Home | About AutomatingInvention.com | About Robert Plotkin | The Genie in the Machine | Contact | Disclosures

November 16, 2008

Gamers Solve Problems in Science and Computing

New Scientist reports on an expanding breed of online games that use human problem-solving skills to make progress on cutting-edge problems in science and computing.  For example, the puzzles at www.fold.it require players to manipulate 3D structures to fit into the smallest possible space.  The web site uses the solutions provided by users to help scientists learn about how proteins fold in the real world.

This and other examples provided in the article are examples of ways in which clever uses of distributed computing are increasingly combining software with human problem-solvers to leverage and combine the strengths of each, and to achieve results that could not have been obtain by either computers or humans individually.

Posted by Robert at 5:56 PM | Comments (0)
category: Artificial Invention | Design & Engineering

October 13, 2008

Software Improvises Musical Accompaniments

Two University of Southern California researchers have created software that can create an accompaniment to any song "in the style of any chosen artist, or even the particular style used in select pieces by the artist."

Posted by Robert at 10:00 PM | Comments (0)
category: Artificial Invention

October 5, 2008

Complexity: Computers Come to the Rescue

New Scientist reports on how computers are increasingly being used to solve problems that are too complex for human minds to handle, such as making sense out of traffic patterns and assigning robots to paint trucks coming off the assembly line.

Posted by Robert at 8:18 PM | Comments (0)
category: Artificial Invention

September 11, 2008

Evolutionary Computation Solves Century-Old Algebra Problem

Professor Lee Spector and others at Hampshire College used evolutionary computation "to solve a century-old algebra problem far faster and more efficiently than any past efforts of humans or machines." The problem involved finding formulas useful for designing electronic switching circuits. Although previous solutions had been found, such methods resulted in formulas that were so large that they were useless in practice. In contrast, Professor Spector's evolutionary computation-based technique found useful formulas containing fewer than 300 characters in a few hours of computing time.

Posted by Robert at 7:32 PM | Comments (0)
category: Artificial Invention

September 2, 2008

Evolutionary Computation Community Experiences Explosive Growth

Gregory Hornby and Tina Yu have published the "Results of the First Survey of Practitioners of Evolutionary Computation," which reveals that "in recent years there has been an explosion not only in the different types of biologically inspired algorithms, but also in the number of practitioners in the field." The survey also found that the biggest obstacle for the acceptance of evolutionary computation in industry is that it is poorly understood. The survey was conducted in 2005-2006 by posting 14 survey questions on the web site of the Special Interest Group on Genetic and Evolutionary Computation of the Association for Computing Machinery.

Posted by Robert at 6:00 AM | Comments (0)
category: Artificial Invention

August 15, 2008

Two Heads (One Silicon, One Carbon) Are Better Than One

Using computers to automate inventing does not mean that humans become irrelevant. To the contrary, the most effective kinds of invention automation often involve cooperation between human and computer, a partnership in which each member does what it does best. Interactive evolutionary computation strives to take advantage of such synergies.

Louis von Ahn has developed a specialty in creating computer games which double as human-computer teams for solving problems that neither could solve by itself, such as:

- The ESP Game, which shows the same image to two people and requires them to type in a word describing it, ostensibly to read each others' minds, but also to create a text-searchable database of images (a problem which computer algorithm designers have yet to crack);

- Tag a Tune, a similar game using songs instead of images; and

- Verbosity, in which one player is shown a secret word and must provide clues from which a second player attempts to guess the secret word, all with the effect of creating a database of word meanings.

Posted by Robert at 6:00 AM
category: Artificial Invention | Design & Engineering

August 12, 2008

Software (Re)Invents the Wheel

Evolutionary software has invented the wheel.

Not impressed? Then consider that a circle is not the only shape which maintains a constant height when rolled across flat ground. Can you figure out what the other shapes are? If not, then don't feel bad. They were not discovered until the late 19th century. Yet an evolutionary algorithm rediscovered them in under an hour.

Posted by Robert at 6:00 AM
category: Artificial Invention

August 9, 2008

See Evolved LEGO Structures in Action

Check out the Brandeis DEMO (Dynamical and Evolutionary Machine Organization) web site for videos of LEGO bridges, cranes, tables, and other structures designed using evolutionary algorithms.

