Entry 2244, on 2022-10-17 at 21:50:50 (Rating 1, Computers)
They say that if there is one thing that can be said in favour of artificial intelligence it is that it is better than natural ignorance. Of course, this doesn't compare AI with natural intelligence, which might be a better choice, but the implication is that using AI, despite whatever its deficiencies might be, is better than doing nothing.
Many years ago I started a postgrad diploma in computer science, which I never finished because I was already doing one and a half jobs and just didn't have time the course demanded. I remember taking about 30 minutes to figure out what even the title of one paper on the subject even meant, it was so full of technical jargon!
This was about 35 years ago and the promise of artificial intelligence was very topical at the time, although nothing much has really happened in that area until the last 5 or 10 years, so it has definitely turned out to be more difficult than many people thought.
But now we are seeing signs that AI is about to deliver on some of its promises, and maybe even exceed them greatly if things go the way some experts think they should. But why do we need it anyway? What could be the benefits, and could they possibly outweigh the potential problems?
Well, like all new technologies, it is hard to tell how effective it might be ahead of time. Many technologies were initially written off as being pointless or trivial. The internet is a classic example, where today it is used to the extent and in ways that people 30 years ago would never have imagined, and AI will almost certainly be the same.
I want to cite a few possible, relatively trivial, use cases where I think existing systems might be enhanced through the use of some extra intelligence.
The first is lifts. My policy at work is to take the stairs for going up 2 or less floors, or going down 4 or more. That varies a bit, and last week I took the stairs for ascending 7 floors. Big mistake; I'm not as fit as I thought I was!
Anyway, the thing I notice with lifts is how stupid they are. I work at a university and the lifts tend to get used at predictable times because of the schedule for lectures and tutorials. Yet the lifts never seem to "know" this. I often see all 4 lifts in one building near the top, say floor 10, while a large group of people are waiting on the ground floor.
Even without knowing about schedules a small amount of heuristic intelligence should be able to ensure that the lifts are distributed across the 11 floors and provide quicker service as a result.
Another behaviour I have noticed is a lift on floor 10 going down to ground might bypass me on floor 9, while I wait for a lift to come all the way up from floor 1. Why? I mean, it's possible that the lift going down is full, but I've never seen any evidence that the lifts know anything like that, so it's just poor design, apparently.
So some artificial intelligence might be good. The lifts know the load they are carrying because they have an overweight alarm. They know where people are waiting, because there are buttons to press. The rest they can learn, if they had a simple machine learning algorithm built into their logic.
So why not use ML to evaluate different algorithms (all 4 lifts would be controlled by the same controller which I think is already the case) and optimise their behaviour based on the average time spent delivering people to their desired floor. It shouldn't be too hard because machines have already shown their abilities in handling precisely specified tasks like that.
Here's another example: traffic lights. I find myself cursing my city's traffic light system almost daily (some days I don't drive near lights so I don't need to curse them!) because of their apparent stupidity.
We have a one way system here, which carries the major traffic through the city, and these are particularly amenable to optimisation. There is no reason why, once I enter the one way system, I shouldn't be able to travel at the speed limit (unusual for me, I concede) without being stopped by any of the 10 or 20 sets of lights. At the very least I shouldn't be stopped by the lights to find there is no traffic on the cross street.
But here's a situation I see quite often: I am stopped by the lights even though the cross street is empty. I wait, and notice a car approaching the lights on the cross street. As it approaches it is stopped by the lights and I am allowed to go. It's the worst possible outcome for everyone, although I admit that if the lights went red in both directions and stayed that way it would be even worse, but I am talking but likely scenarios here, not extremes.
So again, some machine learning could be used to improve the perfomance of the system. There are sensors in the road which detect traffic, so why not use those to measure the total flow of traffic and optimise the lights based on traffic flow? I presume a simple algorithm is already being used, but I doubt whether it uses any form of learning which is the critical component of my proposal.
You might ask why this is so important, given the relatively frivolous examples I have given. Saving a few minutes here and there barely seems worth the trouble to many people, but I disagree. A few minutes a day multiplied by many days and many people can make a real difference to how much time we have for productive activities, and it might also improve our general mood by reducing frustration.
And the two examples I have given are just a start. We could use AI to improve many other aspects of our lives as well, such as appliances which use electricity intelligently, or media which is delivered based on the user's preferences. And yes, I know some of this is already happening, which shows it can be done, but I just want more.
Eventually, once the technology becomes more stable, we could ask it to do more. Self-driving cars are an area where a lot of work is already happening, but the results are mixed. Occasionally AI controlled cars do something really stupid, and might cause an accident, and possibly even death. But statistically would the same car be any safer if it was driven by a human? After all, humans also make stupid errors. But which is worse: artificial intelligence or natural ignorance?
Comment 1 (7321) by OJB on 2022-10-18 at 12:01:55:
You can leave comments about this entry using this form.
To add a comment: enter a name and email (optional), type the number shown, enter a comment, click Add. Note that you can leave the name blank if you want to remain anonymous. Enter your email address to receive notifications of replies and updates to this entry. The comment should appear immediately because the authorisation system is currently inactive.
BLOG PAGE