Google’s Laws For Robots

Google’s AI Researchers are writing their own set of “RULES” or guidelines to be best described on how robots should act. In a new paper called “Concrete Problems in AI Safety,” Google Brain—Google’s deep learning AI division—lays out five problems that need to be solved if robots are going to be a day-to-day help to mankind, and gives suggestions on how to solve them.


Most of the discussion to date has, argues Olah, been ‘very hypothetical and speculative,’ so the team wanted to examine the real-life challenges. It came up with five issues.

  • Avoiding Negative Side Effects: How can we ensure that an AI system will not disturb its environment in negative ways while pursuing its goals, e.g. a cleaning robot knocking over a vase because it can clean faster by doing so?
  • Avoiding Reward Hacking: How can we avoid gaming of the reward function? For example, we don’t want this cleaning robot simply covering over messes with materials it can’t see through.
  • Scalable Oversight: How can we efficiently ensure that a given AI system respects aspects of the objective that are too expensive to be frequently evaluated during training? For example, if an AI system gets human feedback as it performs a task, it needs to use that feedback efficiently because asking too often would be annoying.
  • Safe Exploration: How do we ensure that an AI system doesn’t make exploratory moves with very negative repercussions? For example, maybe a cleaning robot should experiment with mopping strategies, but clearly it shouldn’t try putting a wet mop in an electrical outlet.
  • Robustness to Distributional Shift: How do we ensure that an AI system recognizes, and behaves robustly, when it’s in an environment very different from its training environment? For example, heuristics learned for a factory workfloor may not be safe enough for an office.


Fastcodesign present them in a concise and easy to understand way.

  1. Robots should not make things worse.
  2. Robots shouldn’t cheat.
  3. Robots should look to humans as mentors.
  4. Robots should only play where it’s safe.
  5. Robots should know they’re stupid.



World’s First Fully Programmable and Reconfigurable Quantum Computer Module

A team of researchers, led by Professor Christopher Monroe from Joint Quantum Institute (JQI) at the University of Maryland Physics , has introduced the first fully programmable and reconfigurable quantum computer module in a paper published as the cover article in the August 4 issue of the journal Nature.

For more Info: Read Quantum Computing.



Photo of an ion trap.


The new device, dubbed a module because of its potential to connect with copies of itself is made of five individual ions, charged atoms — trapped in a magnetic field, whose strength is manipulated in such a way that the ions are arranged in a line. A computer program dedicated to solving a particular problem—on five quantum bits, or qubits—the fundamental unit of information in a quantum computer. Quantum computers promise speedy solutions to some difficult problems, but building large-scale, general-purpose quantum devices is a problem fraught with technical challenges.

“For any computer to be useful, the user should not be required to know what’s inside,” said Monroe, who is also a UMD Distinguished University Professor, the Bice Zorn Professor of Physics, and a fellow of the Joint Quantum Institute and the Joint Center for Quantum Information and Computer Science. “Very few people care what their iPhone is actually doing at the physical level. Our experiment brings high-quality quantum bits up to a higher level of functionality by allowing them to be programmed and reconfigured in software.”

The reconfigurability of the laser beams is a key advantage, Debnath says. “By reducing an algorithm into a series of laser pulses that push on the appropriate ions, we can reconfigure the wiring between these qubits from the outside,” he said. “It becomes a software problem, and no other quantum computing architecture has this flexibility.”

To test the module, the team ran three different quantum algorithms, including a demonstration of a Quantum Fourier Transform (QFT), which finds how often a given mathematical function repeats. It is a key piece in Shor’s quantum factoring algorithm, which would break some of the most widely-used security standards on the internet if run on a big enough quantum computer.

Two of the algorithms ran successfully more than 90 percent of the time, while the QFT topped out at a 70 percent success rate. The team says that this is due to residual errors in the pulse-shaped gates as well as systematic errors that accumulate over the course of the computation, neither of which appear fundamentally insurmountable. They note that the QFT algorithm requires all possible two-qubit gates and should be among the most complicated quantum calculations.

The team believes that eventually more qubits—perhaps as many as 100—could be added to their quantum computer module. It is also possible to link separate modules together, either by physically moving the ions or by using photons to carry information between them.


(via Science Daily, IBT)