Mechanisms
Discourse in terms of words like “agents,” “goals,” “motivations,” “success,” “failure,” and so forth is inherently teleological. For instance, it is impossible to talk about whether some action or another has “succeeded” without the idea that that agent had a certain end in mind. So we might say that the very concept of “success” implies teleological causation.
On the other hand, for something to be a mechanism, is to be the kind of thing we can explain purely with efficient causation. Mechanisms are designed to carefully isolate relevant efficient causations and screen the impact of those causes from the larger environment in which the mechanism exists. A mechanic can fix your car because your car is a mechanism carefully engineered to isolate a system of efficient causes from the environment outside of the car.
Digital computers are similarly mechanisms. They are carefully engineered so that only efficient causes enter into their operation.
Such a system, by the very way it was created, does not admit of teleological causation. So if you talk about such a system, and say that it “wanted” to achieve some goal, you are either talking metaphorically (probably attributing the “want” of the user of the mechanism to the mechanism itself), or you are talking about something that has no relevance to the outcome of how the mechanism behaved.
When this is pointed out to AI fanboys, a common response is to claim that “Human beings are just such a mechanism! How about that?”
Well, if that is true, then the consistent reductionists are correct: talking about human behavior using words like “belief,” “goal,” “strategy,” and so on, is just “folk psychology,” and should be banished from serious discourse. (Of course, such a “consistent,” reductionist approach, founders upon the fact that, on a strict reductionist account, there is no way to discern what is “serious” discourse.)