Sources on alpha zero´s preference?

All the games played between stockfish and alpha zero used opening books the first 10-20 moves. Also, alphazero does not have the same evaluating function as stockfish do. It is simply made to come up with a move in a given position, and that move is based upon self training. From the self training games e4 and d4 scored about equal. White won more with e4 but in some games, black also managed to get the full point against e4. While d4 were mostly drawn. I get this information from the official report.

e4 for cute tactics

d4 for real game

A real boring game you mean.

“It’s interesting to compare AlphaZero’s opening choices. As Black, AlphaZero restricts its opening repertoire severely, meeting 1.e4 invariably with 1…e5 – the most popular choice at the top level – and 1.d4 with 1…♘f6 2.c4 e6, aiming for the Nimzo-Indian after 3.♘c3 ♗b4 and the Ragozin Defence after 3.♘f3 d5 4.♘c3 ♗b4.
As White, it favours 1.d4 openings heavily and plays almost exclusively systems where the king’s knight is placed on f3. It also opens 1.♘f3 fairly frequently, and though these games often transpose to standard 1.d4 openings, it also plays the anti-Nimzo-Indian move order 1.♘f3 ♘f6 2.c4 e6 3.♘c3 ♗b4 4.♕c2 with some regularity”

Excerpt From: Matthew Sadler. “Game Changer: AlphaZero’s Groundbreaking Chess Strategies and the Promise of AI" 

This analysis is on how frequent d4 and Nf3 were played as the first move in its training games. In case anybody missed this, alphaZero was not programmed the same way stockfish was, it taught itself how to play. 

D4 openings did occur more in AlphaZero´s games against itself. But for example, the ruy lopez did also occur more often than the queen's pawn game. That does not mean that AlphaZero prefers the ruy lopez over the queen's pawn game. 

But if we are going to look at the games AlphaZero played against itself, AlphaZero´s 3 best scoring opening positions as white was actually the Ruy Lopez, caro kann and the french defence. Where white won more than 50% in all games! As you may notice, none of these openings is d4 openings. But once again this evidence does not mean that AlphaZero prefers e4 over d4, it simply means that when AlphaZero learned the game of chess, it scored the best with these openings.

I´m not arguing for either side, but saying that a complex algorithm that is far from comparable to a normal chess engine such as stockfish, evaluate d4 over e4 while AlphaZero cannot evaluate any positions at all raises some alarms. Especially when some of the evidence given by the speaker suggest otherwise.

E4

why do some people write it E4 instead of e4? The capital letters are for pieces. It's how you differentiate Bxc4 and bxc4, for example.