Hacker News Reader

I seem to remember these were the objections that were made at the time.

The basic tech dates back to SHRDLU and ELIZA. In some ways this is similar to the partially-self-driving-car problem: if it's a mechanical interface where the human is in control it's fine, and if it's a hypothetical human-equivalent AI it's fine, but in the middle people forget to adapt to the limitations of the system. And the limitations are very severe as soon as you go outside the lines.

View on HN

AsI recall, Terry Winograd later claimed SHRDLU was the beginning of the end of the strong AI trajectory, at least at that time. I believe he, or someone close, said the central weakness of the system was that when it failed, it did not have any understanding that it had failed, much less why. Seems this is the current SOTA in chatbots now, given the somewhat snarky examples in the article.

You see, it takes a huge amount of humility to admit that self-driving cars are possible, but we need several long years of supervised training to say we're actually there. Nobody would like that statement, people - including investors - want to have advanced tech here and now, they don't want vague and distant promises. That's why we deal with alpha and beta versions of everything now, including the partially-self-driving cars.

Reply on news.ycombinator.com