/ Thoughts / Automation To our own tasks be false, and automate.

Automation, or what is repeatable.

Generally when someone finds out that one does scripting, they are bored, and roughly likewise when they find out one robotics. However some people, have either at the time of mentioning, or at some later date, a problem that they feel could use the assistance of one or both of those tasks. These problems often are completely impossible as initially phrased to be done by any known robotic process, or indeed often any known human process. Such things include:

Once we get past the initial request sometimes, there is a kernel of a problem I can help them with, and sometimes not. Often times I could help them, but the time needed to help them automate a process would not be worth it for a process done say 10 times a year. However, this experience has taught me that most people aren’t very good at understanding the basics of what even information science can do, much less information science + robotics can do.

So I have decided to write this little document, to help people understand the basics of what they can dream of.

Information and acting on information

Computers and digital systems can act on whatever they are given, and a little bit more. The key challenge though is that computers are rarely given all that they might be. Generally they are given things with at least some loss (or simple lack) of meaning, and while they can generally preserve what meaning they get, that initial lack is deadly. Take a simple thermostat, it can read the temperature, and without getting too philosophical it can act upon that with some degree of meaning. However if you feed a computer something like a text document then it has much less meaning. It has words, and words have meaning, but the whole thing fits together rather loosely. The computer is challenged to make sense of the whole. That isn’t to say it can’t be done to some degree, but it is one of those things where you better really have a good reason, and the means to back your reason.

Something like a picture is even worse. A common request is for someone to block all pornographic images, well that requires judgement even in humans, but that isn’t really a problem computers can judge. The problem is that it requires massive amounts of meaning to be extracted of what is nothing more then ordered colors. That is quite simply just doable to a reliable degree. Of course not all things need perfectly reliable systems, or even mostly reliable systems, and then the computer has a chance, but then again it is also extracting less meaning so that is somewhat expected.

The Environment and Acting upon The Environment

Most people by now realize, one hopes at least, that if robotics were easy they’d have a robotic chef, Nanny, and maid. What they don’t realize is the exact limitations, and well, sometimes they seem to think that the reason they don’t have such things is because no one has asked. Sadly many many people have asked, and even attempted to build such things. However there are several very serious problems that prevent this.

Number one of course is the problem of assembling meaning from sensor input which is a very hard problem on static scenes, and without the right world view moving scenes are even harder. Also sensors in an environment tend to be noisier and more likely to pass things that must be ignored. So to the extent that meaning extraction is needed it is often harder in the environment. Though certain methods of thought can at times reveal methods for using the very noise to reveal a self-filtering meaning aggregator, this has not been shown for the general case.

The other problem is of course interacting with the environment. It is here where technology has simply not the ability to replicate biological systems for flexibility, and crucially for feedback. While computer controlled arms can move things to exacting tolerances they lack the flexibility to operate in a free and unenclosed space. The relatively simple act of finding, and picking up an egg from a table blindfolded is an enormously difficult one for robotics because of the lack of the full range of feedback options in current systems. It is this lack in fact that drives some of the computer vision research, after all a blind man can do an amazing number of tasks in an environment, but it is often felt more fruitful to add vision then add the full range of feedback that would be needed to make a blind robot viable. Whether or not this “vision-mostly” line of thought will be a success in the long term remains to be seen.

Certainly most animals seem to gain many many differing sorts of sensors in great variety and density before they gain the sorts of visual systems that we regulalry try to graft onto a robot. Of course nature does not dictate all things. Wheels are a wonderful tool not found at all in nature. Yet still one must wonder.

The Future

As for, the future… well don’t expect massive changes just yet. Though I do think things are changing. What the world needs is a different way of looking at the problem. As the ways we have now are, almost comical, and progress so far has come mostly from sheer stupidity.

Though who knows maybe that is the way forward sheer dumb brute problem solving applied with really fast chips.

It will be interesting to see.