A tiny worm that can be attached to a computer and control its behavior through a decarbing mechanism has been discovered by a team of researchers from the University of California, Berkeley.
“This is one of the first times that we’ve had a worm that is really capable of doing things that computers aren’t able to do,” said UC Berkeley graduate student and senior author Michael W. Miller.
“This is a very, very early stage, and we are going to need a lot of years to do much more sophisticated work to see if it is capable of much more.”
Miller and UC Berkeley postdoctoral fellow James M. Moore, both of whom were not involved in the research, say their discovery will be a game changer for the field of artificial intelligence.
“The question is whether this is a generalizable mechanism, or if it can be used to do things other than just run a computer,” Miller said.
“We are going from a basic system to something that can do much, much more.
The worm’s ability to learn, to evolve, and to adapt is really a game-changer.”
Miller and Moore developed a technique called decarboxing to remove the need for a computer to read and write the worm’s memory and execute commands.
While a computer can read the worm program, the worm will learn by observing the behavior of the program’s instructions, which are often random or are written out as they are executed.
If the program is a bit-by-bit analysis of the instructions, then a program can be programmed to change the worm.
This is similar to how a human learns, said Moore.
It is possible for a worm to learn new things by watching the behavior and using its decarborization ability to modify its behavior in ways that computers cannot.
To put this into a larger perspective, a computer program can read a program and then modify it in ways it would not be able to modify it.
When the worm decarbates itself to read instructions, it can then make its own changes, without needing to modify the program.
As a result, a human can learn something from a program, and this can be applied to any computer program.
For instance, a student could be able, by watching an artificial intelligence program learn from its own example programs, to modify an existing program to do something they would not have been able to change without modification.
In this case, the program would be modified to take on a different function.
Because a human program is more complex than a worm, a decarbulator will need to be a program with some kind of internal structure to learn from.
A worm decarbonization technique could be used for other kinds of AI systems, such as for teaching robots.
Miller and his colleagues believe their technique is promising enough to warrant more experiments, but they also note that there is still much work to be done before they can begin to develop an algorithm for decarbutizing worm programs.