IBM to build self-healing computers

[Guest.Visitor]

Ralph you may want to think how you bad mouth other careers around here some of us are not programmers. I'm sure the rest us lazy Administrator types could give our opinions of programmers...... Ralph says...... "Last I checked, everybody and their mother wants to be an administrator so they don't have to strain themselves with all that thinking that's required to program." Brian Babineau Systems Administrator

[R.Daugherty]

I was a computer operator for 3 years in the 70's (sys 3) and the Novell 3.11 admin as well as client/server programmer for 2 years in early 90's in an AS/400 shop. Admin is a full time, hard job. I was referring to all the people who I happen to see on the net who are entering schools and ask "should I learn Java or network administration?" If they initially answered programming, they quickly end up in network admin. Then their eyes glaze over with TCP/IP, IPX, OSI, hardware, software, commands.... it's not easy. But as I said, everyone and their mother says they're going into network administration... Ralph

[B.Myrick]

Incidently, this process could actually be made workable. There is an ongoing and now successful attempt by IBM to mass produce Carbon Nano Tubes. Based on the carbon atom, the strength of such an item could be akin to a diamond, or as pliable as copper. IBM has just announced (see CNN.COM) the successful test of the CNT. It has been possible in the past, but with random results in the growing process. Now the process is (supossedly) refined to the point that the CNT is created (grown) as designed. This gives us a simple linear chain of carbon less than 10 atoms across, that can replace copper in chips. The high speed conductivity of the CNT vs copper is too great for me to grasp, let alone explain. What it does do, is put us that much closer to actually being able to send a signal, via hardwire, just that much closer to the speed of light. I can envision a process wherein series of circuits based on the CNT can recognise their failure and 're-route' their neural pathways via self defined shunts and/or self produced (grown) CNT pathways. Once the base software is designed, it would not be out of the question to actually have a voice interpretational system that generates code, within the limits of it's base software, from fractional ideas of the programmer/user. Very simply put, a self learning lattice for a super 4GL. Just food for though from a lazy programmer, who used to be a lazy operator, too..... -bret

[David Abramowitz]

Ralph Daugherty wrote: they're spending a quarter of their research dollars on this. Yeah, , , , , , but they can afford it. Dave

[R.Daugherty]

sounds interesting Bret, but doesn't redundant non-stop systems already handle hardware failures? and doesn't dual side arbitration already handle software failures? what you're suggesting is constant state of unknown versus switching automatically to side B and replacing a board. For this AS/400 research dollars get spent? Ralph

[B.Myrick]

Interesting yes, but at least decades away. IBM has not (that I know of) combined the 'self healing' and Carbon Nano Tubes into one document or press release. I am being somewhat liberal in expousing my vision. Regardless of the ability of the AI to act alone, it must rely on a software set that periodically or constantly analyses itself for defects. Once the defect is found, it can begin restoring itself though healing or actual replacement by constructing new pathways via the CNT. This would mean that the AI would most likely be required to 'surgically' remove the anomaly within itself. Such a possiblity could very likely lead to the development of the positronic brain envisioned by Isaac Asimov and his famous "Robot" and "Foundation" series. Not to mention my favourites, the "Lucky Starr" series he wrote under the psuedonymn of Paul French. Now the sinister part of this. Suppose that such an AI with the ability to heal itself, becomes cognizent. The last thing we need running around is a maniacal "Johnny 5" with the ability to self repair and recognise new threats combined with the capability of countering such threats. Remember the lousy movie, "Colosus: The Forbin Project"? Or worse, they CyberDyne created AI that evolved into Terminators? Myself? I would rather see these mniscule CNT utilise to repair damaged neerve endings. A CNT could conceivably be developed to repair nerves, spinal fibres and possibly bone/joint replacement materiels. This could impact the requirement for stim-cells in such research and thus negate the requirement for such contraversial materiel. -bret

[R.Daugherty]