Posted by Robert at 6:00 AM
category: Artificial Invention

August 6, 2008

Combining Real and Simulated Evolution for Aircraft Design

Most of the examples of automated inventing described on this web site to date were generated using computer simulations. In contrast, Will Regan, Floris van Breugel, and Hod Lipson of Cornell University have used a combination of simulation and a real-world hardware implementation of a "hovering flapping ornithopter" -- essentially an aircraft with flapping wings -- demonstrate the feasibility of this kind of flight (which, as we all know, is not how the airplanes we know and love work). Their paper includes images of both the simulated and physical models they used.

Posted by Robert at 6:00 AM
category: Artificial Invention

August 3, 2008

Computer Game Opponents Evolve

Steffen Priesterjahn and others at the University of Paderborn in Germany have used evolutionary algorithms to generate smarter computer players for the game Quake 3. They generated a set of players which played with an initial set of strategies, then played them against the standard computer opponent. The strategies of the best-performing players were combined (mated) with each other to produce offspring, some of which were also mutated. After multiple generations this evolutionary process created computer players that were significantly more difficult for human players to beat.

Posted by Robert at 6:30 AM
category: Artificial Invention

July 31, 2008

Automated Invention of a Thrombin Inhibitor

Matrix Advanced Solutions has used its "artificial creativity" software to create an orally-available thrombin inhibitor to act as an anticoagulant. The thrombin inhibitor, which was developed without the use of any expert knowledge, is now in pre-clinical trials.

Posted by Robert at 6:00 AM | Comments (0)
category: Artificial Invention

July 28, 2008

Automating Marketing

Affinnova, Inc. has used its "evolutionary optimization" technology to help its clients optimize ad strategies, product package designs, and product promotions. For example, the company used its software to develop a new logo for Cadbury-Scweppe's 7-Up soft drink by displaying millions of possible combinations of design elements (such as backgrounds, color schemes, and logo placements) to consumers in an online selection process which "evolved" the logo in response to consumer preferences.

Posted by Robert at 10:00 AM
category: Artificial Invention

February 20, 2006

Automating Design

IlliGAL Blogging talks about how a company in Singapore named Genometri is using genetic algorithms in product design.

Posted by Robert at 1:40 PM | Comments (0)
category: Artificial Invention

February 4, 2006

Software That Learns on the Job

TRN Research News reports that researchers at Princeton University have developed two "self-improving" algorithms, a sorting algorithm (which sorts data into a logical order) and a clustering algorithm (which groups similar items together).

Posted by Robert at 7:43 PM | Comments (0)
category: Artificial Invention

January 22, 2006

Computer Science Cuts Across Disciplines

An article entitled "Computer Science Growing Into a Basic Science" describes how increases in computing power are fueling the use of computers "to solve fundamental problems across physical, chemical, biological, engineering, medical and social sciences," and includes several concrete examples.

Posted by Robert at 9:24 PM | Comments (0)
category: Artificial Invention

January 16, 2006

Thoughts on Software Design Automation from 32 Years Ago

The January 2006 issue of IEEE Computer magazine has the following quote from the same magazine in January 1974:

Because of the expense it is not practical to develop design-specific D[esign] A[utomation] systems. Designers of DA systems are thus confronted with the task of building 'large, generalized, flexible' (software) systems with very little design assistance from the computer. If only we had systems which apply the computer and computer techniques to automate (or at least facilitate) the design of software: software design automation! Unfortunately, while the theory underlying the application of comptuers in the design of computing hardware has developed thoroughly, keeping pace (or nearly so) with the developing technology, the implementation of this theory remains a difficult, mostly manual exercise in the design of programs and programming systems.

Posted by Robert at 9:21 AM | Comments (0)
category: Artificial Invention

January 4, 2006

Quantum Computing Comes One Step Closer

ACD points out that the IQOQI (Institute of Quantum Optics and Quantum Information) in Austria has produced the first "quantum byte," consisting of 8 Calcium ions. This development brings the "quantum computer" closer to becoming a reality.

