Code and Common Sense

Remember those boxy robots from a couple of decades ago? The ones that spoke to humans and passed witty quips? I remember being so hooked to it that I'd have given an arm for that experience as a kid.

Safe to say, we fell rather short in our attempt to make that a reality. We used the language of computers aka logic, to develop algorithms that understood the syntax and structure of human speech.

But it doesn't feel like a conversation, does it?

John Searle's though experiment explains this best.

Imagine someone in a room with no native understanding of the Chinese language, but well-equipped with a set of dictionaries and rules of grammar. When presented with a sentence in Chinese, the person can translate the target sentence into English with reasonable accuracy.

In other words, the translation achieves fidelity.

That's not how we talk though.

Tiny humans learn languages by allowing it to emerge spontaneously at an acoustic level. These acoustic sounds gain meaning through positive or negative affirmations from others around. We 'train' our brains through countless such iterations.

Computers can be 'trained', just not spontaneously.

Take another example - ‘A man went to a restaurant. He ordered a steak. He left a big tip.’

If someone asked me what the man ate, I would say steak. But there is nothing in that line that suggests that he actually ate steak. We read between the lines.

Machines can't. The human learning experience cannot be replicated.

I admit, something as abstract as common sense is really hard to program.

But the gap in what we set out to build and what we managed seems unbelievably wide.

We set out to have human-like conversations with machines, but solved for board games.

And that leaves the kid in me disappointed.