Bret, if that's based on the book that I'm thinking of, then I read the book. It's the first time I saw the phrase "reprogrammed itself". On the other hand, that's my definition of artificially intelligent, that is generating new code and/or combinations of existing code to perform tasks for which it had not been programmed. Neural networks, datasets, and anything else that records info and has existing algorithms applied to determine actions to take are just more sophisticated versions of what we do every day. Programs that perform tasks that look like thinking, such as chess or image analysis, have always been categorized as AI in the beginning until the hype wears off, then they're just programs and the next level of "thinking" becomes AI. I wrote Double Deck Pinochle in '81. Why was chess AI and DD Pinochle just a TRS-80 program? Level of sophisitication of the algorithms, but they're still fundamentally algoriths. Self-generated code will be created by algorithsm, but the algorithms then have the potential of being modified, and the new code created with existing or modified algorithms creates self-directed action, resulting in true AI. Regardless of the results of neural network analysis, I believe the range of actions that can be taken are predetermined in existing code. Thinking about it, I guess that if a wide range of human generated code, on the net for example, is able to be identified in a search by the program as a desired action to take based on analysis, and the program is able to generate parms to call the code with, that that qualifies as well as generating it's own source code. Just being a smarter programmer, right? Ralph

[B.Myrick]

One of the first 'AI' programs I wrote. In the mid 80's I wrote a tic-tac-toe program for a CPM based 'Eagle' PC. It was unbeatable because it had my strategem built into it. The program could not adapt or temper itself based on the players level of skill. It was just plain remorseless. My employer thought it was genius, even though we both know there are distinct mathematical probabilities (actualities) for winning the game. I never considered it AI, but my boss hyped it as such and it did help get me a 10% raise that year. As for true AI, being an arm-chair theorist (as Dr. Watson would say) I have to admit to being ignorant of the innerworkings of AI, but still reference it akin to an operating system. Given a task, the parameters and resources, the answer can be derived. Any thinking on the part of the AI is however, a product of its interpreted instructions. To learn, the AI has a set of algorithms designed to take input from a source, identify it, determine it's intent (potential or actual) and derive a set of instructions to interact (or counteract) the intrusion. A true AI with the power of self healing could propose a true threat. Even if built in safeguards to the OS or the actual framework of the device are installed, the neural networks of the AI could (in theory) identify such device, grow CNTs to bypass the device and eventually excise the "tumor", so to speak. As for installing a command in the OS to contradict the AI's function or orders of self mastery: They tried that with HAL and he (for a short time) went quite insane. BTW, I think I had a copy of Double Deck Pinochle on my TRS-80 IV. Had the standard Tape Drive I by-passed that and built a connection that interfaced with my Reel-to-Reel, hard drive, Floppy and dot matrix printer. I built a modem connector and hooked up with a unit I acquired from the local University (legally) and hit my first BBS and the Internet during that time. Had poor graphics capabilities but I still remember the hours and fun I had learning as I went. -bret

[R.Daugherty]

Man, that would be too cool if you had my Double Deck Pinochle cassette tape! I pushed the boundaries a little bit on developing that with 32K of memory and a cassette tape drive (in other words, no floppy or hard drive). I wrote the program in 7 modules and hand linked them by defining the address of every label accessed in any of the other 6 modules, generated 7 object binaries, loaded them back to back in memory, then wrote to tape from beginning to end to create an executable program. That was what I would call a learning experience. The patience required to save and load from tape alone was unbelievable. Did you ever try to play the game, Bret? I later rewrote it for CP/M and then DOS. Never got around to a Windows version. I want to rewrite it in Java and make it an automatic seat in Yahoo's Double Deck Pinochle. Ralph

[B.Myrick]

Ralph, I tried a couple of times as I thought pinochle would be a sophisticated game to learn. Too bad I was not sophisticated enough to learn it. I'll stick with some good ol' Draw or Stud Poker!!! That's a lot of work! I wrote a report writing utility for a client and it was not so difficult. 'Course I had Arrays, Logical Files, Data Structures.... -bret