Reflections on “Artificial Intelligence Meets Natural Stupidity”, McDermott. D
I read through a decades old paper criticizing a few unhelpful feature of the AI field and its researchers. Reaction and thoughts are kept here as a record.
A Good Dressing Down
As a student, I am still very much outside the AI field and out of contact with its vanguards. The overall theme of the communications I am exposed to is one of confidence, excitement and urgency about the state of AI and it’s progress towards A-General-I. Only 30 years and we’ll will have our deity in circuitry* apparently, so I’d better hurry up and join the party.
I’ve only read through a few ‘naysayer’ type pieces and all of them including this one, Drew McDermott’s Artificial Intelligence Meets Natural Stupidity, have been really enjoyable for me. I hate thinking that I am naive about something which is important to me, and a little too often I get the feeling about AI that it’s too good to be true.
Successes in Deep Learning have come thick and fast, and naturally I and many other young people are rushing to absorb its lessons. This is it, we may think; with Deep Learning we will build a complete AI. Is it though? I see vaguely the same problems as described by McDermott in this paper existing some 30 years later today. McDermott called out the tendency to imbue AI programs with capacities they do not possess by naming them and their parts with names like THINK, UNDERSTAND, or DECIDE. Strip these ‘well-dressed’ programs of their clothes and is there anything left to be impressed by? As you break more into Machine Learning, some parts of it can be seen to be just “statistics on a Mac”, and an unsatisfying silicon substitute for real learning.
McDermott brings up the Behaviourists to warn against playing loose with the words used in a field. I am not yet informed enough to say whether the Machine Learning world is doing such things, and in effect walling themselves off from inconvenient but important questions. I can at least ensure that I am not, and reading through this paper and through the work of other skeptical philosophers, physicists andlinguists helps with that.
* [Deity In Circuitry] Why isn’t this phrased used more?