According to the current Wikipedia entry for "quantum computer," "It is widely believed that if large-scale quantum computers can be built, they will be able to solve certain problems faster than any classical computer." The implications for artificial invention are clear. Most of the software that is being used for artificial inventing relies on powerful computers and, perhaps more importantly, provides better results when run on even more powerful computers. As a result, people in the field are always seeking more powerful computers at lower cost. Why design better software when you can improve your results just by running the same software on a more powerful computer?

John Koza began using a 1,000-Pentium computer back in 1999 to run genetic algorithms for (among other things) inventing new hardware and software (see photo). Although today's computers can provide the same performance for about 1/10th of the price of Koza's 1999 system, he was able to achieve impressive results using the technology that was available at the time. Fully quantum computers, if they were to become possible, could take this forward by (pun fully intended) a quantum leap.

Posted by Robert at 6:57 PM | Comments (0)
category: Artificial Invention

December 31, 2005

Artificial Invention

I changed the name of the "Artificial Creativity" category on this blog to "Artificial Invention." Invention doesn't always involve "creativity," and I wanted the name of the category to reflect this. Sometimes invention results from serendipity, brute force trial-and-error, or following design rules. The path leading to an invention may involve some combination of these and other features. So "artificial invention" it is.

Posted by Robert at 8:00 AM | Comments (0)
category: Artificial Invention

December 30, 2005

The Latest in Automatic Programming

MSNBC reports on several examples of "automating programming"--the use of computer software to write other software:

• A program, developed by Doug Smith and others at the Kestrel Institute, that "translates a description of a problem into guidelines a computer can understand," and which has been used "to develop software for scheduling cargo deployment for the U.S. military."
• Software from SciComp that "helps investment banks design programs to price financial derivatives."
• An automatic programming program from Tenfold which "can generate corporate software in three weeks, compared with more than a year when done by hand."

The article ends on a note that is consistent with my posting yesterday: "Still, automated code doesn't yet compare in quality to what's generated by hand . . . . Programmers can rest easy knowing that their jobs are safe--at least for now."

Posted by Robert at 7:30 AM | Comments (0)
category: Artificial Invention

August 11, 2005

Never send a computer to do a human's job, especially if the human works for free

Although this blog is about computer automation, humans still outshine computers in the ability to make aesthetic judgments. Despite advances in automated image processing, for example, computers still have a long way to go in recognizing the contents of a photograph or judging whether a new clothing design would be visually appealing to customers.

Interactive evolutionary computation attempts to provide the best of both worlds by combining the ability of computers to generate new designs with the ability of humans to evaluate their aesthetics. Techdirt writes about a researcher at Carnegie Mellon University who has created two online games (here and here) that are fun to play in their own right, and which use the input of the games' users to improve the ability of computers to search for and recognize digital images.

The human players of these games provide descriptive labels of images they are shown and point out key portions of such images, two tasks that computers perform poorly. The human input, however, can then be used by computers to better search for and recognize subsequent images. For example, if many human players of the first game have labeled images of elephants with the word "elephant," when someone then performs a search for "elephant" images, computer software can easily pull up the right pictures just by searching through the human-provided labels, rather than by attempting to recognize the images themselves.

Although I don't believe that either of these games uses interactive evolutionary computation, the philosophy behind both is the same: to form a partnership between computers and humans, using each for what it does best. And when the humans provide input for free, deciding whether to incorporate their superior aesthetic judgments into computer software is a no-brainer.

Posted by Robert at 7:32 AM | Comments (0)
category: Artificial Invention | Design & Engineering

August 10, 2005

Digitizing know-how

IPcentral ponders the difficult question of who should own the technical know-how that is inside the heads of workers at high-tech companies. The posting was motivated by a recent court ruling that temporarily bars a former Microsoft employee from performing search-related work for his new employer, Google, because doing so would violate his non-compete agreement with Microsoft.

Trade secret law and non-compete agreements have long been used to control the movement of know-how and other information stored in the heads of human scientists, engineers, and programmers. But what happens when we "bottle" such know-how, or its equivalent, in the form of software that can design machines and write software? You might think that a company that develops an improved genetic algorithm that assists it in designing new machines should maintain that algorithm as a closely-guarded trade secret. After all, isn't the algorithm the functional equivalent of an engineer's know-how within the framework of the company's business model?

