Do you really need AI?
— for your next project —Suppose, for the sake of argument, that carpentry is your favorite pastime. You have a problem, though: it exposes you to the very real risk of even serious injury. Take table saws, for example. As essential as they are for many tasks, it takes very little to risk losing a finger or even a hand.
At this point, you decide to try to solve the problem by designing a protection system for this kind of injury. The system must be able to:
- Identify when a hand or finger is about to make contact with the blade
- When this event occurs, it should immediately lock the blade
As you begin to gather ideas and opinions on how to deal with this, in this day and age, it would not seem at all strange if someone suggested that you solve the problem by employing AI.
After all, it is not difficult to think of a machine vision system trained specifically to recognize hand-blade contact. Then all that would be needed would be to link it to a mechanism that, as soon as the danger is detected, stops the rotation of the blade.
In this way, we could achieve a system that prevents a good percentage of injuries. But would it be a good system? Not really.
Now, as it happens, this problem has already been solved in a very intelligent way. And the way is very intelligent, in this case, also because it does not resort to AI.
The basic operation is as follows: an electrical signal constantly passes through the blade, even when the blade is cutting wood. Since the human body, however, has different electrical properties than wood and since the electrical signal is specially calibrated for this purpose, a contact of only a few microseconds between any part of the body and the blade is enough to interrupt the signal and trigger the locking mechanism almost instantaneously.
Quite an effective system!
Now, imagine if we had implemented a similar solution instead, but based on AI. Are we sure that the system would have been able to match the performance? Specifically:
- Identify the hazard reliably (without false positives or, worse, false negatives)
- Doing it in such a short period of time
Perhaps some doubts arise.
The key point, without getting too philosophical, is that there is a structural difference between a "mechanical" system, on the one hand, and an "interpretive" system, on the other. So on the one hand, we will have a simple, linear, predictable input-output combination, comparable in readiness almost to a kind of stimulus-response; on the other hand, we will have a complex, non-linear, non-predictable input-output combination.
But, that said, let's also assume that we have an AI system good enough to be suitable for our problem, which satisfies points 1 and 2 (AI is making great strides after all): would it make sense in this case to use AI? Again, no. Having a better alternative, even assuming comparable performance, resorting to AI would mean creating a system that is most likely more resource-intensive and, in any case, over-engineered.
So, is it always better to avoid using AI? Absolutely not, and that is not the message you want to get across. There are plenty of tasks, indeed, for which AI is the best way to deal with theme (or the only possible way).
It is therefore crucial to understand, from time to time, the nature of the problem you want to address in order to decide what might be the best approach to solve it. This step, however, is particularly delicate and may prove not to be easy. Staying with our problem, for example, one realizes a fundamental difficulty: while an AI-based solution would have allowed us to abstract the general aspects of the problem while ignoring more specific ones, the "better" solution, on the other hand, required a deeper and more specific knowledge of the scope of action (e.g., some electrical properties of bodies).
So, do you really need AI for your next project? Well, maybe you do. Or not. You need to fully understand the problems you will face and consider the tradeoffs of the options you have to solve them. And most important, just don't let the hype drive you too much! ✨