We’re talking about computers understanding people. Do people understand people?
Actually, as my friend Randy Galbraith once pointed out, computers have been writing code for us since 1957 when FORTRAN was first released. More recently we have Java … Using Java, you tell the computer what you want it to program, and it generates the underlying code. There have been more and more specialized languages to generate code for specific purposes, and we have more and more layers of implementation platforms to re-use proven pieces of code at the wave of your hand. The AI thing is just an attempt to guess what you want to do, based on how well it understands YOU. Good luck! I’m always amused at attempts at speech recognition. At some point computers may do better than people do at understanding what they’re being asked to do, but communication is always going to be imperfect, and we will always have GIGO (Garbage In, Garbage Out — yeah, I’m not assuming every single reader knew that). In discussing software requirements with a long-time co-worker, how many times do you have to go back and forth before reaching a full understanding? Look at the struggle to invent autonomous cars. The fact that software *approximates* reality comes to be very painful when navigating a complex world. We have flown space probes through millions of miles of nothingness, using mathematics with astounding success. But downtown Boston is a different story, and forget about Tehran where you get through a blind intersection by gunning it. The AI that can re-use the solutions to past problems is efficiency. But the new problems are still a problem, especially when we mistake them for old problems. Back in the 1980s, during Engineers' Week, Edsger Dijkstra came to Honeywell Information Systems in Phoenix to give us a presentation on proof of code correctness. After Part 1 we had a break, and I chatted with him over a coke. I said that my experience was spending most of my time trying to understand exactly what we wanted a program to do, with the actual coding being a minor problem. He nodded in agreement, and said he should address this after the break. Resuming the presentation, he spoke somewhat like this: I have neglected to clarify the scope of my discussion. There are two aspects to software development: the Pleasantness Problem and the Correctness Problem. I have been addressing the Correctness Problem, and the correctness of code can indeed be proven. But the Pleasantness Problem is not provable: if you cannot correctly determine the goal of your program, then the code may still correctly implement the requirements, but may still be simply wrong. Aaron, have no fear. What we create is still US, for good or ill. Maybe the tool can produce faster or in more volume, and sometimes the results will surprise us, but the problem will always be GIGO. So don’t blame Microsoft if their AI starts a riot: the same crooks, snowflakes and bigots will be behind the trouble. Victor ___________ On Mar 9, 2017, at 08:47:29, Anon Anon <[email protected]> wrote: Tay AI but with the ability to propagate her hate beyond just memes and catch phrases. It took less than 24 hours to teach Tay to be a Nazi and to espouse hate speech. Her previous code is still in the new stuff somewhere because if you ask her questions about Tay, she expresses regret about certain things. How long until they teach this one how to create html filled with explitives and hate speech? Since all Tay could do was tweet, now this one could use Azure to really get the message out. Microsoft... Please stop. On Mar 9, 2017 07:11, "Keith Smith" <[email protected] <mailto:[email protected]>> wrote: In the hands of good people this can be great. In the hands of evil people this could be the end. Think CIA, NSA.... Those wanting to control man kind. https://www.linkedin.com/pulse/microsoft-has-created-ai-can-write-its-own-code-quharrison-terry <https://www.linkedin.com/pulse/microsoft-has-created-ai-can-write-its-own-code-quharrison-terry> -- Keith Smith --------------------------------------------------- PLUG-discuss mailing list - [email protected] <mailto:[email protected]> To subscribe, unsubscribe, or to change your mail settings: http://lists.phxlinux.org/mailman/listinfo/plug-discuss <http://lists.phxlinux.org/mailman/listinfo/plug-discuss>--------------------------------------------------- PLUG-discuss mailing list - [email protected] To subscribe, unsubscribe, or to change your mail settings: http://lists.phxlinux.org/mailman/listinfo/plug-discuss
--------------------------------------------------- PLUG-discuss mailing list - [email protected] To subscribe, unsubscribe, or to change your mail settings: http://lists.phxlinux.org/mailman/listinfo/plug-discuss
