy man. He founded and sold off three companies that bear his name-Kurzweil Applied Intelligence Inc., Kurzweil Technologies Inc. and Kurzweil Educational Systems Inc. He still serves as consultant and board member on all three.

He developed the first omni-font optical character-recognition system, the first print-to-speech reading machine for the blind, the first CCD flatbed image scanner, the first text-to-speech synthesizer, the first computer music keyboard capable of reproducing orchestral instruments and the first large-vocabulary speech-recognition system.

He is the recipient of the Grace Murray Hopper Award from the Association for Computing Machinery and served as chairman for innovation at the White House Conference on Small Business during President Reagan's administration.

Kurzweil graduated from MIT, but has since received nine honorary doctorates in science, engineering, music and humane letters from leading colleges and universities worldwide. The Age of Intelligent Machines, was chosen as the Most Outstanding Computer Science Book of 1990 by the Association of American Publishers, and his companion documentary film received seven national and international awards, including the CINE Golden Eagle Award and the Gold Medal for Science Education from the International Film and TV Festival of New York.

Assessing the past year, Kurzweil cites five significant technology developments in 1998 whose exponential growth will eventually result in the spiritual machines of the next half century. The biggest harbinger is speech recognition, which went mainstream in 1998, and which promises to become ubiquitous over the next few years. The others are neural and evolutionary-based investment funds, the retinal display, full-body LANs and automated personalities.

"In 1998 we saw the culmination of a decade of engineering effort in speech recognition, which will become ubiquitous as an alternative to keyboard input in the next century," said Kurzweil. "When we started on speech recognition, the computers would not have been fast enough to run the algorithms that took us a decade to create, but now even small handheld devices have processors fast enough for speech interfaces."

Speech recognition, more than any other technology, hit the knee of its exponential growth curve in 1998, according to Kurzweil. In 1999, he said, speech recognition will become a routine part of the user interface of at least one company's device in every field that has user interfaces.

Speech-recognition technology presages the future, according to Kurzweil, not only because of its imminent success, but also because of the kind of engineering effort that had to be mounted to solve it.

"What is interesting is that we didn't really teach our computers what we thought the rules were for recognizing, say, phonemes, but actually let the learning algorithm discover these things for itself by exposing it to thousands of hours of speech. We slowly taught it in the same painstaking way you would teach a human, and it actually learned a little bit analogously to a human. It made mistakes and we corrected it, and gradually it learned speech recognition over the last 10 years," said Kurzweil.

That decade of painstaking speech-recognition work, plus the exponential increase in available computing power, enabled speech recognition. However, the next leap in technology won't be so long in coming, according to Kurzweil, because computer knowledge offers a striking advantage over biological knowledge: It can be directly downloaded from one computer to another.

"The big difference between humans and intelligent machines is that we only have to teach one," said Kurzweil. "Unlike humans, each one of which has to be taught speech from scratch, we can instead download the speech knowledge learned by one computer into other computers. Eventually all machines will be able to master all human and machine-acquired knowledge."

Solving the speech-recognition problem thus demonstrated that the complex systems of the future will be too intricate to engineer in the ordinary, piecemeal fashion. Increasingly, what will be needed is an algorithm that can be refined automatically.

Its fruits can then be instantly downloaded to other computers. "For instance, a computer will be able to download a fluency in French, in a way that people cannot-there are no direct-memory access ports on humans," said Kurzweil.

Another 1998 technology ready for exponential growth, according to Kurzweil, is investment-funds automation. Today most buy/sell decisions in large funds are controlled by program trading, which automates buy/sell functions according to preset rules written for it by humans. Kurzweil expects most funds to start incorporating neural networks and evolutionary algorithms into their program trading in the next millennium.

Also a technology milestone in 1998, according to Kurzweil, was the retinal display that projects virtual realities directly into the biological eye with a laser. The retinal display heralds an era of non-invasive interface technologies that allow people to directly access the information.

The retinal display achieves that by writing its images directly onto the retina of the eye, transferring pixel data to the brain without any physical connection.

Full-body LANs using wireless technologies, according to Kurzweil, will complement subsystems like the retinal display by connecting all the computing devices you carry in your pockets, such as automatic language translators and holographic projectors.

In 1998 the first standards were proposed for full-body LANs. Eventually, when computers surpass humans in intelligence, according to Kurzweil, and humans augment themselves with neural implants, the body-LAN might be superseded by putting the real brain online with our other computing devices.

"People are already starting to get neural implants to cure disabilities, but eventually neural implants will be able to enhance our perceptions, our memory, our logical abilities and an endless spiral of higher-order abilities unimaginable today," said Kurzweil.

"Already we have implants for nervous disorders, cochlear implants for hearing, but eventually we won't even bother with implants. The new materials will be so superior to biological materials that we will download our entire personalities into them."

Return to: TimesPeople 98