But I don't think the answer is entirely obvious. Perhaps the company should seek a patent on the algorithm, thereby obtaining a period of time in which it can block competitors from using the same algorithm even if they develop it themselves independently. Or maybe they should use some combination of intellectual property protection and licensing mechanisms to secure the maximum value to the company.

The point is that transferring know-how from a human mind to software raises some tricky legal and business considerations that will need to be addressed as the automation of invention continues.

Posted by Robert at 10:07 AM | Comments (0)
category: Artificial Invention | Design & Engineering | Intellectual Property Law | Technology Industry

July 22, 2005

Artificial inventions in simulated worlds

ACD links to an article in New Scientist describing an upcoming simulation of human behavior in an artificial world. One of the goals is to see whether culture will emerge in the simulation.

And what if the simulated humans discover fire, invent the wheel, or invent something completely new? Who would own the patent rights?

I suppose the next step would be for the simulated humans to write a (simulated) simulation of (simulated) humans, and so on, and so on . . .

Posted by Robert at 8:28 AM | Comments (0)
category: Artificial Invention | Intellectual Property Law

July 18, 2005

Imagination Engines Launches New Web Site

I just noticed that Imagination Engines, founded by Steven Thaler, has launched a new web site. The company describes itself as follows:

Imagination Engines is a small company working with what many have recognized as potentially the biggest idea in history, a technology that can invent everything else. Accordingly, largely due to issues of credibility, the company's road to success has been rocky. There have been many skeptics and critics, but there have been more believers and supporters. Now the company thrives upon a significant contract stream and tangible products that speak louder for the technology than words possibly could.

I believe I first heard about the company when I read an article (such as this one) describing how the company had used its patented Creativity Machine to invent the Oral-B CrossAction toothbrush. Although genetic algorithms seem to be getting most of the attention these days, Imagination Engines' Creativity Machine relies on neural networks.

Posted by Robert at 4:52 PM | Comments (0)
category: Artificial Invention | Technology Industry

June 21, 2005

European Parliament Renames the "Software Patent Directive"

ZDNet UK has an article reporting that the European Parliament has proposed changing the name of the so-called "software patent directive" to use the term "computer-aided invention" in place of the old term, "computer-implemented invention." For those of you who can't see what difference this could possibly make, the stated reason for the proposed change is to make it clear that software is not patentable per se, but rather only as part of an innovation that uses software "to aid the performance of the invention."

These and other attempts over the years to find just the right term to define patents on software are, in my view, doomed to fail. One motivation for my "software patent puzzle" (see parts 1 and 2) is to demonstrate that there is no principled way to use the hardware/software distinction as a basis for distinguishing patentable inventions from unpatentable ones.

Instead, I propose that we focus our attention on the patentability of "computer-generated inventions." Examples of computer-generated inventions include software (a computer generates software when you program the computer) and any device whose design is generated by a genetic algorithm or other artificial creativity software.

It is the "computer-generated" feature of software that keeps causing problems for patent law, and that will continue to cause problems for patent law as computers automate the invention of things other than software. I can't justify that claim in a single blog posting, but I tried to make the basic argument here, and will continue to extend the argument in future postings.

Posted by Robert at 4:53 PM | Comments (0)
category: Artificial Invention | Design & Engineering | Software Patents

June 16, 2005

Blogging from GECCO and NASA/DOD EH Conferences

I will be attending and blogging from the Genetic and Evolutionary Computation Conference (GECCO 2005) and the NASA/DOD Conference on Evolvable Hardware from June 27-July 1. (Unfortunately I will miss the first two days of GECCO.) These conferences will provide tutorials and updates on the latest developments not only in computer-automated design, but more generally in evolution-inspired computing.

Instead of attempting to duplicate or compete with Illigal Blogging's live coverage of the GECCO conference, I plan to post my thoughts on the legal and broader social implications of the topics covered in the conference sessions. I will also interview presenters and attendees for a book I am writing on computer-automated invention and the law. (I may post excerpts/summaries of those interviews here, with the interviewees' permission.) If you would like to speak with me, just track me down at the conference (you can see what I look like here).

Posted by Robert at 10:50 AM | Comments (0)
category: Artificial Invention