Posts by Kulumbasik

Hawking: RISE of the MACHINES could DESTROY HUMANITY

Re: Hawkins is correct if

I do not subscribe to such gloomy prospects and I think that kind of reasoning is quite primitive.

For a start, we do not even know now what intelligence actually is and what its constituent properties like awareness, consciousness and sentience together may actually imply. That may turn out quite different as you can imagine now.

Second, you completely miss the idea that humans constantly enhance themselves and in that way the evolution is going on. Mr. Hawking himself is a good example of this. Once you put glasses on yourself you immediately become something different as supposed by nature. You may think you are still human. But where is the borderline of such enhancements, after which you are already a "machine"?

Third, without a real progress in that field (AI), we humans are indeed doomed for extinction. We are too weak now. We haven't even developed ourselves into a Type I civilization (according Kardashev scale). We are in mercy of any big cosmic event like a meteorite that killed dinosaurs or a nearby supernova or a gamma-ray burst. The earth itself is doomed and in a billion years (or even less) will become completely unsuitable for life. We will need to leave the earth, most likely much earlier. That means, we will have firstly to develop a huge infrastructure in the nearby cosmos -- in effect becoming a Type II civilization. How would we do that without some artificial helpers (machines) able to work completely autonomously and withstand all harsh conditions of cosmic environment? Most likely, we will need billions of them!

Re: @Kulumbasik - We are too far away from it

> DARPA already works on robots which can autonomously identify an individual and decide to suppress it with no human interaction or supervision. What can possibly go wrong with that ?

What kind of project DARPA is doing may be not exactly the same as what the media is saying about it. They (DARPA) may be interested to gain more publicity (including with various outlandish stuff), thereby ensuring (directly or indirectly) more funding to it. I know first-hand how it is difficult to get funds -- you need to be creative about this! What they will produce in the end may be even more different thing. I highly doubt that will be on the level of something (robots) depicted, for instance, in "Robocop" movie. You may develop a program that would behave in some situations like a human, e.g. speak with a human voice or recognize your speech (more precisely, convert it into text). But to behave like a human soldier in the field? That seems to me something too much!

Take for instance a lot more modest goal, a software to translate from one human language to another. But what have they achieved so far? Even Google with all its computational power and databases wasn't able to create a decent translator. I frequently need to use one. But what kind of output does it produce? In many cases it is little more than some gibberish unintelligent stuff that without a deep correction cannot be used anywhere. That's because to translate it correctly, ultimately the software needs to understand the meaning of the text. Without similar functionality no truly intelligent robot could exist.

> Now imagine a future when those who programmed this are retired or simply dead.

Modern software projects are not developed by a single person. They are typically well managed, documented an so on. That's the value of that software, not just the lines of code! There's a whole branch of software industry (maybe even larger than AI) dedicated exactly to management of other software projects (that is called "Application Lifecycle Management"). By the way, that actually only stresses how laborious the software development actually is (and, therefore, how far away from AI).

> How about a firmware update going wrong ?

All the same as it is now. What would you do when your "intelligent" vacuum-cleaner isn't working after the last firmware update?

Re: We are too far away from it

> There's a whole sub-genre of computer science devoted to it.

Sure. That field does exist and I am a sort of working in it by myself. But my feeling is that all this is basically old-style programming work re-branded as something new and called "AI". Those programs indeed do something previously only humans did (like recognitions of a human face and finding it in a database). But are they really sentient? Are they able to think anything of their own (let alone to redesign themselves)? They are still just calculators, however very complex ones.

Any device we can think or develop now as "AI" will inevitably be a Turing Machine (TM) -- even quantum computers, which are supposed to be exponentially more powerful on some tasks. But without the input of external information a Turing Machine cannot produce anything new -- that's the mathematically proven fact. What that "new" (in the informational sense) is may not be only about brilliance or creativity. It may be actually a critical component of awareness and sentience.

Of course, a computer (that is a software that powers it) may use various input data to improve itself. So, it would be considered as some kind of open system, thereby breaching the TM barrier. But would that environment input be enough? After all any animal on earth has that kind informational input, which doesn't make them intelligent. You need also to consider the intensity of that environment data flow -- it doesn't depend much on the design of the AI device. If evolution is any example, it took billions of years to "design" anything.

Overall, my feeling is that the current "AI" field is mainly about marketing buzz, and the notion itself is highly overblown. Indeed, there is lots of research going on there. But that's all essentially old-style computer science and programming (that is the development and implementations of various algorithms for TM).

But really few researches does exist about truly fundamental things. The last I've read so far was "Shadows of the Mind: A Search for the Missing Science of Consciousness" by Roger Penrose (and some additions to it).

We are too far away from it

Talking about AI is something the same as talking about interstellar travels. Is it good or bad? Doesn't matter! Whatever it is, not only don't we have a working starship, we have no even the fundamental physics needed to build one.

The same is with AI. Any AI we could imagine now will be some kind of computer. However, any device we can think now as a computer, regardless of its processing speed and memory size, will inevitably be an equivalent of Turing Machine -- the so called Church's Thesis.

But Turing Machine, a mathematical construct invented specifically to analyze any computational processes, proved to have some fundamental limitations. For instance, there cannot be a program that can create other programs, even for some quite narrow classes of tasks. We humans somehow do it... But the main limitation of any computer system (a Turing Machine, that is) is that it cannot create information by its own. A computer is always just a transformer of information. Yes, its capabilities can be extended indefinitely by adding new programs. But it cannot create those programs by itself. Human programmers are needed for that.

Being a programmer by myself, it always strikes me what meticulous efforts are need to teach (i.e. program) a computer to do quite simple things. The whole new specialized computer languages pop up all the time to do precisely that in particular fields.

It is only humans, who create information! But how can we do that? Maybe our brain listens to the cosmos, catches the information emanated from it in the form of entropy.... Anyway, we don't even have a physical theory about that (the same as with interstellar travels).

In the end, I think, in spite of all this anti-AI buzz raised by various celebrities, currently we are actually far away from the creation of a truly sentient being, and since we don't even understand what it is, all those fears are essentially baseless.