Sometimes it's lower, sometimes it's much higher. Usually the more people that are using the blockchain the higher the fees. Bitcoin right now mostly follows a trend where it gains in popularity, which causes a surge in transactions, which causes fees to go up, which in turn causes people to stop using it, etc.
Low on-chain fees are actually a disaster for Bitcoin. Once mining rewards dwindle to nothing due to halvings fees would have to rise astronomically to maintain hash power.
How much hashpower do we need to keep the blockchain secure? Maybe it's less than what we have today, though I suspect it's more. (ASIC mining reduces the risk of 51% attacks because there's no huge mass of "dark silicon" that could be temporarily redirected from running Slack or Monster Hunter to mount a 51% attack; the vast majority of hashpower that has been manufactured so far is busily mining away to support the blockchain.)
I'm not convinced that higher transaction fees would be a huge disaster. Maybe Lightning will work out (El Salvador seems to be betting heavily on it), or maybe people will use "custodial wallets" like LocalBitcoins, or simply run up a tab at the bar; with any of these approaches, relegating the public blockchain to weekly, monthly, or trimestral settlements for most people would be a mild inconvenience rather than a disaster.
