AI for self-driving cars? I keep thinking that occasionally in driving we encounter: (1) A situation where we need to stop the car and converse in our natural

AI for self-driving cars? I keep thinking that occasionally in driving we encounter: (1) A situation where we need to stop the car and converse in our natural language, that is, for just our half of the conversation, do speech recognition and natural language understanding. So, the AI approach would be to get the list of the 100,000 most common conversations, tune some speech recognition to those, f'get about the actual language understanding, and, instead, for each of the 100,000 cases implement the most common resulting action or response in the data? Sorry -- in that case I'd rather not be in that car! (2) A situation where the driver needs actually to have real human understanding of a situation that, really, has never occurred before and, thus, is not in any AI training set. E.g., the vehicle ahead is a pickup truck and has some liquid dripping out of the truck bed out back. Somehow the liquid doesn't look like water. Taking a whiff, it smells like gasoline. Hmm. That stuff could catch fire, move to ignite the stuff in the bed of the truck, and maybe something could go "Boom". So, what the heck to do? Sure, slow down, get back, well back of the truck, change lanes and move ahead of the truck, pull off the right side of the road and stop, etc. IIRC, so far such general deductive reasoning is beyond AI. IMHO, such reasoning requires real AI, or whatever we are calling that now, and we don't know how to program computers to do that now. IMHO, first cut, for self-driving cars, the best chance would be to do some extensive re-engineering of the roads. Here is a general point: We don't yet understand how general human intelligence works and, thus, don't know how to program it. So, we are having trouble evaluating the automation we now have and, thus, are vulnerable to overestimating how far to real AI the current work really is. Besides, AI hype, if this is a case, and overestimating how much progress there is to real AI is a very old story: As I've heard, way back in the days of vacuum tube computers, IBM was pushing publicity about their "giant electronic human brains". Looks like IBM is still doing this.