What’s the basis for this assertion? There are still tons of young players, and young grandmasters. It’s true that Karjakin’s record is still unchallenged, but he broke the previous record by so much that’s not really surprising. It took 33 years for Judit Polgar to come along and break Fischer’s record, and that set off a torrent of new youngest-ever GMs that culminated in a 12 year-old doing it. In the 15 years since, about 20 13/14 year-olds have done it also, including a new record for youngest-ever to reach 2700 rating. That certainly seems like a fairly robust number of youths still view chess as worthy of serious pursuit, even if you grant that becoming a grandmaster is easier now than it was in Polgar or Fischer’s day (because rating inflation has continued, but GM is a fixed threshold). The top 20 players in the world are all under 50, and 12 of them are under 30.
There are lots of things that could be done to make a version of chess that appeals to a broader audience, but there’s not that much desire to actually do so. People who are passionate about chess like the intricacy of the tactics, the crazy skill ceiling, and the traditional forms of the game. People experiment around the edges of that for random products, but those rarely hit home, because if you bend the game too much you might as well just invent a new one. The more interesting question to me is how computer chess can help make the game more accessible. It will never be truly accessible on the scale of Candy Crush, but you could certainly create a version where each move earns you points based on the computer’s assessment of how good a move it was, and where the computer itself plays to keep the game interesting*, so that you can keep playing regardless of skill and end up improving as you learn to play well. Similarly, you could use the power of computer analysis to make live commentary more interesting, highlighting key variations that could be played next, providing an instant score for each move, and attempting to give viewers a sense of the complexity and difficulty faced by each player**. Then live announcers could draw on the computer variations, scores, and complexity assessments to provide the storylines and excitement that people want from a spectator sport. When people watch baseball, the commentators’ job is to build excitement and contextualize the plays: “3-2 count, tying run on first will be in motion with the pitch”; “Just outside, ball 2. That brings the count to 2 and 1. That really changes the at-bat here, now he’s got to come after him with the next pitch and that’s where it can get dangerous” and so on. There is a bit of “they should do this” but the drama is built from the game situation, the playoff situation, and the individual and team storylines like rivalries and position battles. These same things apply to football (where the average viewer only vaguely understands the complexity of each play), soccer, basketball, etc. Knowing what the best thing to do is doesn’t make those games less exciting, it makes them more exciting as you watch to see if the players can pull it off live. This could definitely be done with chess, and games like Starcraft and LoL have shown the way. Chess is starting down these lines with things like the chess.com pro leagues, but I think it needs much better 'casting and tournament staging to reach that level.
So tl;dr is: chess should embrace computer analysis, not just as a tool to make humans better, but as a way to assess and present human achievements.
*what I mean by interesting is having the computer play you and try to choose moves so that the best possible response stays within a couple points of zero without forcing a draw (that is, within a couple points, but not if that means every response leaves the game at 0). Perhaps it occasionally forces a takeback, using up a “life” when your move would hang a queen or whatever, to make sure it doesn’t completely ignore major blunders or create really absurd positions where a piece has been hanging for 10 turns.
**complexity and difficulty in this sense would mean how significant the best move is (that is, the difference in assessed value between the best, second best, and third best moves), how hard to find it is (that is, how many moves appear similarly good if you only look ahead 5, 10, or 15 moves), and also perhaps other measures like how volatile the line is (that is, how precisely do you have to understand what to do to achieve the assessment - are there many paths where neither player is making a big mistake, or just one viable path?) and how dangerous it is for the opponent (this is the reverse of the significance measure: how big is the spread of best responses for the opponent?).